This PDF file contains the front matter associated with SPIE Proceedings Volume 12246 including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

Ship noise has always been one of the important sources of marine noise. In order to study the sound source characteristics of underwater radiated noise from ships, this paper is based on the monitoring data collected in the sea area of Zhoushan Zhairuo Island, to get the underwater noise signal of the measured ship under different driving speeds. Through the timefrequency domain feature analysis, the short-term energy spectrum and spectrogram of the noise signal are drawn, and the following conclusions are obtained: The frequency of ship underwater noise is mainly concentrated in the interval of 0~700Hz, with the increase of ship speed, the main frequency range of underwater noise also increases, and the noise energy at different frequencies also increases. The sound pressure level of the underwater noise of the ship selected in this experiment is in the range of 75dB~100dB. With the increase of ship speed, the sound pressure level of ship underwater noise also increases, and has a certain linear relationship. The slope estimate m is 2.442dB/knot.

To the high accuracy real-time landslide early warning information sent to the disaster area, reduce the landslide caused loss of life and property of people, design a kind of based on Beidou short message communication mode to release the method of high-precision landslide early warning information, optimize the landslide early warning information release, the standard of Beidou short message protocol and CAP (Common Alerting Protocol) combined general alarm agreement, enrich and perfect the early warning information content, high precision of landslide early warning information for the package transmission, and automatically by the Beidou terminal receipt Beidou short message warning message, implements the information based on Internet and Beidou short message communication connectivity, provides the way of warning issued emergency measures, improve the timeliness and reliability of the landslide early warning information release.

In order to better solve the problem of land pollution, it is necessary to analyze and judge the complex types of land pollutants. In view of the poor detection effect of land pollutants, the regional pollutants are identified and analyzed in combination with distributed sensing technology, and the regional land characteristics are collected and analyzed in combination with distributed sensing technology to judge the types and degree of pollutants, The results show that the analysis results of soil pollutants by distributed sensing technology are basically accurate, and the evaluation results of soil environmental quality are completely consistent, which can be used for soil environmental quality evaluation and quantitative identification of soil pollutants.

The development of short-range wireless communication technology has gradually matured in China. Its application in medical electrophysiological (ECG) is the key technical point for the measurement industry to enter the large service system of telemedicine industry. It is made a detailed research and comparative analysis on various short-range infinite communication technologies in this paper. We select ZigBee technology with low power consumption, low cost, high reliability, large network capacity and "one to many" data transmission to be applied to the wireless measurement and detection system of ECG signals, and design the overall structure of the wireless measurement and detection system of medical signals. The design principle and software program flow of hardware circuits are also given. Test results show that the ECG waveform distortion, discontinuity and deformation are very small, which could meet the needs of wireless parallel detection of ECG monitors in a single department unit or independent ward building.

The corresponding arrhythmia often occurs before the onset of cardiovascular disease (CVD), electrocardiogram (ECG) can more intuitively detect any abnormality in the heartbeat as a sign of arrhythmia. There are many traditional ECG classification methods, but these methods are constrained human costs and inaccuracy since they rely on manually extracting features, and cannot fully mine the deep pathological information hidden in the data. Consequently, a novel deep learning model based on single-lead ECG signals and inter-patient paradigm is developed to improve the shortcomings of traditional ECG heartbeat classification. The residual connection is adopted in the proposed method to improve classification accuracy and alleviate the gradient disappearance issue. Heartbeat complexes that included a targeted heartbeat and an adjacent heartbeat is selected as the input of the model. The MIT-BIH arrhythmia database is employed to valid proposed method. Besides, a focal loss function is used to address the classes imbalance of the database. The experimental results show that the positive predictive values of the proposed classification method for N, S, V, and F are 99.10%, 96.80%, 61.32%, and 95.36%, respectively. In addition, the sensitivity values are 95.83%, 95.54%, 90.43%, and 84.79%; the specificity values are 92.96%, 99.88%, 96.05%, and 99.97%, respectively. Compared with the art-of-state inter-patient ECG heartbeat classification approaches, our proposed approach achieved better results. Therefore, the proposed deep learning model of heartbeat classification is effective and feasible for the single-lead ECG signals and interpatient paradigm.

