This PDF file contains the front matter associated with SPIE Proceedings Volume 13441, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In the digital era, the transformation of the book industry has promoted the rapid development of book e-commerce platforms, emphasizing the importance of accurate classification of book categories. This paper proposes a deep learning model BookGCN that combines text attributes and book link relationships, aiming to improve the accuracy of book classification. By using Bert for text feature extraction and Graph Convolutional Network (GCN) to extract the link relationship between books, the model effectively integrates these two types of information. Experimental results show that the SAGE graph convolution model performs best in different graph convolution networks, reaching an accuracy of 0.602. At the same time, the ablation experiment verified that the link relationship and residual connection significantly improved the performance of the model. Although there is a long-tail problem in the distribution of categories, this paper not only provides new ideas for book classification, but also lays a foundation for the development of personalized recommendation system, and injects new vitality into the sustainable development of the book industry.

With the rapid advancement of drone technology, unmanned aerial vehicles (UAVs) have found widespread application in military domains. Ensuring the optimal performance of UAVs in military applications necessitates the assessment of their intelligent capabilities. Traditional live-flight test methods for UAVs pose inherent challenges such as high risk, substantial costs, and low efficiency. Consequently, research into intelligent level test methodologies based on simulated scene holds significant importance. To address the modeling requirements for UAV intelligent level test scene, this paper proposes a modeling approach founded on the Meta Object Facility (MOF) framework for constructing scene model tailored to evaluating UAV intelligence. Concurrently, this paper establishes a complexity model for test scene based on the Analytical Hierarchy Process (AHP), and employs this model to perform a graded assessment of the UAV's intelligence level according to the complexity of the simulated scene.

Intelligent workshops have become one of the hotspots in current manufacturing industry. Production logistics simulation technology in intelligent workshops can proactively identify potential issues. Entities within the workshop, such as production and logistics entities, comprise the workshop, utilizing the MBSE method and SysML modeling language for modeling. The structure and behavior of entities within the workshop are modeled based on BDD and STM. The workshop's structure and processes are modeled using BDD, IBD and SD. Instances in BDD are used to instantiate workshop components. Through template-based code generation, the SysML-based intelligent workshop production logistics model is automatically transformed into executable code, which is then simulated and run. The results demonstrate the feasibility of the SysML-based unified approach: Compared with the commercial software FlexSim, the outcomes are favorable.

The volume of data generated by the manufacturing industry has exploded at an exponential rate in structured, semi-structured and unstructured forms, increasing the risk of errors in the multi-modal data fusion process. To address this challenge, this study proposes an innovative multimodal information fusion strategy aimed at improving data processing capabilities in manufacturing digital industrial scenarios by integrating federated learning optimization algorithms. The multi-modal data in the digital industrial scene of manufacturing industry is collected and redundant components are eliminated. The feature information is extracted from the dimensionally reduced data and input into a multi-modal data fusion framework optimized by federated learning to generate a tensor integrating three key features. The fuzzy set theory is used to guide the local decision process of each module in the framework, and the fuzzy membership degree corresponding to the multi-modal information is calculated to realize the multi-modal information fusion effectively. The experimental results show that the information fusion degree of this method is not less than 0.35, and the fused information has high integrity and consistency. In most cases, the fusion error can be significantly reduced.

Spread spectrum and channel coding are two main anti-interference methods in wireless communication. In order to enhance the anti-interference ability of communication systems, this paper proposes a joint coding and modulation technology combining RS channel coding and CCSK. We have conducted research on the combination of RS encoding and CCSK modulation, with a focus on the research and Simulink programming implementation of CCSK demodulation algorithm based on FFT operation. The simulation results show that under the same channel interference conditions, the joint coding and modulation technology of RS and CCSK can effectively improve the system's anti-interference ability.

Sparse code book multiple access (SCMA), due to its good link performance, adapt to large-scale access scenarios, helps to cope with the surge of 5G and next generation mobile communication equipment and the challenge of access capacity improvement. The implementation of SCMA system faces two major challenges: optimal codebook design and efficient decoding algorithm. At present, the method based on deep learning has been studied in SCMA system. In this paper, a SCMA system scheme based on generative network introduces Transformer attention mechanism in the encoder and uses context information to solve the problems of high complexity and insufficient flexibility. In the decoder, Patch GAN is adopted to reduce the number of parameters and calculation of network model, and solve the problem of high complexity of traditional decoding algorithm to improve the performance of code error. The research of SCMA system based on generated adversarial network has important theoretical significance and practical value for the selection of multiple access scheme of 5G and future mobile communication system.

