This PDF file contains the front matter associated with SPIE Proceedings Volume 13229, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Wafer defect classification is a critical issue in the semiconductor field, yet the detection accuracy of existing methods still can be improved. In this paper, a novel method based on YOLOv8n is proposed, it integrates the EMA mechanism and ODConv. The EMA mechanism assigns higher weights to important features to enhance accuracy, while ODConv reduces computational complexity through parallel computation. Experimental results on the WM-811k dataset demonstrate that our method achieves an accuracy rate of 97.5%, which represents a 2.2% higher accuracy than the baseline model, while reducing computational complexity by approximately 27% compared to the baseline model. This result confirms that the proposed method can maintain high accuracy in the semiconductor field while having lower computational complexity.

With the advancement of semiconductor technology, the lighting field has stepped into a brand new period - the LED light era, LED lights have obvious advantages, including long service life, environmentally friendly, high conversion efficiency and stable performance, etc. The performance of LED lights greatly depends on the switching power supply, and in the real environment the LED switching power supply will have the problem of In the real environment LED switching power supply will have the problems of low efficiency and unstable voltage, so to improve the performance of LED lamps need to start from the power supply. This paper proposes a method of using resonance and power feedback loop to solve the shortcomings of the current LED switching power supply and realize the design requirements of LED switching power supply with high efficiency and constant voltage.

With the intelligentization of urban construction in our country, the use of modern technology and information communication technology to enhance the sustainable development of urban construction, smart manhole covers have gradually gained widespread attention. This article mainly discusses the design and implementation of smart manhole covers. Intelligent management of manhole covers can achieve functions such as remote monitoring, real-time data feedback, and real-time data analysis. By designing the STM32 microcontroller, sensor, and NB-IoT module for smart manhole covers and implementing remote monitoring on the OneNET cloud platform, this article demonstrates that the design can achieve real-time monitoring of manhole cover status and timely response, ensuring that manhole covers on city roads are in normal condition. It also improves the maintenance efficiency of manhole covers and prevents traffic accidents, as evidenced by experimental results.

In this paper, an electromagnetic compatibility design scheme of a traveling-wave tube amplifier is introduced. Based on the traditional EMC scheme of TWT amplifier, a scheme of wrapping absorbent tape on the surface of the high voltage lead between EPC and TWT was proposed to minimize the energy density of low frequency leakage, thus effectively improving the EMC capability of TWT amplifier. Provide technical support and experience support for other satellites' TWT amplifier equipment to meet aerospace EMC requirements.

In the field of optical devices, numerous specialized materials exist, and conventional detection techniques are inadequate for determining the residual capacity of these materials during daily device usage. This uncertainty poses significant challenges to device utilization. Ascertaining the remaining amount of special material in a container is unfeasible, necessitating refilling before each use; however, opening the container causes evaporation of the special material, resulting in wastage of valuable resources. Due to its chemical properties, transparent containers cannot be used for storage or estimating remaining capacity through visual inspection. Therefore, employing X-ray imaging and computer vision-based image processing offers an effective solution to determine residual capacity by capturing images of metal containers and subsequently analyzing them using edge detection and image restoration technologies proposed in this study. This method successfully extracts the contour boundary of the special material in the original image, and provides necessary support for the subsequent calculation.

With the development of various precision-guided weapons, new radars, and detection means, stealth technology and stealth materials are paying more and more attention. Stealth materials have been widely used in aircraft, missiles, ships, UAVs, and other military fields. Barium ferrite (BaFe₁₂O₁₉) is a hexagonal crystalline ferromagnetic oxide that has good chemical stability, corrosion resistance, and uniaxial magnetic crystal energy anisotropy. The easy magnetization axis parallel to the C-axis makes it have a large involuntary force and magnetic energy product, so it has been widely used in the field of stealth technology and electromagnetic shielding. In this paper, pure ferrite barium nanoparticles were prepared by sol-gel self-combustion technique. The reaction mechanism and the critical temperature of the solid-state reaction of the dried gel were studied by TG/differential scanning calorimetry (TG-DTA) and Fourier transform infrared spectroscopy (FT-IR). The structure and properties of barium ferrite were studied by scanning electron microscopy (SEM), vibrating sample magnetometer (VSM), and X-ray diffraction (XRD). The results show that the particle size of the products increases gradually with the increase of calcination temperature and is greater than the critical particle size of a single domain. When the precursor was held at 400℃ for 2h and calcined at 1200℃ for 3h, the maximum particle size of 82.7 nm was obtained, greater than the critical particle size of the single domain. The maximum saturation magnetization of the product was 64.72 emu•g⁻¹ , and the minimum coercivity was 1181.07 Oe•g⁻¹.

The rectangular block graphite heat exchanger is one of the three commonly used types of impervious graphite heat exchanger. The heat transfer and flow resistance characteristics of rectangular block graphite heat exchanger were studied experimentally in this work. Firstly, the overall heat transfer coefficient and heat transfer coefficient in transverse holes were obtained. Secondly, the pressure drops in both transverse and longitudinal holes were acquired. Finally, the Nusselt number was fitted as the function of Reynolds number and Prandtl number and the Euler number was fitted as the function of Reynolds number. It is found that the overall heat transfer coefficient increases with the rise of fluid velocity due to the higher heat transfer coefficient in transverse holes. The pressure drop in transverse holes is much greater than that in longitudinal holes because of the baffling flow and reinforcing bar. The Nusselt numbers in transverse and longitudinal holes are respectively about 90% and 110% of the classical value. And the Euler numbers in transverse and longitudinal holes are respectively about eight times and two times of the classical value.

The aluminum-silicon alloy (Si: 25wt.%) is modified by the Pr modifier. The results show that Pr has a remarkable refining role on primary silicon in aluminum-silicon alloy (Si: 25wt.%). When the additional amount of Pr is 0.6%, the grain size is refined from 180-220 μm without modification to 60-80 μm after modification. The shape of primary silicon has changed from coarse five-petal star-shaped and other complex shapes to fine polyhedra and plate-like, while the size of long needle-shaped eutectic silicon decreases and the distribution density remarkably decreases. The X-ray diffraction analysis results indicate that no new phases are generated in the modified alloy. The scanning electron microscopy analysis results show that after Pr modification, a white trapezoid structure AlSi_xPr_y ternary compound appears in the aluminum-silicon alloy (Si: 25wt.%) sample. The hardness testing results show that the Pr modification additional amount is 0.6%, and the hardness value of aluminum-silicon alloy (Si: 25wt.%) has been improved by about 32.5% from 56.3 HB without modification to 74.6 HB after modification.

The water flow characteristics in a fishway largely depend on the geometric shape of the partition. This study is the first to compare two classic fishway forms and study their hydraulic characteristics using CFD method. This study was conducted in a water tank with a slope of 10%, a length of 43.6m, and a width of 1.5m. Set up a rest area for every 10 baffles. The simulation results show that on the one hand, the vertical-slot fishway has better energy dissipation effect and more significant flow velocity reduction, and on the other hand, the Daniel fishway has a larger rest area at the bottom.

The problem of cable fault detection will greatly affect the safety of aircraft operation, requiring timely detection and handling by staff. However, the problem of aircraft cable faults relies more on manual detection methods and traditional offline testing methods, and the amount of voltage data under normal operating conditions will be much larger than the fault data. Direct detection using existing deep learning models cannot achieve good results. To surmount this challenge, we present a noval model, annotated as CFDDR: Algorithm Based on Deep Learning and Cable Signal Reduction. CFDDR consists of three basic components: an aircraft cable data balancing module, a feature extraction module, and a deep neural network module. This method has excellent performance in detecting aircraft cable faults.

To enhance the performance of the PMU, a voltage reference based on a voltage bandgap has been proposed. During the power-on process, the startup circuit is activated due to parasitic capacitance coupling. After the power-on process is completed, the startup circuit self-turns off to reduce power consumption. Subsequently, a high-order compensation circuit is designed to generate a current with only constant and high-order temperature terms, compensating for the temperature coefficient of the voltage bandgap. An RC filter is then added to the output to improve the power supply rejection ratio in the high-frequency range. To further reduce the power consumption of the voltage bandgap, almost all MOS transistors are operated in the subthreshold region, achieving a decrease in static current consumption. Circuit design and simulation were completed based on the TSMC 0.18 μm CMOS process. The results indicated that at a supply voltage of 1.8 V, the bandgap output was 1.232 V; within the temperature range of -40 ℃ to 140 ℃, the temperature coefficient was 4.8×10^–6 ℃^–1; with the addition of the RC filter, the power supply rejection ratio at high frequency was -73 dB @ 10 MHz; furthermore, the static current of the entire bandgap was only 3.7 μA

Traditional machine vision based on von Neumann architecture suffers from the limitation of memory and computing separation. With the explosive growth of data, meeting the application requirements has become challenging due to the architecture. Artificial neuromorphic vision systems (ANVS) inspired by the human visual system (HVS) are considered a viable solution. Organic synaptic devices have great potential in neuromorphic computing. We propose an artificial neural visual system (ANVS) based on organic electrochemical synaptic transistors (OEST) and a photodiode array strategy in this work. The basic biological synaptic behaviors can be successfully achieved by optical stimulation. This system successfully simulated the human visual system and was able to perform feature extraction on color images, producing results that aligned with simulation expectations. Color images can achieve a recognition rate of 99.4% after OEST-based hardware convolutional kernel processing, whereas this rate drops to only 33.4% after processing through software convolutional kernels. Comparison of the recognition rates shows that the OEST-based hardware convolutional arrays are more closely matched to the mechanisms of the human eye's retina and can recognize various color images. Overall, this work demonstrates that OEST has great potential in the field of neuromorphic vision.

As a key module of the phase-locked loop in the RF transceiver system, the performance of the voltage controlled oscillator directly affects the performance of the entire system. This article focuses on the application of low-power phase-locked loop systems and designs a low-power and low phase noise voltage controlled oscillator circuit. The circuit structure is improved on the C-class LC voltage controlled oscillator structure, using a resistor array and current mirror to achieve gate dynamic bias, thereby solving the startup problem of the C-class voltage controlled oscillator and ensuring that the circuit maintains the C-class working state after stable oscillation. CMOS complementary cross coupling is adopted to achieve current reuse and reduce power consumption. By removing the tail current MOSFET, the source of phase noise is reduced, making the overall phase noise of the circuit lower and more suitable for low voltage application scenarios. The design of this article is based on TSMC 40nm CMOS technology. Simulation results show that under a power supply voltage of 0.8V, the overall power consumption of the oscillator is 0.871mW. It can cover the frequency range of 2.37-2.44GHz at all process angles, and the phase noise at the 2.4GHz frequency point is less than -122.28 dBc/ Hz@1MHz Meet the performance requirements of phase-locked loops.

Cables play an integral role in ship and marine engineering applications. By simulating cables in ship simulators, crew safety and work efficiency can be effectively improved. Therefore, in conjunction with the latest developments in flexible object simulation, geometric, physical and hybrid modeling methods for cables are comprehensively explored and the advantages and disadvantages of the three modeling methods are examined. The physical simulation algorithms are analyzed in depth to ensure the accuracy and reliability of the simulation results. According to the latest research progress in the field of collision detection, cable collision detection algorithms with real-time capability are summarized. Finally, the development trend of cable simulation in ship operation simulators is outlooked.

With the rapid construction of a new power system with new energy as the main body, aiming at the problems of high voltage fluctuation intensity and randomness caused by new energy access to the grid, this paper proposes an electric energy measurement method for new energy access by optimizing hardware and software based on high-precision analog signal sampling and measurement algorithm design. The hardware scheme of the measurement module is constructed by using the optimal design of zero-flux current transformer, sampling resistor and high precision ADC chip. In order to improve the synchronization of current and voltage signals and the accuracy of power measurement, a strategy of compensating the phase deviation of current and voltage is proposed. Finally, different experimental conditions are set to verify the effectiveness of the proposed method. The results show that the proposed method achieves high measurement accuracy, which is better than the technical requirement of 0.2S of active power accuracy.