At present, there are many kinds of wearable ECG devices in the market, but there are many problems such as large volume, low precision, high power consumption, and discomfort in wearing. The wearable ECG monitoring system is designed, which be small, deformable, comfortable to wear, and can continuously measure ECG signals in real-time. In this paper, STM32F411 is used as the microcontroller. Furthermore, TI's 24-bit ADS1292R is used as the acquisition module of the ECG signals. The measured ECG signals are passed through a 0.5-70Hz band-pass and 50Hz notch digital filter to remove motion noise signals. The processed ECG data are stored in a Micro-SD card. At the same time, the data are sent to the upper computer (Labview) through Bluetooth to display the ECG waveform and heart rate in real-time. The device was used to measure the ECG signals of people of different ages and genders under other exercise states. The results show that the device has the advantages of comfortable wearing and low power consumption. It can display the electrocardiogram and heart rate in real-time. The ECG data transmission delay does not exceed 0.5s, and the accuracy of heart rate measurement is 92.45%.

In order to investigate the feasibility of acoustic emission phenomenon aroused by electromagnetic excitation in metal specimens, electromagnetic acoustic emission model based on COMSOL software was established. On this basis, electromagnetic acoustic emission experiments were carried out and real signals with a lot of noise were obtained. To remove the noise of the electromagnetic acoustic emission signal, wavelet packet transform and complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) was introduced. Firstly, the noise signal of low-frequency was removed by the wavelet packet transform, and the retained node signal was reconstructed. Then the reconstructed signal was resolved by CEEMDAN, and the effective component signal was obtained by the variance contribution rate. Finally, to obtain denoised electromagnetic acoustic emission signal, the effective component signal was reconstructed. The result of simulation and experiment shows that the electromagnetic acoustic emission signal could be effectively obtained by loading electromagnetic excitation, which has greater practical value in engineering applications.

Aiming at the problem that the traditional classification and recognition scheme has few effective features and low recognition accuracy in the process of UAV acoustic signal recognition, a new method based on Gramian Angular Field (GAF) and Convolutional Neural Network (CNN) is proposed. Using the methods of Gramian Angular Summation Field (GASF) and Gramian Angular Difference Field (GADF), one-dimensional UAV acoustic signals are mapped into two-dimensional images. Then, a convolutional neural network model suitable for UAV acoustic signal classification is designed, and the two-dimensional image is used as input, thereby realizing UAV acoustic signal recognition. The experimental results show that compared with the GASF algorithm, the GADF algorithm has higher recognition accuracy and better reliability. The GAF algorithm has more significant features of the UAV acoustic signal than the traditional feature extraction algorithm, and is more suitable for using convolutional neural network for recognition. Compared with other UAV acoustic signal recognition schemes, the GAF combined with CNN method designed in this paper has better performance.

Aiming at the multistatic continuous wave detection problem, the target positioning and motion analysis of bistatic continuous detection are studied. This paper discusses the multi-base continuous detection signal model, target positioning accuracy and target motion factors. The algorithm uses the continuous single-frequency signal as the bistatic continuous detection waveform, and analyzes the target location method. The extended Kalman filter is used to provide the positioning accuracy. The target motion elements of the bistatic sonar system are further discussed, and the Monte Carlo method is used to simulate and analyze the method in this paper. The simulation results show that the pure azimuth positioning method of bistatic continuous detection has better positioning accuracy than the traditional sonar. The main factor affecting the target velocity vector is the estimated radial velocity of the transceiver station.

Fire seriously endangers people's safety. As an important feature of early fire, smoke should be the primary target of fire early warning detection. Aiming at the shortcomings of traditional fire detection accuracy and real-time performance, a smoke detection model based on Yolov3 is proposed. Since there is no community smoke dataset on the Internet, I established my own dataset in various ways, and then improved the Yolov3 network through multi-scale detection training and K-means-dimensional clustering, so that the training network has a certain pertinence in community smoke detection. The features extracted by the Darknet-53 network are reassembled, and multi-scale training is performed using GPU to obtain the optimal weight model to detect smoke targets in images. The experimental results show that under the same data set, the experimental model can detect 96%, and the test set of the community smoke data set also reaches 91.34%. The speed can also be detected in real time, and the smoke and alarm can be detected in advance, which has practical value.