For high resolution spaceborne spotlight/sliding spotlight SAR, it is essential to consider the error of the range equation and signal azimuth spectrum aliasing, both of which are caused by the long illuminate time. In order to address these issues, we present an imaging algorithm for high resolution spaceborne spotlight/sliding spotlight SAR. Firstly, it utilizes "curve orbit compensation" to compensate for range space variation error prior to RCMC, thereby avoiding defocusing issues resulting from the error of the range equation. Secondly, through the use of the "extended TSA", variations in Doppler center frequency and Doppler frequency rate with instantaneous range frequency are employed for azimuth pre-processing, effectively overcoming problems associated with azimuth spectrum aliasing. Simulation results for point targets demonstrate the effectiveness of the proposed algorithm.

Continuous monitoring of Hematocrit (HCT) is essential for the treatment of patients during cardiopulmonary bypass. Noninvasive monitoring is helpful to prevent the formation of thrombosis and reduce the risk of treatment. However, the problem of insufficient sampling data further makes it difficult to improve the accuracy of existing prediction methods. Aiming at the problem of sampling difficulty, a data simulation model was proposed. The data simulation model is based on Monte Carlo (MC) photon simulation algorithm. Then, the light intensity data after passing through the blood obtained. Aiming at the problem of low accuracy of existing prediction models, two prediction models are further proposed to estimate HCT: backward iterative interpolation estimation model and BP neural network prediction model, and then the estimated values were compared with the actual values. The experimental results show that the MC data simulation model can reflect the real data change trend. When the simulated data are applied to the reverse interpolation estimation model and the BP neural network prediction model, the root mean square error of the former is 1.39%, and the latter is 0.61%. Bland-Altman analysis was also performed to confirm the accuracy.

The health of bearings in mechanical equipment is crucial for production safety and stable operation. The vibration signal of a bearing contains abundant fault information, accurately reflecting its health condition. In this paper, the feature extraction algorithm for vibration signals is being studied and optimized for various bearing fault characteristics. In the time domain, statistical methods are applied to extract the peak value, root-mean-square (RMS), and other indicators reflecting the signal amplitude and energy. In the frequency domain analysis, Fast Fourier Transform (FFT) is used to find the characteristic frequency, and the operating quality of the bearing is determined through harmonic analysis^[1]. In the time-frequency domain analysis, Wavelet Transform (WT) and Wigner-Ville distribution are utilized to extract the features of the non-stationary signals^[2]. Combining these methods optimizes the feature selection logic further. Dimensionality reduction techniques, such as Principal Component Analysis (PCA), are utilized along with adaptive filtering and other methods for reducing noise in the original signal^[3]. For the extracted features, a support vector machine (SVM)-based classifier design and optimization are conducted^[4]. Simulation experiments are carried out using MATLAB to validate the effectiveness of the proposed algorithm for identifying fault types such as outer ring, inner ring, rolling body, and cage. The experimental results show that the algorithm exhibits high precision and recall in fault diagnosis, especially in complex noisy environments, while still maintaining good performance. This performance is superior to that of traditional methods. The optimization process and analysis method proposed in this study significantly enhance the accuracy of bearing fault diagnosis, offer reliable technical support for condition monitoring and maintenance of mechanical equipment, and hold significant application value for intelligent manufacturing and predictive maintenance.

Direct sequence spread spectrum (DSSS) signals, due to their complex signal structure, pose challenges for accurate modulation mode recognition and parameter estimation using traditional methods. This paper proposes a deep learning-based DSSS signal mode recognition and parameter estimation algorithm. It employs cyclic spectrum analysis to estimate the carrier frequency and code rate parameters, and combines the feature extraction capabilities of a convolutional neural network (CNN) with the modeling and recognition strengths of a long short-term memory (LSTM) network to achieve high-precision DSSS signal recognition. Experimental results demonstrate that the cyclic spectrum method can accurately estimate the carrier frequency and code rate parameters, and the CNN+LSTM mode recognition achieves an accuracy rate of over 85% even at a signal-to-noise ratio below 0 dB, validating the effectiveness of the proposed algorithm for DSSS signal mode recognition.