The monitoring of insulated gas leakage in the power industry mainly relies on infrared imaging means, which realizes gas diffusion imaging display by measuring the regional temperature difference caused by the selective radiation absorption of insulated gas. Aiming at the difficulty of identifying trace insulated gas leakage under complex background interference, this paper proposes the research of insulated gas leakage imaging enhancement technology based on Gauss Mixed Model (GMM), which enhances gas diffusion display by deducting complex dynamic background. Through the simulation experiment of insulated gas leakage, the infrared multi-frame images of gas leakage were collected and processed to realize the diffusion marking of insulated gas leakage, which preliminarily verifies the feasibility of the proposed method, and has certain application potential in intelligent imaging monitoring of insulated gas leakage in power industry.

This paper combines the advantages of machine learning and satellite remote sensing to study a method for predicting the tropospheric NO₂ column concentration based on machine learning, so as to quickly and efficiently obtain the tropospheric NO₂ column concentration directly from satellite spectral data. Extreme gradient boosting, random forest and BP neural network algorithm models were constructed respectively. Among them, the BP neural network model has the best performance, with an average percentage error of 10.4% on the test set, a symmetric average absolute percentage error of 9.5%, and R² is 0.94, and the average percentage error is 1.8% and 12.1% lower than the random forest model and extreme gradient boosting model respectively. The optimal model was used to predict the daily tropospheric NO₂ column concentration and compared with TROPOMI's NO₂ product. The average percentage error was less than 10.3%, which fully demonstrates that the constructed BP neural network algorithm model can effectively predict the tropospheric NO₂ column concentration.

Graphite composite bipolar plate is one of the core components of the fuel cell, which is mainly composed of graphite and polymer resin, and its electrical conductivity is greatly related to the ratio and distribution of these two materials. In order to study the electrical conductivity of graphite composite bipolar plates, the surface layer of the sample was removed by manual micro-grinding and the distribution law of polymer resin and graphite was investigated by resistance measurement using a through-plane resistance meter. RESULTS: Sample 1 was milled from 1 mm to 0.2 mm in thickness, and the resistance decreased sharply from 0.97 mΩ to 0.51 mΩ when the thickness was milled from 1 mm to 0.9 mm, while the resistance jumped between 0.42 mΩ and 0.57 mΩ when milled from 0.9 mm to 0.2 mm. Sample 2 was milled from 1mm to 0.76mm in thickness and the resistance jumped from 0.93mΩ to 0.48mΩ when the thickness was milled from 1mm to 0.94mm, whereas the resistance jumped between 0.38mΩ and 0.51mΩ when Sample 2 was milled from 0.94mm to 0.76mm. CONCLUSION: By measuring the resistance values of sample 1 and sample 2 after grinding and comparing them, it is known that the thickness of the resin on the surface layer of the sample is about 0.06mm,because of the high content of resin, which leads to the high resistance of the surface layer, and the high content of graphite inside the sample, which leads to the small resistance. This paper provides some research value for improving the electrical conductivity of the surface layer of graphite composite bipolar plates.

Electrocatalytic reduction of CO₂ to formic acid is an effective method for addressing excessive greenhouse gas emissions. However, CO₂ Reduction Reaction (CO₂RR) still suffers from issues of low selectivity and energy efficiency. Herein, we report the preparation of flower-like bismuth nanowires (Bi NWs) that can be synthesized via in situ electroreduction. Bi NWs provided more active sites with improved intrinsic activities. Compared with urchin-like Bi nanorods (Bi NRs) and Bi nanoparticles (Bi NPs), Bi NWs exhibited the highest activity over a broad potential window, showing a 90.7% Faradaic efficiency and 20.4 mA cm⁻² current density at −0.9 V vs. reversible hydrogen electrode.

Surface unmanned platforms can realize the purpose of attacking the enemy boats by loading counterload. In order to solve the problem of maximally traumatizing the enemy boats under the premise of limited damage capability, a cascade detection algorithm for key parts of boats based on YOLOv7 is proposed. Firstly, the detection strategy is reconstructed, and the detection of the key parts of boats is divided into two stages. The cascade detection algorithm is designed to detect the whole boats first and then the key parts. Secondly, the boats dataset NSDS and the key parts dataset NSKPDS are constructed. Thirdly, considering that multiple boats targets may be detected in the first stage, a multithreaded design is introduced in the second stage. And the relevant control experiments are carried out. The results show that the multithreaded design can effectively shorten the detection time of the cascade detection algorithm. Finally, the proposed cascade detection algorithm, YOLOv7, YOLOv5 and YOLOv8 algorithms are tested on the constructed datasets. And the results show that the cascade detection algorithm can not only effectively complete the detection of key parts of boats, but also has a higher total detection rate and average confidence than other detection algorithms.

In the NISQ era, exploring quantum approaches capable of addressing real-world data analytics tasks has been a focal point of research. The detection of anomalous images is one of the classic data processing problems that can be considered to tackle using quantum computing. Inspired by the proposal of quantum circuit autoencoders (QCAE), we design a quantum method to tackle image anomaly detection tasks. In this work, we propose a method for encoding an image into a quantum circuit. Then, we employ the QCAE model to detect the anomalous images. Moreover, we create a scoring function to predict whether test data is an outlier. Finally, combining the trained QCAE with the scoring function, we obtain an anomaly detector. We use the handwritten digit image dataset to train detectors under various training setups and validate their performance on a test set. Our method achieves an f1-score of 0.8190 and a recall of 0.8600. The experimental results demonstrate the potential of quantum computing in handling data anomalies in the classical world, encouraging further exploration of QCAE applications in practical problems.

In the past few years, there has been a rise in incidents involving gas leaks, raising serious concerns for human safety and well-being. It is of utmost importance to develop an advanced gas detection platform that ensures accurate identification of gases and enables timely alerts to prevent accidents. Field-effect transistor (FET) gas sensors have a significant role in real-time environmental monitoring, medical pre-diagnosis, industrial control, and other scenarios due to their unique ability to amplify gate electrode signals and detect even the smallest amounts of gases. In this study, we have developed a gas sensor based on carbon field effect transistor (FET) technology to detect hydrogen (H₂) at room temperature. The sensing gate layer of the sensor consists of Pd/Ag, while the conductive channel is made up of carbon nanotubes (CNTs). To mitigate the negative effects caused by channel exposure, we have incorporated yttrium oxide as the gate dielectric layer. Our fabricated carbon-based FET gas sensor demonstrates an impressive detection limit of 70 ppm for H₂ at ambient conditions and can effectively function as an early warning system for gas leakage incidents. Moreover, our designed sensor exhibits good selectivity, repeatability, and anti-humidity property. We anticipate that our research will contribute significantly to advancing high-performance and highly integrated trace gas sensors by establishing solid theoretical and experimental foundations.

Handwritten ancient documents present unique challenges due to paper aging, ink fading, and blurred handwriting, making their text detection more difficult than standard tasks. Simultaneously, the layout structure is notably intricate, featuring double columns interspersed with single columns along with a blend of images and text, presenting challenges for detection. Therefore, considering the challenges of images from ancient documents, a single stage text detection method on Scale wise Feature Aggregation Module (SFAM) is proposed. It builds on fully convolutional networks to directly generate character level predictions, that identifying redundant and slow intermediate steps Furthermore, by fusing feature maps of different scales to encode information from different receptive field sizes and introducing channel attention mechanisms to allow issues caused by scale variations among different object instances, effective and accurate detection of characters in ancient documents is achieved. In order to assess the effective of the approach, we carried out experiments utilizing the MTHv2 dataset. Our findings indicate that the proposed method surpasses the majority of other text detectors in terms of precision, recall, and F1 score.

During the operation of apple picking robots, it is necessary to accurately identify branches to prevent collisions with them. However, obtaining branch images in complex environments is often fragmented, making it difficult to reconstruct the entire tree. To solve this problem, this study obtained branch images by removing non-branch areas, and extract tree skeletons and endpoints based on 8-connection and thin line methods. This study investigates the growth characteristics of fruit tree branches and the distribution of endpoints between branches at the same level, using this as a constraint to connect broken branches. Finally, the reconstruction of the entire tree is achieved. The experimental results show that the algorithm has a branch recognition rate of 92.1% on sunny days and 84.2% on cloudy days. Therefore, the algorithm proposed in this study can provide accurate branch information for apple picking robots.

To study the modified application of graphene aerogel composites in mainstream energy storage devices. Firstly, AFMMMSEMTEM electron microscope was selected to observe the size state of the modified mainstream energy storage devices; Secondly, Raman was used for spectrum electron energy spectrum ray diffraction analysis, and finally BET specific surface area test was carried out to obtain the performance indicators of the mainstream energy storage device after modification, thus completing the modification test of the mainstream energy storage device. Using graphene aerogel composites to modify mainstream energy storage devices can effectively improve the energy storage capacity, energy storage coverage, operating performance and thermal stability of energy storage devices, which proves that graphene aerogel composites have good modification performance and high reliability.

The current highway CCTV monitoring equipment control mechanism is generally set as a target structure, the control mode is relatively single, resulting in control efficiency is too low, resulting in the eventual control frequency decline, which puts forward based on the improvement of the expectation maximization algorithm of highway CCTV monitoring equipment control technology design and practice analysis. Combined with the actual control needs, want to thank you for closed-circuit television monitoring equipment state real-time capture, the use of adaptive methods to improve the control efficiency, the design of adaptive sensing control mechanism. Based on this, the establishment of highway closed-circuit television monitoring equipment to improve the expectation maximization measurement control model, using tracking sensing processing to achieve equipment control. Test results show that: compared with the infrared image visualization highway CCTV surveillance equipment control technology, task behavior technology highway CCTV surveillance equipment control method and the design of the improvement of the improvement of the expectation maximization measurement of highway CCTV surveillance equipment control method ultimately results in a relatively high frequency of unit control, which indicates that the design of the highway CCTV surveillance equipment control technology is more efficient, complete and specific. This shows that the designed highway CCTV monitoring equipment control technology is more efficient, perfect, specific, targeted has been significantly improved, in the complex background conditions, the controllable difficulty has been further controlled, stability has also been improved, with practical application value.

This study successfully synthesized TiO2 nanotube films doped with Ni, Mo, and Ag ions via the spin-coating technique, exploring the impact of these metal ions on the photocatalytic efficiency of TiO2 films. The films' structure, morphology, and light absorption characteristics were analyzed through X-ray diffraction (XRD) and ultraviolet-visible (UV-Vis) spectroscopy. Photocatalytic activity was evaluated by conducting photodegradation experiments with methylene blue on the doped TiO2 films. Results demonstrated that suitable metal ion doping markedly enhances TiO2's photocatalytic efficiency by boosting visible light absorption and optimizing the photogenerated carriers' separation and transfer, thus amplifying methylene blue degradation. Experimental findings suggested that by fine-tuning the doping concentration, optimal photocatalytic activity is attainable, with Ag-doped TiO2 films displaying superior performance.

The identification of steel corrosion is a critical task in the field of engineering maintenance; however, the high cost of collecting corrosion samples has limited the application of traditional identification algorithms. This challenge is particularly acute for Transformer architecture models, which require a substantial amount of training data and thus face significant difficulties in corrosion identification. Addressing this issue, this study explores a pre-training strategy for Transformer architectures aimed at recognizing the state of steel corrosion. By employing masked self-supervised learning for pre-training the model under conditions of limited training samples and subsequently refining it through fine-tuning, this approach enables precise identification of steel corrosion states. Experimental results on a custom datasets demonstrate that this method, without increasing the cost of additional annotations, effectively utilizes the limited sample information for learning and modeling, thereby significantly enhancing model accuracy, which provides new insights and tools for addressing similar engineering problems and suggests that self-supervised learning could become an important technological trend in engineering maintenance, especially in contexts where sample acquisition costs are high.