In response to the challenge presented by the wide range of scenarios and targets in intricate disinfection environments, this study introduces an improved target detection algorithm. The algorithm integrates a Coordinate Attention mechanism, thereby enhancing the ability to capture information related to small targets. Additionally, the SPPFCSPC module effectively enhances the feature extraction capabilities of the backbone feature network. To address issues related to slow convergence and imprecise regression outcomes, the Focal and Efficient IOU loss function is utilized. Experimental findings demonstrate that the proposed algorithm achieves a remarkable mean Average Precision of 91.4%, representing a notable enhancement of 2.4% over the YOLOv8n model.

3D laser scanning improves the convenience of point cloud data acquisition. However, owing to the impact on multiple environmental factors during the scanning process, The targeted point cloud data contains numerous noise points, which will directly effect the subsequent segmentation and three-dimensional reconstruction of the point cloud data, and thus need to be denoised. Usual denoising ways cover the average filtering method, straight pass filtering method, and so on. In this paper, by studying the Spatial position between the point cloud and its domain points in space, A denoising algorithm for point clouds is proposed based on the spatial distribution of domain points, which can effectively remove the noise while retaining the point cloud body well.

Crowd counting is a computer vision task that involves using computer algorithms to analyze the number and distribution of crowds. With the world's population increasing, stampedes caused by too many people continue to occur frequently, and the number of researchers paying attention to the task of crowd counting has been growing steadily. This article first introduces about the traditional crowd counting algorithm and the deep learning-driven crowd counting algorithm. It introduces two performance indicators used to assess the crowd counting algorithm, the mean absolute error (MAE, which can be represented by E_MAE) and the root mean square error (RMSE, which can be represented by E_RMSE). Finally, discussed some public datasets about crowd counting work .These datasets make much contributions to the development of algorithms, and comparing the experimental results of different algorithms can help improve crowd counting algorithms.

With the deepening application of Marketing 2.0 and Smart Brain, the problem of not being able to find, not being able to read, and not being able to use the data has been the difficulty and pain point of plaguing the business people to carry out data analysis independently. Therefore, this paper proposes a distributed encrypted storage of large-scale atomic data for cloud computing. In-depth study of the smallest computing factor with free combination and assembling ability, i.e., atomic data management application system, to realize efficient and safe storage of large-scale data in the cloud computing environment. At the same time, the study of atomic data business application system is directed by business requirements, aiming at consolidating the basic analysis base of marketing intelligent brain, empowering grass-roots business, lowering the threshold of data analysis at the grass-roots level, improving the ability of business risk identification, creating a good atmosphere of independent analysis, helping to cultivate the grass-roots digital team, and promoting the digital transformation of marketing. The research results show that the distributed encrypted storage method for large-scale atomic data oriented to cloud computing can effectively respond to the challenges of huge data volume, high access frequency, and high security requirements in the cloud computing environment.

To effectively improve the performance of multiple earth observation satellites and obtain more observation benefits, it is very important to plan tasks according to the characteristics of observation targets in advance. The problem of satellite collaborative mission planning is studied in this paper, and a multi-satellite mission planning problem model is proposed. A novel genetic algorithm is proposed to obtain the observation mission sequence of each satellite. The detail design and process of the algorithm are introduced. Simulation results show that our algorithm is a good approximation algorithm for solving the NP-hard planning problem, and can enhance the rapid response capability of satellites, and improve the efficiency of satellite observation missions.