Surface flashover caused by electric field distortion usually results in the insulation failure of electrical facility. Coatings with nonlinear electrical conductivity contribute to the regulation of the surface electric filed. Herein, we prepared a fluorocarbon resin coating performed nonlinear electrical conductivity via compositing with SiC particles. Besides, the relative permittivity of the fluorocarbon resin/SiC coating (FR-SiC coating) were measured. Then, we established a simulation model to indicate the surface electric field regulation. Results shows that the surface electric field distribution of the sample with nonlinear electrical conductivity was regulated effectively. The electrical distortion among the electrode, the water droplet, the coating and the air was relieved.

Driver fatigue is one of the leading causes of traffic accidents and poses a serious threat to traffic safety. Therefore, realizing real-time monitoring and early warning of driver fatigue status is especially important. Currently, most fatigue driving detection technologies are based on physiological signal detection, but this method cannot accurately determine fatigue behavior, and there are limitations to the use of high sensor costs. To this end, this paper proposes a fatigue determination algorithm based on deep learning face key point detection. The algorithm uses the RetinaFace network for face detection and intercept, the PFLD face key point detection network to extract facial key points and calculates fatigue indicators based on these key points to achieve the detection of driver fatigue state by integrating multiple evaluation criteria. The experimental results show that the algorithm has significant accuracy and detection speed advantages, providing a new technical solution for realizing real-time monitoring and early warning of fatigue driving.

With the rapid development of the integrated circuit industry, the complexity of the internal structure of the chip is increasing, to improve the inspection efficiency of defects and analyze the types of defects, is the guidance of the production process and the test equipment of each process segment is an important guarantee to improve the output. This paper develops a defect detection device for the back channel of wafer, splices and analyzes the captured images, and compares and classifies the types and sizes of defects on the surface of existing silicon wafer samples through image analysis. Combined with deep learning technology, automatic identification and classification of surface defects is realized. The system has good real-time performance, high efficiency and strong industrial application detection ability. The technical problems of defect detection such as long time, low efficiency and strong subjectivity are optimized.

Hollow nanostructure not only can reduce the transmission distance of ions and electrons, but also hinder the volume expansion during the charge and discharge cycle, as a result for excellent rate and cycling performance. In this work, hollow nanocubic Co-NC was synthesized by calcination method using cubic Co-NC as precursor. The capacity of Co-NC composite is around 102.20 mA h/g in the potential range of 0.01-3.00 V at the 0.10 A/g after 320 cycles.

Based on big data technology, this study designed and implemented a substation hazardous waste tracking and early warning system. Through detailed evaluation and comparative analysis of system performance, the results show that the system has excellent performance in data processing speed, early warning response time and prediction accuracy, which is better than the industry average. The experimental results show that the system has high efficiency and reliable operation characteristics, and has important application value in hazardous waste management. However, a number of issues were also identified during the experiment, including large-scale data processing performance and data quality challenges. Some suggestions for system optimization and improvement are put forward, including optimizing system architecture, improving data quality and improving early warning mechanism.

Elliptic Curve Cryptography(ECC) is widespread in practical applications because compared to other asymmetric cryptosystems, ECC can achieve the same level of security with shorter key lengths. With the development of the Internet of Things(IoT), a great deal of constrained devices may require software ECC implementations. However, point multiplication, which is the most time-consuming operation, restricts the use of ECC in these devices. In order to improve the efficiency of the SM2 algorithm based on ECC in IoT, we propose an efficient and secure distributed scheme for SM2 decryption. In this proposed scheme, the SM2 private key is covertly divided into key shares of different lengths, which are then distributed to various devices to ensure the privacy of the key. Additionally, high-performance devices handle auxiliary point multiplication calculations, improving the decryption speed and alleviating the burden of local calculations. Performance analysis and experiments demonstrate that the scheme achieves nearly 3.8 times faster decryption than the standard SM2 algorithm, making it suitable for IoT applications.

With the continuous development of new power system, there are new changes and requirements for the types and functions of physical assets in power grid and the relationship between physical assets. In view of the current situation of assets in power grid enterprises, it is necessary to carry out efficient physical asset management, evaluate and manage physical assets. In this paper, according to the characteristics of grid physical assets under the new power system, the integrated fuzzy evaluation method based on Pythagorean fuzzy sets and contributing characteristics is used to evaluate the grid physical assets. Evaluating the physical asset condition of power grids in each region with this method, the results show that the best comprehensive level is in Jiangsu Province, and the worst is in Tibet Autonomous Region. Therefore, for the regions with poor asset quality, it is recommended to strengthen the technical transformation and upgrading of assets, and to strengthen the overhaul of assets with serious abnormalities in order to ensure the normal use of assets.

Crowdsourced testing is a useful way to quickly identify defects and save time. However, the quality of crowdsourced testing reports varies widely since most testers are non-professionals. Effectively reviewing these test reports has become an urgent issue. To address this issue, researchers have proposed various techniques for classifying, clustering, and prioritising reports. These methods often directly extract text-related features without further understanding the textual content. By delving deeper into the textual information of crowdsourced testing reports, sentences within the text descriptions can be categorised into two types: defect descriptions and step descriptions. This paper presents a prioritisation method (TPLSI) for crowdsourced testing reports to enhance the efficiency of report review. Initially, a pre-trained LSTM is employed to classify the textual part of the crowdsourced testing reports. Subsequently, extract textual and visual features from the reports, merge these features, and use clustering techniques to group similar reports. Finally, the inspection sequence of the testing reports is determined by considering the severity of the issues reported in the crowdsourced testing reports. Furthermore, the experiments in this study were conducted on industrial crowdsourced projects from six different domains. The results indicate that the proposed method is capable of detecting errors more quickly within a limited timeframe.

Aiming at the limitations of traditional machine learning in data feature extraction, an integrated Capsule Network and multi-layer Bidirectional Gated Recurrent Unit (Bidirectional Gated Recurrent Unit) were proposed. BiGRU and the BACN fall detection and recognition model of Attention mechanism. In this model, the capsule network is responsible for capturing the spatial features, while the bidirectional GRU module effectively extracts the hidden temporal features of the data. At the same time, the attention mechanism is used to further highlight and screen the fine-grained features to improve the accuracy of model recognition. Through repeated experimental training and model optimization, we determined the optimal hyperparameters of the network and constructed an efficient fall detection model. The accuracy of the model on Sisfall, Mobiact fall data set and self-collected data set reached 98.3%, 97.8% and 94.4% respectively, which fully proves its effectiveness in practical application.

Image-to-image unsupervised translation involves the visual level and the graphics level. In its process, the main principle of its performance is to learn the correspondence between two different image sets without using specific image pairs. However, without an image paired data set, under such conditions, if you want to obtain the best possible results, it is difficult to avoid a deeper network structure, resulting in a huge model, which means more calculations. This paper introduces an efficient module using parallel feature extraction, which helps reduce depth while retaining high performance. It is valuable to prioritize the structural aspects of image translation to enhance the coherence and quality of the translated output, thereby mitigating the risk of producing fragmented results. This article uses an attention network - the pyramid attention network, so that it can pay more attention to high-level features of images. The results show that this method is indeed optimized in image translation.

Despite deep learning models have made significant breakthroughs in the magnification and precision of single image super-resolution (SISR) reconstruction. However, current methods which focus on extracting rich texture details, always ignores the influence of byproduct of artifacts in high factor super-resolution construction. Therefore, a super-resolution method based on improved diffusion probabilistic model (DPM) is proposed to eliminate artifacts while obtain texture details. Firstly, residual skip path (ResPath) is designed to enhance the constrain of initial information on the reconstruction result to mitigate artifacts. In addition, overfitting may also generate artifacts, so blueprint separable convolution (BSConv) is introduced to reduce redundant network parameters. Secondly, in order to retain the texture details, the efficient channel attention (ECA) block and the enhanced spatial attention (ESA) block are incorporated to extract more comprehensive channel and spatial implicit information. The effectiveness of this method is verified on 4 public datasets with ×8 factor. Compared with other advanced SISR methods, the proposed method achieves the best performance on perceptual index (PI) and visual perceptual quality.

Multivariate time series classification (MTSC) methods as a key research topic in the field of data mining, attracting extensive attention and in-depth study. However, existing multivariate time series classification methods often struggle to simultaneously capture both the complex interactions between variables within individual samples and the global relationships among different samples. To address this issue, a spatio-temporal graph attention network (STGA) classification model for multivariate time series is proposed. Firstly, a new feature extraction module comprising convolutional layers and spatio-temporal attention layers is introduced, aiming at capturing the relationships between multivariate time series variables. Secondly, a graph construction method based on Triangle Inequality and Point Clustering Dynamic Time Warping is designed to capture the relationships among samples. Based on the above techniques, a classification model based on graph convolutional modules is proposed. This model fully explores the relationships among samples and between variables within samples to obtain more accurate feature representations, thus enhancing classification accuracy. Experimental results on the UEA dataset demonstrate that this algorithm outperforms other comparative algorithms in time series classification tasks.

This study is based on the network teaching platform of colleges and universities. We use advanced big data analysis technology to conduct a comprehensive and objective evaluation of teachers and their teaching quality. We deeply mine all kinds of data of online learning, including course information, students' learning trajectory, subject scores, etc. elaborate what methods you used. Based on the analysis, we reveal the internal correlation of learning effectiveness, identify the characteristic differences of student groups, and predict learning results. Based on these analyses, we are committed to building an online education model that suits China's national conditions in order to improve the quality and effect of education and teaching.

Rumor propagation is affected by various factors, and its mechanism is still under exploration. Based on the principle of drought propagation, this paper added rumor deniers to the classical rumor propagation model, and established a CDTR propagation model of network rumors. Then the model parameters were analyzed to explore the influencing factors in the process of rumor propagation. The experimental results show that rumor deniers play a very important role in inhibiting the spread of rumors. That is,government departments should increase positive intervention or official media should increase the spread of true information to increase the proportion of rumor deniers, so as to achieve the purpose of rumor governance.

After years of accumulation, our department has accumulated a large amount of knowledge, which is stored in Oracle and MongoDB database and uniformly displayed in the knowledge management app of the IT Service Management (ITSM) System. In order to quickly and efficiently search for knowledge, this project is based on Elasticsearch (ES) and implements a knowledge full-text retrieval system. Compared to traditional SQL based retrieval methods, the retrieval platform built in this project is capable of retrieving knowledge attachment content and sorting it by relevance. The implementation effect after deployment shows that it can effectively improve the efficiency of knowledge retrieval and effectively display the hit content in attachments.

With the ongoing advancement of the State Grid information system, accurate forecasting of cloud data center resource loads becomes pivotal in enhancing the efficiency of cloud computing resource utilization, minimizing power consumption, and optimizing operational costs associated with human resources. Addressing the challenges of low prediction accuracy and the intricate time series characteristics of historical data, a novel approach, ABLCL, is proposed. This method leverages a multi-model fusion strategy incorporating ARIMA, BPNN, and LSTM models, integrating time series factors. Through comparative experiments on public datasets, the efficacy and stability of the proposed model in predicting cloud computing resource loads are validated, demonstrating superior performance over alternative forecasting models.

To realize the effectiveness and real-time of computing power scheduling of power grid cloud platform, this paper proposes a load forecasting and computing power scheduling method of power grid cloud platform based on distributed computing. The short-term and ultra-short-term computing power of each virtual resource pool load task is predicted. Based on the decomposition-ensemble model, LSTM (long short-term memory) model and XGBoost model are used to predict the computing power. The cloud platform center server groups each virtual resource pool according to the prediction curve, and only the virtual resource pool in the same group can carry out resource scheduling and information transmission. The scheduling method is verified by experiments, and the test results are good.