Convolutional neural networks have become a key research object in computer vision related fields due to their powerful feature extraction and classification recognition capabilities. However, as CNN models continue to update and iterate, their structures become increasingly complex and the number of model parameters also increases. However, general CPUs cannot meet the large and intensive computing needs, and using GPUs has a series of problems such as high power consumption and noise. The cost of designing ASICs is too high, and they do not have the ability to update and iterate. With the rapid development of FPGA, the resources integrated on FPGA chips are constantly enriched, and some FPGA chips also integrate high-performance digital signal processing units. These advantages make FPGA an important platform for accelerating convolutional neural networks. This article focuses on the computational requirements of parameter intensive shallow CNN network models. By fully exploring their parallelism, structures such as convolutional layers with multiple convolution kernels in parallel, pooling layer pipeline and convolutional layer synchronization, and fully connected layer hierarchical computation are designed. Finally, it was deployed on the XCZU7EV-2FVC1156I chip of the Zynq UltraScale+MPSOCs EV series, and simulated using Vivado. The simulation data was validated against the results of external calculations to determine the logical and data correctness of the design. At the same time, compared with an accelerator design based on LeNet-5, this design has a shorter inference time for a single image at 86.68us and a computational performance of 19.96GOP/s under higher computational requirements, fully verifying the performance advantages of this design.

In response to the satellite attitude (SA) stability problem faced by the RMMFFSR in capturing NCT, firstly, the forward kinematics (FK) equation of RMAFFSM based on RMMGJM was established. Secondly, RMAFFSM decompose motion speed control (DMSC) method based on RMMGJM was proposed, and DMSC algorithms for RMAFFSM multi arm coordinated motion(CM) to stabilize SA and RMAFFSM multi joint CM to stabilize SA were designed. Finally, an RMAFFSM SA control system (MSACS) and a SA controlled DMSC algorithm were designed. The RMAFFSM multi arm planning system was combined with the MSACS to propose an RMAFFSM SA motion control (MC) algorithm based on attitude Interference Prediction (IP) and an RMAFFSM autonomous coordinated MC algorithm for stabilizing SA.

Incorrect motion state detection is a significant obstacle of motion monitoring systems in application of the real world. To address this issue, this paper proposes a multi-source feature fusion-based intelligent motion monitoring (IMM) system. First, an edge-end (EE) motion monitoring module is designed to recognize physical movements. The EE module adopts MoveNet model to detect pose keypoints in video frames, then the extracted keypoint coordinate information is classified to the EE motion features. Second, a Kalman fusion filtering (KFF) algorithm is designed for the to extract device-end (DE) module features. The KFF algorithm combines acceleration and angular velocity features to acquire the accurate inertial features of the device-end module, which effectively mitigating external interference caused by device movement. Based on these, the IMM system is built based on the wireless communication between the EE module and the DE module. Then, the IMM system integrates multi-source features from computer vision and inertial sensors, realizing the real-time monitoring of physical movements. The experimental results validate the IMM system for real-time movement monitoring in real-world scenarios.

A novel wideband CPW-fed elliptical slot antenna is proposed for ultra-wideband (UWB) applications. The design of the antenna includes a wide elliptical slot cut in the ground plane and an L-shaped radiator fed via a coplanar waveguide (CPW) line. By introducing a wide elliptical slot, the antenna achieves a broad impedance bandwidth. Additionally, the circularly polarized (CP) bandwidth is significantly enhanced by the introduction of a L-shaped radiator. For this antenna, the simulated impedance bandwidth reaches 99.8% (4.01GHz – 12GHz) and the simulated CP bandwidth is 91.9% (3.19GHz – 8.31GHz). The antenna, measuring 42mm × 42mm × 1mm, achieves a peak gain of 5.1dBic on the overlapping bandwidth (|S11| < -10dB, AR < 3dB).

This study focuses on the coherent integration (CI) of space-based distributed radar for space high-speed maneuvering targets. The platform space difference and high-speed motion result in the envelope position offset and phase difference of the inter-channel signals. Meanwhile, the high-speed and maneuvering characteristics of the target will lead to the range walk (RW) and doppler walk (DW) in the pulse accumulation in single node, which brings challenges to the detection of space targets. This paper presents a method for spatial-temporal joint coherent integration and parameter estimation based on the time-space range history. Firstly, based on the coupling relationship of spatial and temporal in the range history, we established inter-channel envelope correction and phase compensation function to address the inter-channel echo range and phase difference caused by spatial location differences between nodes. Secondly, we combined inter-channel synthetic compensation with generalized Radon-Fourier transform (GRFT). Thus, target range, angle, velocity, and acceleration parameter estimation are integrated into a unified framework. Finally, joint Spatial-temporal joint CI and parameter estimation for space-borne distributed radar system can be implemented. Simulation results demonstrate the effectiveness of the proposed method.