The radar antenna pedestal usually works at very low speed and the vibration signal of antenna pedestal is composed of multiple interference signals. Influenced by multi-component signals and strong noise, the characteristics of early faults are weak. To solve the problem, a fault diagnosis method for radar antenna pedestal based on VMD and energy entropy reconstruction is proposed. The vibration signal is decomposed into multiple IMFs using VMD. Then fault-informative IMFs strong are selected based on cross- correlation and energy criterion. Finally, the analytical signal is reconstructed by selected IMFs and demodulated. Experimental results show that the proposed method could strengthen modulation components caused by weak faults and the assembly fault of the antenna pedestal is successfully diagnosed in real application.

With the rapid development of information technology, digitalization has gradually integrated into the daily lives of modern people, becoming an indispensable and important component. Especially, DingTalk software, as an office application that integrates communication and collaboration, has been widely used in daily office work. This paper studies the authenticity of electronic data produced by DingTalk software and its evidence collection methods. By analyzing the data storage mechanism of DingTalk, it identifies the challenges faced by electronic data evidence collection and proposes a method to ensure the authenticity of data. Meanwhile, it discusses strategies to counter data tampering techniques. The paper ultimately offers suggestions to improve the accuracy of evidence collection, aimed at ensuring the legal effectiveness of electronic evidence to meet security challenges and evidence collection needs.

This paper introduces a classification methodology of aircraft requirements on Quality Function Development(QFD) , which divides quality into Must-be Quality,One-dimensional Quality and Attractive Quality. Accordingly, the fuzzy clustering method is used to analyze the customer requirements in details, which determines the subordinate relationship of each kind of requirements information, outputs requirements classification results. Finally, the research method is applied to the classification design of aircraft requirements capture in order to prove its effectiveness.

Knowledge extraction is a significant issue in fields including knowledge graph construction and natural language processing. Triple extraction, as a crucial issue within knowledge extraction, is intended to acquire structured triples from unstructured texts, that is, to identify entities, their types and relations. Typically, it provides technical support for downstream tasks such as information recommendation, semantic search, and question answering. However, present triple extraction methods fail to sufficiently utilize the information related to relation types and entity types during the sentence embedding generation, and cannot fully exploit the auxiliary semantic information about relation for entity extraction. To address the aforementioned challenges, this paper proposes a triple extraction approach founded on Meta-type Prompt learning and Bidirectional relation Collaborative Attention (MPBCA). That method utilizes meta-type prompt learning which introduces relation types, entity types and their correlations to optimize the token embedding. Thereby, the generated word embeddings can be more adaptive for the triple extraction task by exploring the intrinsic logical connections of target triple components. Furthermore, our approach designs a bidirectional relation complementary attention mechanism to identify entity head and tail positions. That mechanism not only strengthens the semantic modelling capabilities for relations between entities, but also improves fault tolerance and captures the potentially accurate target triples by using bidirectional sentence structures, not unidirectional structures. The results of experiments on two public datasets show that our triple extraction approach MPBCA proposed in this paper is superior to the existing methods, confirming the effectiveness and superiority of the proposed model.

In response to the low accuracy and high false alarm rate of anomaly identification in energy big data, this paper proposes a dynamic weighted identification method for energy big data anomalies based on optimal sub segment deep learning. This method deeply mines the optimal sub segments in energy big data, uses deep learning techniques to automatically learn advanced feature representations in the data, and combines dynamic weighting strategies to accurately identify abnormal behavior. By selecting the optimal sub segment, this method can focus on the data features most relevant to abnormal behavior, improving recognition accuracy. The experimental results show that the proposed method can capture subtle changes in abnormal energy big data, further enhancing the accuracy of recognition and reducing the false alarm rate. In summary, the method proposed in this article has high accuracy and low false alarm rate in the field of energy big data anomaly recognition, providing strong technical support for decision-making, optimization, and management in the energy industry.

This study advances computational modeling techniques by integrating propagation and logistic models with transportation network analysis. Utilizing a dual-model approach, we explore the application of compartmental models (SIR model) and population dynamics (logistic model) to simulate and predict dynamics in a structured network. Model I (Propagation Model) employs a SIR framework to predict the spread through potential transmission routes among cities, utilizing differential equations to model infection-like spread dynamics. Model II (Logistic Model) focuses on population dynamics through a logistic growth equation, which is tailored to describe population variations over time under certain constraints. The innovative aspect of the research is further demonstrated in Model III, which incorporates traffic data (AATD) to assess how major traffic flows might influence propagation patterns. Model IV, the road traffic model, extends the analysis to explore how transportation nodes and pathways can be integral to the dissemination process, leveraging network theory to map potential spread scenarios. By repurposing epidemiological and logistic methodologies within the context of network and transportation analysis, this research not only showcases the adaptability of these models outside their conventional biological applications but also highlights their potential in tackling complex systems analysis, providing a robust framework for addressing diverse challenges through mathematical modeling. This approach exemplifies the intersection of computational modeling with practical applications, offering a blueprint for future interdisciplinary research.

In recent years, with the increasing market demand for Cordyceps sinensis, the identification of artificial breeding varieties and wild varieties has become particularly important. The traditional identification method of Cordyceps sinensis is time-consuming and inefficient. Based on the advantages of hyperspectral detection technology, a simplified spectral method based on PCA inter-class distance evaluation was proposed. By performing PCA dimensionality reduction on the hyperspectral data of Cordyceps sinensis samples under different spectra, extracting principal components, calculating inter-class distance, and selecting the most recognizable spectra, the identification process was simplified. The preliminary results showed that the inter-class distance between artificial breeding and wild Cordyceps sinensis was the largest when SWIR-1 (1000-1800nm) spectrum was used only. When the Mahalanobis distance is used as an evaluation index, when only SWIR-1 spectrum is used, the Mahalanobis distance value reaches 5.21, which is larger than that of other spectra only. By simplifying the spectrum, the identification of artificial breeding and wild Cordyceps sinensis using only SWIR- 1 spectrum can be effective, and the hardware cost of the spectrometer can be reduced.

In order to address the issue of optimal pathfinding in autonomous mobile robot navigation, an improved path planning scheme based on the traditional A* algorithm is proposed. Firstly, aiming at the problem of collision and low planning search efficiency in traditional A* algorithm, a safety distance is set, and the Euclidean distance calculation method is selected in the heuristic function, constructing a cost function with dynamically adjustable heuristic function weights. Secondly, to smooth out the non-smooth paths generated by traditional A* algorithm, a Bezier curve smoothing algorithm is employed for path smoothing. Then, through simulation experiments, the significant improvements of the algorithm in terms of planning efficiency, safety, and path smoothness are verified. Finally, through autonomous navigation experiments, the feasibility of the improved A* algorithm is demonstrated. The research demonstrates that the algorithm designed in this paper can plan the optimal path and safely and efficiently reach the target point.

The power grid control system is facing the threat of cyber attacks. Based on the Naive Bayes algorithm is an effective way to detect the abnormal traffic of the grid network. First, design the abnormal detection framework based on the semantic characteristics of Naive Bayes. Then, the data pre-processing of the data in the industrial control system was conducted, and the feature vector was constructed according to the functions, instructions, points and other fields in the network data of the industrial control system. However, designing a Naive Bayes Gaussian abnormal detection model, the state of normal/abnormal systems in the network data stream, control behavior recognition and detection. Finally, this method detects the abnormal MMS protocol of the power grid system to verify the detection effectiveness of the method.

Mission planning for UAV swarms is an NP-hard combinatorial optimization problem. Genetic algorithms find approximate solutions to this problem within an acceptable computational time. However, in complex real combat scenarios, the computation of individual fitness becomes exceedingly difficult, which severely restricts the efficiency of genetic algorithms in exploring the solution space for mission planning. To solve this problem, this study proposes an improved genetic algorithm based on a weighted graph attention network. This method encodes the chromosomes of individuals into a graph structure and makes predictions about the fitness of individuals by introducing a graph neural network, which ensures prediction accuracy while reducing the complexity of fitness calculation. To eliminate the prediction error of the graph neural network as much as possible, the actual fitness value is employed for corrections when required.

The emergency response of transnational oil and gas pipelines involves multi-party linkage, and there is uncertainty in the actual reporting and execution process. In this paper, we study the intelligent body linkage scheme modeling language and establish the linkage scheme optimization algorithm, and take the crude oil pipeline leakage accident as an example to realize the optimization and representation of the linkage scheme. The results show that in the process of emergency response to large-scale accidents involving the linkage of multi-level agencies, the addition of emergency response agencies to form an organized trans-level information reporting helps to improve the disposal efficiency, and the optimal scheme is accurately selected by comparing the response end time, cumulative disposal efficiency and yield of the three linkage models.

In order to reduce the loss value in the data recovery process and ensure the integrity of the procurement information, a data recovery method for online procurement of power grid materials based on local weighted reconstruction is proposed. According to the principle of clustering interval set division of online procurement data of power grid materials, the missing data of online procurement of power grid materials is mined by using the density peak clustering method; the characteristics of procurement missing data are extracted by the combination model formed by the generative model and the discriminant model in the generative adversarial network through the dual semantic perception of the characteristics of the missing data in the procurement, the reconstruction of the missing data in the procurement is realized; after the reconstructed online procurement lost data is detected by the autocorrelation fusion filter detection method, the lost data is optimized and restored by combining the local weight reconstruction, And by using the local weighted learning method to realize the optimal control of lost data recovery. The experimental results show that the method can effectively cluster the missing data; the Rand coefficient is adjusted to keep above 0.5; when the reconstructed data is reconstructed 90 times, the position of the reconstructed data is basically the same as the original data; the restored data is more complete and no loss value is generated. It can effectively mine the lost data of procurement, and realize the recovery of lost data of online procurement of power grid materials.

In response to the complex construction content and inconsistent construction progress of various trades in electrical substations, this article uses LiDAR point cloud technology to achieve rapid modeling of substations and transmission lines. With the help of post-processing technology from Unity 3D engine, a BIM model that can be used for substation roaming inspection and construction design is constructed. At the same time, a substation collaborative construction detection system was developed, which coordinated the progress of collaborative construction operations for different types of work, providing support for rapid construction and early planning of substations.

In order to isolate each chromosome more accurately and completely, a two-stage method of processing overlapping chromosomes was proposed. In the first stage, In the first stage, we propose a DA-UNet (dual attention U-Net) network based on improved U-Net network to segment overlapping and non-overlapping regions of chromosomes. In the segmentation network, the encoder part was replaced by the pre-trained ResNet50 extraction feature to accelerate the convergence speed. A dual attention module was designed at the jump junction, combining the features of a multi-level encoder and a parallel decoder. In the second stage, the output of the segmentation network was matched geometrically to obtain a single chromosome. Experiments were carried out on the chromosome segmentation dataset, and the overall accuracy of this method reached 94.70%, which was 5.16% higher than the existing chromosome segmentation and reconstruction methods. Experimental results show that the proposed method can achieve accurate segmentation and reconstruction of chromosome images.

This paper introduces the N10 security chip, a cutting-edge processing solution equipped with advanced features to meet the evolving security demands of both national and business levels. Developed using the RISC-V intelligence system and manufactured using a 28nm tape-out process, the N10 chip offers unparalleled performance. Notably, the chip supports reconfigurable cryptographic algorithms, including SM4, SM3, AES, DES, and more than 20 others, ensuring adaptability and compliance with stringent security standards.

This study investigates the development of single-page applications using the Taobao Midway.js framework, Prisma ORM, and Vue.js. Midway.js, a robust Node.js framework, offers extensive features and flexibility for efficient backend service development. Prisma ORM simplifies database operations with a user-friendly abstraction layer, reducing data access complexity. Vue.js enhances front-end development with its intuitive component-based approach and responsive design. The research begins with an overview of the theories and technical aspects of Midway.js, Prisma ORM, and Vue.js. It then details the full-stack development process with these technologies, covering project architecture, backend implementation, database ORM operations, and front-end development. Case analyses and practical experience summaries confirm the viability and effectiveness of the Midway.js+Prisma ORM+Vue full-stack development approach, offering new solutions and insights for the web development field.

With the development of the Internet and IT technology, single application architectures are difficult to meet the needs, and distributed system architectures have become the mainstream. The traditional RPC call mode will result in strong dependency and high coupling between applications, which cannot realize Asynchronous communication and application decoupling, and cannot adapt to high concurrency scenarios. In addition, with the continuous expansion of business scale, some business scenarios may encounter sudden situations of high traffic, which traditional architectures are difficult to resist. A mechanism is needed to peak shaving and buffer these sudden traffic. To address the above issues, we propose a distributed message middleware that can easily access large-scale application clusters. The distributed System Architecture Evolution evolves to the message oriented architecture (MOM), realizing event driven and Loose coupling. Application decoupling and loose coupling can also be achieved through asynchronous messaging.

With the development of the digital economy, tourism resource promotion of information technology has become the key to the transformation and upgrading of the cultural and tourism industry. The traditional collection of visitor information relies mainly on manual records, which have problems such as low data accuracy, insufficient real-time, and difficulty in predicting visitor behavior. To change the current state of low promotion of information technology of the Grand Canal National Cultural Park, it is extremely important to establish a visitor information management system. Based on Java SE, this paper designs and develops a visitor information management system with a clear interface, convenient operation, and strong interactivity, which realizes the organic combination of data collection, big data analysis, and tourism resources. It can effectively improve the intelligence level of the Grand Canal National Cultural Park, optimize the visitor experience, and better manage visitor information.

In order to quickly respond to the real-time monitoring requirements of the plane and shorten the task preparation cycle, the mapping template of parameters and the controls in the monitoring interface is designed. NPOI analyzes the Excel template, and the C # reflection mechanism completes the data source binding of the controls. On the basis of this, we can increase the effectiveness of monitoring software development. This method is currently being utilized to successfully ensure the flight test duty. And it can cut the development cycle in half and the amount of software code by at least 50%, as confirmed by flight test. As a consequence, the development and maintenance costs of software monitoring may be decreased, and the reusability and scalability of software modules can be enhanced. The results demonstrate that this strategy can greatly reduce the amount of software code and avoid the complexity of repetitive editing by developers. It offers a fresh concept for the creation of software for flight test monitoring.

MBSE, or Model-Based Systems Engineering, is an approach that facilitates system design, analysis, verification, and validation. It is integral throughout the entire product development life-cycle, from initial conceptualization to the eventual end of the product's lifespan. This text focuses on the preliminary design stages of a smart cabin management system using MBSE methodologies, outlining the process of formulating a system concept guided by MBSE principles.

Closed box structure is usually adopted in traditional reinforced server. Once fault appears, the managers need to power off the chassis and fix it up, which brings up some troubles of system maintenance. This paper introduces a design scheme of embedded intelligent management system based on ARM and FPGA for reinforced server. The server remote management is realized by chassis management module and IPMC module. The experimental result shows that the designed intelligent management system can improve system management efficiency, reduce maintenance cost and enhance server maintainability and stability.

The abundant degrees of freedom provided by a large number of antennas at the base station are harnessed in MU-MIMO-OFDM downlink systems. We formulate the OFDM modulation, MU precoding, and peak-to-average power ratio (PAPR) constraints into a non-convex optimization problem for transmit power minimization. This problem is subject to predefined thresholds for PAPR of each antenna and multi- user interference (MUI). Instead of using existing relaxation- based convex optimization methods, we directly address this non-convex PAPR-aware precoding problem using the linearized alternating direction method of multipliers (LADMM). Simulation experiments confirm that the proposed LADMM method significantly reduces PAPR and minimizes symbol error rate (SER). Importantly, compared to existing methods, our LADMM method exhibits faster convergence speed and lower complexity

Gear reducers, as crucial mechanical transmission devices, have been extensively utilized in various fields. Owing to the complexity of gear reducer structures and the diversity of operating conditions, maintaining these devices poses significant challenges, and training technical personnel can be both time-consuming and costly. To address this issue, this paper proposes a gear reducer operational state maintenance system based on Virtual Reality (VR) and Augmented Reality (AR) technologies. Utilizing VR, the system provides disassembly and assembly training, displays component structures, and conveys relevant parameters, while AR facilitates maintenance guidance and remote expert assistance. This integrated system combines a VR training platform with an AR maintenance support tool, enabling not only VR simulation for virtual assembly training but also AR-based on-site maintenance guidance and remote consultancy, making upkeep tasks more convenient and efficient.

Due to the continuous upgrading and iteration of software products, many off-site software upgrades have been brought about. In order to improve the convenience and reliability of software upgrades after product delivery, based on DSP secondary boot technology, a DSP software online upgrade method using TMS320F28235 is proposed. The architecture of the online upgrade hardware platform is described, the DSP software secondary boot loading process is introduced, the generation method of upgrade files and program update instructions are explained, and the transmission process of upgrade files and FLASH programming process are described in detail. This method can achieve online upgrade of DSP software without changing the hardware state and without using an emulator. In addition, this method can effectively avoid upgrade failures due to transmission problems, improving the stability and success rate of online upgrade.

This study presents a Bluetooth data processing platform designed using MATLAB App Designer, aimed at addressing the data reception and processing challenges encountered in wearable device algorithm research. This platform provides convenient and fast data analysis and visualization tools, mainly used to solve the problem of cross platform transmission after wearable devices collect data. The platform is structured into modules for data reception, processing, and visualization, with defined protocols for Bluetooth data reception and transmission. Additionally, we designed a real-time dynamic heart rate algorithm based on the collected data and integrated it into the platform to verify its reliability and stability. This research provides essential tools and platform support for algorithmic research in wearable devices and other Bluetooth-enabled devices.

The widespread adoption of Building Information Model (BIM) technology has profoundly impacted the architecture, engineering, and construction (AEC) industry, with the Industry Foundation Classes (IFC) serving as a pivotal open-standard file format facilitating BIM data exchange. Despite IFC's significant advantages in promoting interoperability and data sharing, challenges such as complexity, performance issues, and potential information loss persist. This study explores the utilization of the IFCNetCore dataset and training with Vision Transformer (VIT) models to process IFC images and achieve intelligent classification of BIM elements. Compared to traditional Convolutional Neural Networks (CNNs) like ResNet152, the proposed approach exhibits superior robustness and generalization, thereby enhancing efficiency in architectural design, construction, and management while fostering support for the development of intelligent buildings and virtual simulation domains.

Since the original dung beetle optimization algorithm suffers from the drawbacks of slow early convergence, low solution accuracy, easy to be trapped in locally optimal, a dung beetle optimization algorithm incorporating Lévy's flight strategy and lens imaging learning is proposed (OLBOA). Introducing firefly perturbations in the DBO algorithm for updating the position of the rolling ball dung beetle to prevent it from getting stuck in a partial optimum; The Levy flight strategy is used to effectively optimize the ability of the algorithm to search globally and locally; Mutation of optimal individuals in a population using convex lens imaging reverse engineering strategy to form new individuals and to increase the acceleration of acceleration of convergence of the algorithm. Compare the improved algorithm with other cluster intelligence methods on four of the functions tested. The outcome demonstrates that the modified DBO algorithm converges faster and the optimization accuracy is higher, which verifies the feasibility of the algorithm.

E2V's EV10AQ190A sampling chip introduces a time-interleaved ADC architecture as a means of improving the sampling rate. Still, the non-ideality of the circuitry leads to the occurrence of timing mismatch among the sub-ADCs, which seriously affects the overall performance. To address this challenge, this study proposes a temporal mismatch error calibration scheme for the EV10AQ190A chip. The scheme incorporates statistical theory for mismatch estimation and a farrow-structured fractional-delay filter optimized by the Lagrange interpolation algorithm to construct an efficient and simple blind calibration system. Experiments verify the effectiveness of the system through simulation and actual sampling. The results show that the time-interleaved ADC's Spurious-Free Dynamic Range (SFDR) is improved by an average of 17.7474dB and the Effective Number of Bits (ENOB) is increased by an average of 2.3 bits after calibration.

In recent years, UAVs have played an increasingly important role in the military and civilian fields, especially UAVs swarm with high mission execution and diverse execution scenarios. The operator should have high proficiency in manipulation due to the complexity of UAV swarm, and cognitive analysis is an efficacious method used to train and assess operators of UAV swarm. This paper put forward an UAV swarm simulation system for operator cognitive analysis. First the requirements of simulation system for operator cognitive analysis is introduced. Subsequently the architecture and detailed module design of simulation system is proposed. Finally, the result of vehicle searching mission simulation and EEG data of operator verify the effectiveness of the simulation system.

The 'Technology Winter Olympics' was a major highlight of the 2022 Beijing Winter Olympics, aimed at exploring better solutions for future urban living worldwide. This system, focusing on ice rinks as its primary application scenario, addresses the issue of elderly and children skaters who fall and cannot receive timely assistance due to mobility problems. The project utilizes programmable devices, FPGA (Field-Programmable Gate Array), to collect and process signals from multiple sensors, including angles, acceleration, and rotation speed; it implements fall detection through differential threshold method and uploads the results to a mobile app via 4G signal. The system's effectiveness was validated through online calibration in the ice rink, where, after about ten experimental tests, the fall detection success rate was 85%, and the success rate for other functionalities exceeded 90%. The results demonstrate that the system can accurately detect falls while ensuring high operational efficiency, and it holds potential for extensive application in public places for detecting falls among the elderly and children.

Due to the existence of a large number of different brands and models of equipment and systems in the power system, the CPU control system in the power system is prone to system integration and compatibility problems. In order to ensure the smooth application of CPU control system in electric power system, the research on the application of CPU control system in electric power system based on ARM processor is proposed. The CPU control system architecture of ARM processor is established, and the irregular characteristics and task attributes of computational tasks such as sparse matrix operations in power system topology analysis are analyzed. Subsequently, a general optimization criterion for GPU parallel algorithm design is proposed. Experimental results show that the above CPU control system for topology analysis algorithms in power systems achieves a 6-fold acceleration ratio relative to the 8-threaded CPU algorithm on a 30,000-node system, and more than a 4-fold acceleration in the full process of topology analysis.

Modern CPU’s memory data access speed is much lower than the calculation speed. Data visits are often a bottleneck of program performance. Cache has a higher data access speed compared to memory, which can load data in memory in advance to improve the efficiency of CPU access to memory data. The cache is usually a multi-level structure. The closer to the CPU level, the faster the cache visits. The cache miss is that the data which CPU needs to be visited is not in the current cache. The processor needs to read data from the next level of cache or memory. The occurrence of cache misses will reduce the memory performance of the CPU. Prefetch can load the data visited by the processor to the cache in advance, reduce the occurrence of a cache miss, improve the efficiency of access to access, and then improve program performance. Related scholars have studied key technologies in different scenarios and on key technologies such as prefetch data, prefetch distance, and prefetch methods. However, related studies are mainly aimed at specific scenarios or access modes, and less prefetch research on general programs. For the prefetch optimization problem of general programs, this article proposes value evaluation model and automatic implementation method of software prefetching in general program optimization. The model can automatically analyze the data that needs to be prefetch and the method can automatically insert the data prefetch into the program. Based on the value evaluation model and automatic implementation method, this article implements the automatic software prefetch tool SP-Frame under the LLVM compiler. In the Crono Benchmark, the SP-Frame tool can reduce 54 %of the L1 cache miss on average, and the average speedup has achieved 23%.