The rapid development of Internet of Things (IoT) technology has made the interconnection of all things a reality. However, multi-source data fusion still faces challenges such as the diversity of different device interfaces, the non-unification of device clocks, and the high integration of video acquisition devices. The data synchronous acquisition of serial communication interface and image video is of great significance due to its wide application. Based on the digital signal detection of the serial communication data interface in this study, and in combination with the application of the OpenCV image processing control, the data synchronous acquisition of the serial communication interface and the USB camera is realized. The system is experimentally verified by using the radar velocimeter. The measured data synchronization accuracy is less than 1ms, providing a feasible technical means for multi-scenario applications.

The transmitter of GEO-LEO system spaceborne bistatic radar located in geosynchronous orbit, and the receiver located in low orbit, which has the advantages of large beam coverage, anti-stealth and anti-interference, and has great application potential. However, the feature of large beam coverage requires the detection efficiency. For improving the detection efficiency, the best beam coverage area should be designed. In this paper, the detection area is divided by transmitting beam and receiving beam for simulating the real beam irradiation condition and the clutter suppression ability is used as the evaluation index to find the change rule of detection performance and the best detection position of the detection region. The simulation results show that the detection effect is best when the azimuth angle of the receiving beam is 90°, and the regional detection performance gradually becomes better with the extension of the detection distance.

Common simulation experiment evaluation is mostly static post-evaluation, which takes a long time and cannot meet the requirement of concurrent experiment and evaluation. To solve the above problems, this paper designs a real-time simulation evaluation system based on network technology, which consists of simulation computer, data acquisition and preprocessing module, and real-time evaluation module. The three are connected through a real-time optical fibre reflective memory network. The system can calculate evaluation indicators based on real-time simulation experiment data and display evaluation results in time, supporting the timeliness of simulation experiment and evaluation.

The demand for precise monitoring and analysis of user behavior in power grid terminals has surged dramatically. Traditional monitoring methods struggle to adapt to complex behavior patterns. this paper proposes a method based on constructing behavior portraits using statistical and behavioral features, aimed at uncovering the underlying connections in user behavior to effectively preempt and manage potential risks in the power grid system. Firstly, statistical features are extracted from user behavior sequences using data mining techniques and transformed into statistical feature labels. Then, through neural network training, these sequences are deeply modeled to generate behavioral feature tags. Finally, by combining these two types of tags, a behavioral profile of terminal users is constructed, and graph matching technology is used for terminal situational awareness, effectively identifying and addressing potential security threats. Experimental results validate the effectiveness of this method, which not only precisely captures and analyzes user behavior features but also significantly enhances the security management capabilities of power grid enterprises.

An automatic control system with temperature, humidity and illumination detection is introduced. The system compares the values with the range of temperature, humidity and illumination set by us through buttons. If the temperature, humidity, and illumination in the surrounding environment are not within the range we set, a buzzer will sound, and the relay will activate temperature and humidity control devices such as a small fan for dehumidification and cooling, a water pump for dehumidification, a heating element for heating, or an LED light to increase illumination until the temperature, humidity, and illumination return to the specified range. At the same time, the microcontroller will connect to the Bluetooth module and the phone to send the detected temperature, humidity, and illumination to the phone.

At present, it is critical to protect radar and communication systems from interference. As technology continues to advance, ensuring the stability and security of these systems is becoming more and more important to maintain normal operation and communication quality. In this paper, an anti-jamming technique based on cyclic prediction optimal Kalman filter (CPOKF) based on pseudo-code phase modulation fuse pulse is introduced. In order to improve the anti-jamming ability of the pseudo-random code pulse Doppler fuse, on the basis of the analysis of its working and anti-jamming principle, the Kalman filter parameters were optimized by using the Crested Porcupine Optimizer (CPO), and a CPOKF algorithm was proposed to simulate the pseudo-code pulsed Doppler signal under active suppression interference. The results show that the anti-interference performance index of the filtered signal of the proposed algorithm is better than that of the fixed-parameter Kalman filter signal, and the anti-interference performance is better.