The fusion of panchromatic and multispectral images is a prerequisite for obtaining multispectral images with high spatial resolution (HRMS). The traditional data processing mode requires that the acquired panchromatic data and multispectral data be transmitted to the ground through the star-ground link, and then a series of correction, registration and fusion operations are conducted in order to obtain HRMS. This processing mode has low timeliness and cannot meet the demand of on-board real-time processing. To solve this problem, we perform efficient registration and fusion processing for the images containing the target region acquired in real time during satellite orbit. Specifically, we propose a fast virtualized registration method based on object-space-consistency, which is capable of obtaining accurately registered images under the premise of ensuring real-time performance. On this basis, we utilize GPU acceleration to parallelize the time-consuming steps of the traditional fusion algorithm, thereby achieving rapid fusion processing while guaranteeing the fusion performance. Through experiments on an on-board simulation system, our method has been able to achieve 15.10 times speedup compared to the traditional registration and fusion processing while guaranteeing good performance.

Multi-domain unmanned vehicle joint operations is an important application direction of unmanned vehicle, The efficiency and stability of cooperative command system is the key to improving the flexibility and effectiveness of unmanned vehicle. A typical multi-domain unmanned vehicle joint combat system is presented and open cooperative command system is designed in this paper. Finally, the key technologies such as interconnection and interoperability of heterogeneous unmanned vehicle, Multi-source intelligence processing based on video and image, Cooperative mission planning of heterogeneous unmanned vehicle is analyzed. The system realizes Cooperative Command of Multi-domain Unmanned Vehicle, Improving the effectiveness of joint operations.

There are various challenges in text detection of handwritten document images, such as dense text, complex layout, and noise interference. In order to better detect handwritten text lines in document, this paper combines ViT with PAN++ network models, utilizes ViT's powerful image understanding ability, and combines pixel aggregation algorithms to propose the PAViT network. The improved network model can effectively address various challenges in document images, such as dense text, mixed print and handwriting, background noise, etc. It accurately detects handwritten text lines in document images and performs experiments on the SCUT-HCCDoc dataset, achieving good results.

Unsafe driver behavior is a significant cause of traffic accidents, and accurately identifying and warning of dangerous driving behaviors is crucial for improving road traffic safety. Single-modal data alone is insufficient to comprehensively describe complex driving behaviors. Multimodal data fusion technologies, by integrating heterogeneous information sources such as vision, speech, and motion, provide robust support for driving behavior analysis. This paper comprehensively reviews the current status, key technologies, application cases, challenges, and future development trends of safety driving behavior warning models based on multimodal fusion. It introduces typical architectures and algorithms of multimodal deep learning models, such as early/late/mixed fusion strategies, attention mechanisms, and temporal modeling, and analyzes application scenarios such as intelligent driving assistance systems and driving simulation training. Finally, it highlights challenges such as model performance, data quality and computational requirements, and discusses future directions. The development of multimodal safety driving behavior analysis technology will make important contributions to the construction of intelligent transportation systems.

When the parameters of material or size of structure are uncertain, the optimization result of the structure will also be uncertain. In this paper, interval mathematics is used to define the uncertain parameters, and the optimal mathematical model of the uncertain problem is established. The interval objective function with respect to the uncertain structural parameters is transformed into a deterministic function and the deterministic constraints are obtained from the uncertain interval finite element structural equation. The sparrow search algorithm is used to solve the uncertain structural optimization problem, chaos algorithm is adopted in order to enhance global search capability. The feasibility is proved by the example provided.

Network security is a critical concern in the development of interconnected and shared networks, given their vulnerability to various attacks. Intrusion detection technology has emerged as a crucial defense mechanism against malicious activities. Traditional methods face limitations in detecting complex intrusion behaviors, leading to a shift towards machine learning-based approaches. Deep learning models like Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory Neural Networks (BiLSTM) offer promising solutions by capturing intricate patterns in network traffic data. This paper introduces a fusion network intrusion detection model, leveraging Grey Wolf Optimizer (GWO) and Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM), addressing emerging challenges in network security. The model's efficacy was assessed on the CICIDS2017 dataset and compared against Random Forest and CNN-BiLSTM models. Results confirm the viability and efficacy of the proposed approach, offering a novel strategy for network intrusion detection.

With the advancement of technology and social development, significant changes have occurred in people's production and lifestyle, with prolonged sedentary work and study becoming the norm for contemporary individuals. However, incorrect or irregular sitting postures for extended periods have increasingly severe health implications, affecting people's learning and work efficiency, and even leading to spinal diseases, causing inconvenience in daily life. The identification and monitoring of human postures have always been a research focus. The core of this system lies in the pressure array sensors embedded in the seat, which can real-time collect the force data on the chair surface and perform in-depth analysis through the algorithm embedded in the hardware system, accurately identifying the user's sitting posture. Once the system detects an improper sitting posture, it will immediately display relevant information and issue reminders to guide the user to adjust to a correct sitting posture. The system also features data uploading and cloud storage capabilities. By uploading the sitting posture recognition results to the cloud platform, users can conveniently check their prolonged sitting time and the cumulative time of various sitting postures through a mobile phone app. This function not only facilitates users to understand their sitting habits, but also plays a corrective and reminding role, helping users establish healthy sitting habits.

With the continuous development of video surveillance in distribution station buildings, the pressure of safety risk prevention and control under the massive data on the edge has been increasing year by year. The traditional operation and maintenance methods that heavily rely on manual labor can no longer meet the requirements of lean management. The article proposes an artificial intelligence based intelligent terminal application scheme for the edge side of distribution rooms, which realizes on-site intelligent analysis of video images. Accurate identification of equipment status such as meter readings for daily inspection images; At the same time, in response to real-time video surveillance requirements, various security risks such as personnel intrusion can be detected in a timely manner. By analyzing actual inspection images, the safety risk detection rate of this method is over 98%, greatly improving the inspection efficiency of distribution station buildings and effectively reducing the workload of operation and maintenance personnel.

The "satellite data acquisition and ground processing" mode is difficult to meet the characteristics of detection and recognition in high real-time tasks such as target tracking and disaster detection caused by the large width, high resolution, and massive data of remote sensing images themselves. Propose a master-slave parallel computing architecture based on switching nodes. This architecture uses domestically produced PSOC (Programmable System-On-Chip) intelligent heterogeneous AI(Artificial Intelligence) chips, combined with SRIO(Serial Rapid IO) and Ethernet buses to achieve decoupling between broadcast image data distribution and parallel scheduling of MPI(Message Passing Interface) systems. On the basis of better utilizing the computing performance of single chip processors, it explores parallel computing modes with higher computing power and reliability. This mode enables image detection algorithms with large-scale and complex operator characteristics to be deployed and applied faster and more conveniently in spaceborne environments, which has certain reference significance and research value.

In order to solve the problem that the alarm of traditional CNC machine tools is not timely when faults occur, resulting in reduced machining accuracy and equipment damage, a condition monitoring system of CNC machine tools based on STM32 is proposed, which has the functions of remote alarm, real time monitoring, data acquisition and so on. When the network is normal, the status of CNC machine tools can be viewed in real time on the WeChat applet. The system test shows that the average relative errors between the system ambient temperature monitoring data, ambient humidity monitoring data, axis temperature monitoring data, and instrument measurement results are 1.9 %, 4.9 %, and 2.4 %, respectively, which have good timeliness and reliability.

With the deepening development of the gas and oil field, it is very important to data processing workload, low utilization of historical resources and high cost of introducing commercial software. Based on technical characteristics of BS and CS software architecture, this paper researched reservoir dynamic modelling and trace simulation system combined with adaptive neural network and FVM, and established an integrated platform for dynamic geological modelling and reservoir simulation over time. In addition, according to the newly added dynamic and static data, iterative fast update and tracking simulation of the reservoir model using OpenGL technology. The system fully demonstrated the practical significance of dynamic modelling and tracing computation, it is also play an important role in cutting production costs and optimizing production measures.

Music can relieve people's stress and is an indispensable element in modern life. There are already many music systems that can meet people's needs for listening to songs. Still, people have to download and update the corresponding software to use them, which is undoubtedly tedious and frustrating. With the rapid development of mobile Internet technology, WeChat applets are easy to spread, easy to use and easy to access. As a result, they have become an inseparable part of our life, work and business activities. This research aims to design and implement a music-playing system based on WeChat applets. Although there are many music software, the quick and easy-to-use features of WeChat applets make it necessary to develop music applets. When people want to listen to music, they need to open the app and select the music they like to play. In addition to the basic functions such as music playing, the app also has functions such as video square and song list.

In order to better develop the industrial biogas treatment engineering, it is necessary to conduct real-time data collection and monitoring of the anaerobic fermentation process, and then make statistics, analysis and scientific decision-making. In this paper, 5g^[1] and remote radio (lora) technology are introduced to monitor the anaerobic dry fermentation system, and various architectures of the online monitoring and control system for anaerobic fermentation are designed. Finally, the monitoring and diagnosis experiment was designed, remotely collecting the data of anaerobic dry fermentation process as the observation object and conducting statistical analysis. The results show that 5g and lora technology have strong applicability in the anaerobic fermentation process, which can effectively improve the level of online monitoring technology of biogas treatment.

RF magnetic probe array system installed on Experimental Advanced Superconducting Tokamak (EAST) is one of the important means to measure the parametric decay instability (PDI) of lower hybrid wave (LHW). Recently, the probe system was upgraded and tested to address the accuracy decline issue caused by insufficient high-temperature resistance of the original probes system. By replacing the probe material and solder, the high-temperature resistance of the probe array has been significantly improved. Experiments assessing the magnitude-squared coherence and measurement accuracy of the upgraded probe array are conducted and investigated. Results show that the improved probe has a coherence greater than 0.99, and it can distinguish the changes of parallel refractive index in an order of 0.2, providing more accurate data for the analysis of PDI experiments.

Recently, deep generative models have made significant progress in handling missing data, with the most notable being the use of variational autoencoders (VAEs). However, most of the VAEs methods project the original data into a lowdimensional vector space by the deep latent models. This can cause the loss of spatial information for matrix data, such as images, resulting in biased imputation. To address this issue, we propose an unsupervised learning method based on matrix-variate variational autoencoder, which generates a low-dimensional latent matrix through a convolutional network and reconstruct it through deconvolution. Besides, we proposed an iterative imputation method for better image reconstruction. Through testing on classic image datasets, the experimental results effectiveness of the proposed method.

This paper proposes a sound recognition algorithm for home environment based on deep learning. This algorithm uses the fusion feature extraction method to fuse the two feature extraction methods GFCC and FBank, and uses the fused feature value as the input of the ResNetSE network. Compared with related research, this method is optimized at the model input layer. By fusing two feature extraction methods, it replaces the traditional MFCC feature extraction method and improves the model's classification accuracy for home environment sounds. Experimental results show that the classification accuracy of the FBank-GFCC ResNetSE model proposed in this paper reached 97.4% on the Urbansound8K public data set. Compared with related work, the accuracy is improved by 6.4%. This shows that the algorithm has better performance in home environment sound classification tasks, and provides a more accurate and reliable solution for applications such as security monitoring in smart home systems.

Despite persistent efforts have been made in untangling the information source on complex network, a little attention is investigated for the relation of the underlying information spreading networks and information source detection. Here, we first exploit the relationship between users and the information spread among them, constructing a user-text coupled network based on two independent networks, which are users' network and texts' similarity network respectively. After that, an algorithm for information source detection based on bread-first search (ISD-BFS) is proposed, in which the spread centrality of nodes is calculated. Theoretical proof is also given that the probability of a node being the information source is proportional to the value of node's spread centrality. Experiments on the real-world data collected from Sina Weibo indicate the feasibility and effectiveness of the proposed algorithm compared with six other state-of-art methods.

With the rapid development of Internet and multimedia technology, people tend to express their feelings and views through video and other media. The key to sentiment analysis in user videos on social media is to fully utilize the embedded multimodal features, such as text, audio, and facial expressions, to establish efficient deep learning models. The traditional processing methods of simply fusing feature vectors or using multiple models to comprehensively predict results cannot effectively extract the intra modal characteristics and inter modal commonalities of multiple modal data, resulting in unsatisfactory accuracy of sentiment analysis results. In response to the above issues, this article takes monologue videos posted by users on social media as the specific research object and proposes a cross modal sentiment analysis model CMRL based on modal representation learning. By establishing constraints for both independent and fused modal modules, the fused modal module can fully consider the intrinsic characteristics of the modes. In order to enable the model to fully learn the intra modal characteristics, a loss function based on Pearson correlation coefficient is established by combining the sentiment analysis results of the independent modal module's speech modality, text modality, and expression image modality data with the sentiment analysis results of the fusion modal module. In order to prevent loss or confusion of intra modal features after feature fusion, the speech modal features, text modal features, and expression image features extracted by the Transformer in the independent modal module are fused, and a loss function based on Spearman correlation coefficient is established with the fused features of the fused modal module.

In this study, we propose an improved LPA* (Lifelong Planning A*) algorithm for multi-path planning tasks. This algorithm adds a step of path priority division based on the initial LPA algorithm and performs path planning in priority order. The study divides the planned paths into higher priority and lower priority, where the higher priority paths have less variation, ensuring the stability of the routes, while the lower priority paths have more variation. During the planning process, weighted processing is introduced for the nodes of the already planned higher priority paths to avoid conflicts between paths of different priorities and maintain the stability of the higher priority paths.

With the rapid development of modern traffic network, it is very important to accurately predict the trajectory of key traffic nodes. Aircraft navigation path, as a kind of data with complex temporal characteristics, is often affected by climate change and environmental disturbance, which makes the dynamic prediction of path characteristics a challenge. In this context, a novel trajectory prediction technique combining LSTM model is introduced in this study. The technique achieves accurate prediction of the target path through deep training of a large amount of data. The experimental results fully demonstrate the significant advantages of this method compared with traditional algorithms in the accuracy of flight target prediction and data processing efficiency.

Aiming at the problems of branch loss and fracture in the fundus retinal vessel segmentation algorithm, a blood vessel image segmentation method based on an improved simplified pulse coupled neural network (ISPCNN) and a gray wolf optimization algorithm with chaotic mapping and inverse learning (MLGWO) is proposed. First, the input domain and connection domain of PCNN network neurons are simplified, the gradient information factor is added to the internal activity items, the model parameters are reduced, and the binary pulse output is changed to multilevel gray intensity output; second, by improving the population initialization and the position update mechanism, a nonlinear gray wolf optimization algorithm based on the cubic chaotic map and reverse learning mechanism is designed to balance its global search ability and local search ability, automatically determine the optimal parameters of ISPCNN, and then a segmentation algorithm combining the improved ISPCNN with MLGWO is constructed for fundus retinal vessel. Finally, the test in the public data set DRIVE shows that the sensitivity, accuracy, and specificity of the algorithm are 0.8071, 0.9695, and 0.9862, respectively, compared with other algorithms, the simulation experiments show that the MLGWO-ISPCNN algorithm solves the influence of low illumination, optic disc highlight, and foveal shadow on vessel segmentation, and shows excellent vessel connectivity and peripheral sensitivity.

This paper introduces a single-image super-resolution method based on improved convolutional neural networks. This method aims to improve the spatial resolution of images by designing new network architectures, employing multi-scale loss functions and data enhancement techniques. Experimental results show that the proposed method achieves significantly better results than conventional methods and state-of-the-art SRCNN-based methods on multiple super-resolution datasets, which can better recover image details and generate clear and realistic high-resolution images. The study in this paper is essential for single-image super-resolution. It provides a reference for future improvements in network architectures and training strategies and explores the application potential of other deep learning techniques.

Acoustic scene classification is the task of assigning scene labels to audio based on the recorded environment. Although deep learning models perform well in the field, they usually rely on hardware platforms with high computing power, which limits the popularity of deep learning methods in practical applications. To address this issue, we propose a lightweight High-Resolution network (Lite-HRNet) with only 76.224K training parameters. The structure is based on Lite-Net, a lightweight model constructed using inverted residual modules. Based on the Lite-Net, we introduce the High-Resolution (HR) structure to maintain high-resolution in the frequency axis direction, which effectively fuses high-resolution and low-resolution features in parallel, maintaining low complexity. In addition, the coordinate attention mechanism (CA) is introduced to direct the network's focus towards critical information. Experimental results show that Lite-HRNet improves the classification accuracy of various scenes in the TAU Urban Acoustic Scenes 2022 Mobile Development dataset, achieving an average accuracy of 52.4%, which is a 9.5% improvement compared to the DCASE baseline system.

The need for real-time processing of multi-frame images with complex geometric and optical corrections makes the algorithm more computationally intensive and requires higher processor performance. In order to ensure the real-time and resource consumption problems of the multi-motion target monocular vision multi-level detection algorithm in practical applications, a multi-motion target monocular vision multi-level detection algorithm for ARM isomorphic processors is proposed. The algorithm groups multiple cores in an isomorphic multicore processor evenly, so that the processor cores in the same group are a processor cluster set. The tasks to be scheduled are arranged in ascending order of utilization, and the tasks are assigned to each processor cluster set based on the worst-case adaptation algorithm, and it is guaranteed that the tasks with greater utilization than in each cluster set do not exceed the number of cores in the cluster set. In addition, the moving targets are separated from the background using methods such as background subtraction, inter-frame differencing, or optical flow. The detected targets are matched with their positions in the preceding and following frames to track the motion trajectory. Detection is performed at different resolution levels to handle targets of different sizes and velocities. The experimental results show that the algorithm can utilize the computing resources of the processor more efficiently, achieve faster target detection and tracking, and meet the needs of real-time applications.

Compared to conventional methods of collaborative filtering for recommendations, algorithms that employ matrix factorization are adept at tackling the challenge of sparse data and enhancing the performance of recommendation processes. However, matrix factorization algorithms still have the drawback of not fully utilizing user feature information. To address this, we propose the TransH-MF model. This approach incorporates graph representation learning techniques grounded in matrix factorization. Initially, the TransH graph representation learning algorithm is utilized to project both entities and their relationships into a space, whose vector dimension is lower. Subsequently, the derived vectors are employed to assess the semantic proximity between entities. Ultimately, this proximity data is merged into the process of solving the optimal matrix factorization result. Experiments on the more recent public movielens-latest dataset have proven that the TransH-MF model outperforms the single matrix factorization in regard to both MAE and RMSE metrics.

Recently, with the rapid growth of the Internet, a considerable amount of textual data has emerged, injecting vitality into natural language processing. As a result of the diverse characteristics of natural language descriptions, humans and computers often face ambiguity problems when processing natural language text, and they are unable to combine context and prior knowledge to deeply understand the deep meaning of the text, which limits further intelligent processing of textual data. Entity linking, as a basic task, aims to link the entities mentioned in the textual data to the corresponding entities in the knowledge base. However, short text entity links face difficulties such as limited contextual information, irregular expressions, and incomplete grammatical structures. Furthermore, the flexibility and syntactic diversity of the Chinese language increase the challenge of understanding short Chinese texts on a deeper level. This paper puts forward a Chinese short text entity linking approach via data enhancement and multi-dimensional feature fusion. Experimental results demonstrate that our approach overtakes the baseline model on the Chinese short text entity linking dataset, indicating that our proposed model exhibits better competence and adaptability in handling Chinese short text entity linking tasks.

Synthetic Aperture Radar (SAR) imagery finds extensive applications in both military and civilian domains due to its inherent advantages, such as all-weather capability, high resolution, and complete coverage. However, SAR images encounter several limitations, including unclear edge profile information, multi-scale representation, high sparsity, and a high percentage of small target ships. Consequently, these factors contribute to relatively low accuracy, poor model positioning capabilities, and difficulty in feature extraction in target detection. To overcome this limitation, the present study introduces a novel SAR ship detection method, FL-YOLOV8. It enhances the 160X160 detection feature map by incorporating FHP(four-head prediction) to identify targets larger than 4X4 and replaces the original detection head with LSCDH(lightweight shared convolutional detection head). First, due to the relatively large downsampling multiples in yolov8, it becomes challenging to capture the feature information of small targets. By incorporating a feature head, it becomes feasible to integrate shallow and deep feature information. Second, LSCDH enhances the feature representation of the model, accommodates inputs at various scales, and minimizes both the number of parameters and computational effort. Furthermore, comprehensive experiments conducted on the benchmark dataset HRSID demonstrate the superior performance of FL-YOLOV8 in ship detection, achieving an accuracy of 92.8%.

This study investigates an approach to dialogue act classification leveraging a pre-trained model, with a specific focus on the efficacy of employing the ERNIE model for this task. Dialogue act classification is crucial for deciphering the intentions, actions, and objectives underlying conversations. In this research endeavor, we selected the ERNIE model as our pre-trained backbone, augmented it with fine-tuning techniques, and synergistically incorporated it with an RCNN architecture to achieve precise classification of dialogue acts. Through a series of experiments, we rigorously assessed the model's performance using both publicly available and proprietary datasets, comparing it with conventional methodologies and alternative deep learning frameworks. Our findings revealed that the proposed dialogue act classification methodology, anchored in the ERNIE model and RCNN integration, yielded notable improvements in accuracy and generalization capabilities. This underscores the prowess of the ERNIE model in dialogue act classification tasks, offering new insights and methodologies for analyzing dialogue text. Subsequent research avenues will delve into exploring more intricate model architectures and harnessing richer data reservoirs to further elevate the performance and applicability spectrum of dialogue act classification.

The frequent occurrence of falls among the elderly has attracted the attention of numerous research institutions and scholars at home and abroad as an important obstacle on the path to a healthy global aging population. Injuries associated with falls pose a great challenge to the elderly in maintaining the normal function of the organism, which seriously threatens the safety of the elderly's lives. Therefore, real-time fall detection based on deep learning is crucial for the health monitoring and rescue of the elderly. To address the problems of large model size, poor timeliness, and inability to accurately recognize multi-target poses in existing pose recognition methods, a lightweight and improved OpenPose real-time fall detection algorithm is proposed to replace the VGG-19 feature extraction network with a lightweight MobileNet network. The experimental results show that the proposed algorithm effectively reduces the model size and computational volume, and has good real-time performance and robustness, which meets the application requirements of real-time multi-person pose detection.

Medical image segmentation refers to segmenting pathological regions in medical images. It is an essential step in medical image analysis and an important technical prerequisite for doctors to determine the quantitative extent of pathological changes and treatment plans.In the present paper, we introduce a medical image segmentation method utilizing URV-Net, aiming to address the challenges of ambiguous boundary delineation, limited feature information extraction, and insufficient segmentation precision encountered in current medical image segmentation techniques.The model of this method consists of three major feature extraction modules, namely, U-Net+SE enhanced encoder, multihead sensory field feature extraction module (RFB) with feature residual optimization module (RRM), and ViT encoder with non-local self-attention module (Non-local). Using multipath input, different levels of features extracted from different coding blocks are complemented and optimized with each other by the feature optimization module, which improves the model’s ability to capture global dependencies and capture image features from coarse to fine, simultaneously, it enhances the adaptive learning ability of the importance of feature channels. By applying our proposed model, we conducted the training, validation, and testing phases of the model on the GlaS 2015 dataset, and performed comparative experiments with several other deep learning models. Analyzing the experimental result data, the intersection and merger ratio IoU increased by about 10% on average, the Dice coefficient increased by about 5% on average, the accuracy (Acc) increased by about 6.9% on average, and the precision (Prec) increased by about 5.4% on average, and by comparing the visual segmentation results, the performance of the segmentation model proposed in this chapter is superior to other comparative models in all aspects. Furthermore, ablation experiments have also demonstrated the improvements brought by the combination of different modules in the model to the overall model performance. The code for the model proposed in this article is available at https://github.com/manreren/URV-Net.