5G is a new wireless communication technology, which is characterized by faster data transmission speed, lower delay and higher bandwidth. It is highly consistent with the needs of power communication. Integrating 5G communication technology with power grid services can effectively improve the intelligence level of the power system. This paper analyzes the application scenarios of 5G in power systems, introduces the power 5G slicing architecture adapted to these scenarios, proposes a terminal design scheme with strong processing capabilities, rich interfaces, and 5G functions, and tests the performance of the terminal. It will play an important role in all aspects of smart grid including power transmission, transformation, distribution, and power consumption.

For change detection in multi-source heterogeneous images, this paper proposes a change detection method using a refined hierarchical clustering approach. For multi-source heterogeneous multi-temporal images, Stacked Denoising Autoencoders (SDAE) is used to extract deep features from multi-source heterogeneous images. On this basis, the comparability of heterogeneous data in deep feature space is guaranteed by iterative transformation of features. Finally, the correlation between deep features of heterogeneous data is described by introducing a variety of distance measures, and the hierarchical clustering method is improved. The classification of change types is gradually realized through multiple clustering, while improving the accuracy of change detection.

In SAR images, the radar antenna pattern has a significant impact on the energy intensity of the target. Therefore, it is essential to correct for this influence during processing in order to eliminate any fluctuation gain changes caused by it. With evolving application requirements, new SAR systems often operate in squint mode, rendering the current antenna pattern correction method developed for broadside looking mode unsuitable. This paper proposes a novel correction method for SAR antenna patterns that effectively enhances the accuracy of pattern correction in squint mode. The effectiveness of the proposed method is validated through a SAR simulation experiment.

At present, many processes of construction in China are still highly dependent on manual work, which has high labor intensity, low efficiency and high risk, and cannot be scaled and industrialized production. The binding point of steel bars is an important part of reinforced concrete structure, which directly affects the continuity, tightness and force transmission performance of steel bars, and thus determines the safety and durability of the structure. For the above problems, this paper designs and implements a method of video reinforcement strapping quality detection based on deep neural network. Shallow features in the backbone network of YOLOV4 algorithm are directly input into the SPP module for feature fusion. Meanwhile, CBAM modules and RFB modules are added to the backbone network and neck network of the target detection algorithm, respectively, to increase the algorithm's sensitivity field and increase the algorithm's attention to target information. The mAP@0.5 of the improved CBAMy-RFB algorithm is 4.4% higher than the mAP@0.5 algorithm of YOLOV4 algorithm. The reasoning speed of the new algorithm is 60FPS, which is 5FPS lower than that of YOLOV4 algorithm. The improved object detection algorithm meets the requirements of the application scenario.

In a distributed network environment, multiple mutually independent trust domains are formed between different enterprises. During network access, access requests for resources may come from local trust domains or from field trust domains. Therefore, when a user from a local trust domain accesses a resource from a foreign trust domain, or a user from a foreign trust domain accesses a resource from a local trust domain, it involves cross-trust domain authentication and session key negotiation. In this paper, we propose a distributed heterogeneous network environment security monitoring study based on microservice architecture, which collects and analyzes the horizontal traffic of microservices in real time to monitor the network performance. By collecting the virtual network packet capture data under the microservice architecture of the SouthNet cloud environment, the network communication quality between microservice applications is monitored and alerted in real time, and the network communication fault localization between microservice nodes is supported to assist in troubleshooting. The experimental results show that the distributed heterogeneous network environment security monitoring system based on microservice architecture can improve the flexibility, scalability and security of the system, which is applicable to the monitoring needs in different fields.

In this paper, a new method of adjusting the resistance value of the memristor is proposed by studying the existing resistance control methods and improving the variable gate voltage control method. The experimental results show that the method can realize the resistance control quickly, and the precision range of the resistance value is about 5%, which can meet the requirements of using the memristor array to complete the neuromorphic calculation. Based on the variable gate voltage control method, the precision control method of the memristor resistance value designed in this paper can realize the precise control of the resistance value by changing the amplitude and width of the pulse voltage applied to the source on a certain basis, and can achieve the effect of rapid convergence. Through experiments, the research results can be applied to the accurate regulation of the resistance value of the memristor, which can solve the problem of slow convergence speed of the current memristor regulation method to a certain extent, and can better achieve the regulation of the resistance value of the 1T1R array through further optimization, so as to better serve the neuromorphic calculation and help the development of artificial intelligence.