In the diagnosis of clinical cardiovascular diseases, the calculation of left ventricular (LV) ejection fraction (EF) is a key step in the assessment of cardiac function. The accuracy of EF depends on the precise segmentation of the LV region. Traditional methods require clinicians to manually outline the LV boundaries in 2D ultrasound images, which is inefficient and subjective. To solve this problem, we propose a deep learning method for automatic segmentation of LV in echocardiography. The method fuses a time series feature enhancement module with a 3D convolutional structure, aiming to achieve a balance between the accuracy and computational efficiency. Further, based on the segmentation results, we also propose an algorithm to automatically calculate the left ventricular ejection fraction (LVEF). Experimental results on the EchoNet-Dynamic dataset show that our segmentation method works well on both 2D images and 3D videos. The average absolute error (MAE) of the computed EF is 0.08, which is lower than the range of normal physiological variability, confirming its potential value in clinical applications.

With the advancement of the digitalisation wave, the importance of 3D point cloud data is becoming more and more prominent. In response to the current lack of 3D point cloud text line datasets, this paper constructs the DocPointCloud dataset, which is generated by 3D reconstruction of document images and provides rich data resources for 3D text line detection. At the same time, for the problem that existing approaches cannot accurately detect 3D text lines when processing document point clouds, this paper proposes the DocPointNet model. By introducing the feature pre-extraction module and the attention mechanism, this model significantly enhances the feature representation ability of the model on point cloud data, thus achieving the accurate detection of text lines in point cloud. Experimental results show that the mIoU of the model on the DocPointCloud dataset is 0.72, which verifies its effectiveness in the 3D point cloud text line detection task.

Movie reviews can help viewers understand basic information about a movie in advance, and film ratings play a vital role in the audience's choice of which movie to watch. Due to the large number of comments left by viewers on movie ticketing websites, it is difficult for viewers to distinguish between good and bad movie reviews and cannot judge the quality of the movie. Therefore, it is necessary to perform sentiment analysis on movie reviews. This paper uses 104,215 movie reviews collected from web pages as experimental subjects, uses TF-IDF to extract text features, and performs sentiment analysis on each movie review through logistic regression algorithm and naive Bayes algorithm to determine the emotional state represented in it. Positive or negative. The results of the two algorithms were compared based on 5 evaluation metrics (accuracy, precision, recall, F1 measure, and area under the curve (AUC)). Experimental results show that logistic regression is better at predicting movie reviews than naive Bayes.

Current optimization methods of learned spatial indexes are often based on data distribution. However, when in coldstart scenarios whose query distributions are unknown, these methods are not suitable. In this paper, a sophisticated learned spatial index, K-Index, is designed for such cold-start environments. It employs a clustering-based data partitioning strategy, utilizing the K-means algorithm to cluster the spatial dataset. Then, cube-shaped partitions are formed with these clusters. For the sake that data should be stored in a certain order to speed up query performance, a dimension reduction technique, inspired by the Lebesgue measure, is used to convert a multi-dimensional spatial sample into one-dimensional format. Finally, a regression model is trained on these dimension-reduced samples for each partition, to enhance performances of cold-start spatial queries. The effectiveness of K-Index has been validated through experiments on both real and synthetic datasets, representing a significant advancement in learned indexing technologies. The K-Index method has been proven effective through experiments on both real and synthetic datasets, enhancing the adaptability and efficiency of handling spatial data and marking a significant advancement in learned indexing technologies.

Birds play an important role in the ecosystem. Using machine automatic recognition of bird species can be used to monitor bird activities, help researchers understand bird behavior, and have important significance for the protection of wild resources and scientific research of bird species. This paper designs a method for automatic recognition of bird species based on deep learning. We extract image features based on pre-trained EfficientNet-B0 and use MLP for classification. Our method achieves an accuracy of 87.3% on the validation set, which exceeds human recognition accuracy for complex bird species.

Images acquired under foggy conditions have blurred lighting and low resolution and contrast, while the feature extraction of traditional target detection designs does not take into account the fog obscuring the target, affecting the recognition rate of targets such as vehicles and pedestrians. Based on this, this paper proposes YOLO-FOG. Based on Anchor-Free, the TSCODE module is inserted to further separate the feature encoding of the two tasks; MobileViT is introduced into the pyramid structure of the feature graph to construct the long-distance dependencies of the features and integrate the features. When evaluated using the RTTS dataset, the mAP values of Bicycle, Bus, Car, Motorbike, and Person with their five types of targets can reach 82.19%, 81.36%, 89.47%, 75.89%, and 84.01%, respectively, and meanwhile the simultaneous detection speed up to 65.52 fps, which proves that this paper's method combines both accuracy and speed, and has good application prospects.

Current representation learning-based multivariate time series classification models typically only consider temporal features and fail to model hidden relationships between different variables. Therefore, this paper proposes a multivariate time series classification model based on dynamic-static graph ranking contrastive learning. The model represents the features of different dimensional variables in multivariate time series as nodes and the latent relationships between variables as edges. While modeling the hidden relationships between multivariate variables, it captures the dynamic variation characteristics of variables through dynamic graph representation. Additionally, it utilizes discrete wavelet transform and its inverse transform to obtain frequency domain features of time series and enhances the data to generate strong and weak views. Meanwhile, by employing node-graph contrastive loss and graph ranking contrastive loss, the model learns robust and discriminative graph representations. This study compares the proposed method with current mainstream multivariate time series classification methods as baselines through comparative experiments on 30 publicly available datasets from UEA. Experimental results demonstrate that the proposed classification approach significantly outperforms existing methods.

The safety of chemical equipment is influenced by multiple parameters, and there are complex interactions and correlations between these parameters. In order to improve the quality of chemical equipment safety level assessment, a chemical equipment safety level assessment method based on information entropy and time series is proposed. Firstly, based on the information entropy method, useful evaluation data is extracted from a large amount of safety level evaluation data of chemical equipment. Then, based on the time series method, evaluate the safety level of chemical equipment. Finally, comparative experiments were designed to verify the effectiveness of this method. The results indicate that the information entropy and time series methods have high accuracy and efficiency.

This paper present a novel blockchain CP-ABE (Ciphertext-Policy Attribute-Based Encryption) data ciphertext sharing scheme, leveraging decentralized proxy re-encryption. It utilizes Ethereum blockchain and zero-knowledge proof technologies to develop a blockchain key management system for multiparty secure computation. This system aims to mitigate heavy computation and key leakage risks associated with traditional centralized CP-ABE systems. The approach begins with the design of a distributed master key generation and distribution protocol within the data ciphertext access control, deploying the UMBRAL-based proxy re-encryption technology on blockchain for key management system. Moreover, the paper uses ERC1155 to mint transferable NFTs for blockchain transactions in data sharing and access control. Through simulation experiments, our scheme demonstrates improved scalability in blockchain ciphertext data transactions and storage, alongside a significant decrease in the Gas costs of on-chain data asset transactions. Furthermore, our methodology not only augments the safety of distributed key management but also realizes high computational efficiency and swift query response, particularly in scenarios of high concurrency.

In response to the issue of low accuracy in predicting the navigation state of existing AUV motion models, a predictive model for AUV navigation state based on a combination of a time-domain convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM) recurrent neural network enhanced by a multi-head attention mechanism is proposed. Initially, the CNN extracts the feature sequence of the AUV navigation state. Subsequently, the BiLSTM network is utilized to learn the trajectory sequences of preceding and following conditions to extract dependencies within each navigation state in the sequence. The multi-head attention mechanism is then introduced to adjust the weights of different attribute features, prioritizing those with a greater impact on the navigation state. This ultimately facilitates the accurate prediction of the AUV navigation state. A comparative experiment using AUV lake test data demonstrated that the CNN-MABiLSTM model achieves higher prediction accuracy and better fitting than the CNN-BiLSTM model, thus verifying the effectiveness and practicality of the designed CNN-MAiLSTM model in predicting AUV navigation states.

The tobacco industry plays a crucial role in our country's financial stability and economic development. In the production assembly line of tobacco silk, tobacco drying is a vital process, directly impacting the quality and flavor of tobacco[1]. Tobacco drying involves complex simultaneous heat and mass transfer, and physicochemical transformations. To enhance control strategies for silk drying and study parameter changes in the drying process, this paper focuses on the parameter characteristics of the silk dryer. Analyzing from the perspective of heat energy balance and moisture changes, this paper constructs a model accurately describing the characteristics of the electromagnetic vortex-type silk dryer[2]. The aim is to predict the moisture content of exported tobacco and provide deeper theoretical support for the technical development of the tobacco industry[3].

The traditional approach for text classification tasks commonly involves fine-tuning existing general-purpose or base models. However, when training data is extremely scarce, this approach may lead to overfitting or getting stuck in local optima. This paper investigates meta-learning-based methods for low-resource text classification, taking a diabetes dataset as an example, and proposes a meta-learning framework combined with data augmentation techniques. Firstly, we utilize an abstract summarization method to effectively augment the original dataset, alleviating potential issues caused by data imbalance while enhancing the diversity and generalization of the samples. Then, by employing a meta-learning algorithm, we achieve optimization and adjustment of the global initialization parameters. Subsequently, these parameters will guide the fine-tuning of the pre-trained BERT model to adapt to the diabetes text classification task. Experimental results show that our method significantly improves classification performance under low-resource conditions, providing new insights for handling low-resource text classification problems in similar domains.

In response to the current situation of low voice wake-up rate, low recognition rate, false wake-up, false recognition, and recognition errors in car voice interaction, which cannot be detected after execution, manual error correction verification is required. However, manual testing faces a large number of test cases and is prone to negligence. Based on the visual recognition technology of artificial intelligence, this article can automate testing and calculate the car voice wake-up rate, recognition rate, false wake-up, false recognition, and word error rate. Thus, problems can be identified and corrected in a timely manner, improving work efficiency and accuracy, thereby improving the accuracy of in vehicle speech recognition and the recognition rate of speech execution.

Aiming at the problems of lack of emotional resonance and unsatisfactory anthropomorphic effect in the responses generated by the existing dialog models, a dialog generation model based on the prediction of emotional intent is proposed. The model predicts the intention information of the context to obtain fine-grained emotion information to construct a high-quality dialog generation system, combines with an emotion classifier to identify the emotion category of the input utterance, and generates corresponding empathetic replies based on the emotion comparison mechanism according to the different intention information and emotion information. The experimental results show that compared with the traditional dialog generation model, the proposed model can recognize the user's emotional intent and emotion information, generate responses that are more in line with the user's expectations and adapt to the context, and realize a more empathetic emotional dialog process.

The integration of blockchain technology with Ciphertext-Policy Attribute-Based Encryption (CP-ABE) has addressed key challenges in distributed systems, such as mutual trust and collusion, while enhancing auditability of key management. However, existing blockchain-based CP-ABE models face a trilemma, unable to achieve decentralization, scalability, and security concurrently. Therefore, we propose ZK-CPABE, which merges Zero-Knowledge Proofs with blockchain-based CP-ABE, significantly improving verifiability and scalability. This system combines a secure private key distribution and zero-knowledge proofs to construct off-chain computation and scalable transactions for CP-ABE. Our Ethereum-based ZK-CPABE prototype demonstrates its effectiveness and potential for secure, scalable data sharing and access control.