With the in-depth implementation of national environmental protection policies, gas will be more widely used in our lives, which requires better design to upgrade the gas safety. On the basis of the project, our laboratory constructed a gas monitoring and warning visualization platform through a variety of front-end and back-end technologies, enabling managers to conduct real-time monitoring of equipment status in different gas stations through web pages. The system focuses on the collection and analysis of video information, which can effectively identify the identity and action of the personnel entering the gas station, and then mark the personnel and issue an early warning on the video window of the visual page of the system, thus improving the management ability of the gas company of each gas station.

With the progress of science and technology, human-computer interaction technology has become a research hotspot. Many new interaction methods have been developed. As an important organ to receive information, eye plays an important role in human-computer interaction. In this paper, an electrooculogram acquisition system is designed by using five electrodes to measure the amplitude of electrooculogram. The 3Hz and 8Hz components of electrooculogram are extracted by variational modal decomposition. The maximum, minimum, distance between maximum and minimum, mean and standard deviation of these frequency components are used as features. KNN, SVM, CART and Bagging classifiers are used to classify electrooculogram. In the two-dimensional coordinate system, the highest accuracy of the classification of 44 eye rotation angles is 88.40%, which is 12.50% higher than the accuracy of only using the time domain features. It proves the effectiveness of the proposed method and provides a richer interaction mode for human-computer interaction.

In the field of road surveying, we use vehicle laser scanning to get the cloud point data of road. It can generate huge cloud point data. It is difficult to display huge cloud point data quickly, you should organize the data first, then we can display the data quickly. But it is difficult to organize huge data quickly. In this article, we use the file mapping to read the whole data into memory. We cut the whole data into different block, then each block organizes the cloud points into LOD data structure. If the block is in the Frustum, it shows different detail as the sight distance changes.

With the modernization development of the city, the community policing management has become an advanced management mode, and under this management mode, the community video system plays an important role in this management mode. To ensure the normal progress of community governance requires the efficient and high-quality operation of the community video system. In order to ensure the smooth and efficient development of the community policing work, we need to propose the necessary solutions for the color bias phenomenon of the monitoring images. This paper mainly puts forward the solution of this problem from four aspects: resource sharing, technology governance, hardware guarantee and emergency plan, so that such problems can be encountered in the public security practice and have a three-dimensional and rapid response mechanism to deal with such problems. In this paper, the histogram statistical method is used to detect the color bias of the monitoring images. The advantages of this method are the intuition and the rich detection information.

With the rapid development of low earth orbit (LEO) satellite systems, the interference between LEO satellite constellations using the same frequency has become increasingly prominent. Because of the large number and high dynamic characteristic of LEO systems, the traditional interference avoidance methods have great limitations. Based on existing interference avoidance methods, this paper establishes a mathematical model of the interference avoidance for LEO satellite systems, and proposes an interference avoidance method using spatial processing. Aiming at the fatal main lobe interference avoidance problem, the method turns main lobe interference into side lobe interference by switching the beam of earth station to other satellites within the visual range. Then it uses adaptive beamforming method of side lobe null broadening to suppress side lobe interference in high dynamic environment. The interference scenario between OneWeb and Starlink constellations is used for simulation. The results turn out that the proposed method can effectively reduce the interference between LEO constellations and improve the link quality.

The midcourse of ballistic missile is the longest section of ballistic missile flight time. Accurate tracking of ballistic missile in this section can gain more time for warhead identification and determine the best interception point for ballistic missile defense system. In this paper, the method of midcourse target tracking of ballistic missile is studied. The motion acceleration of the mid-course target of ballistic missile in EUN coordinate system and spherical coordinate system is given, and the state equation and measurement equation in the mixed coordinate system and spherical coordinate system are established. The key functions are expressed explicitly, and the UKF algorithm is used to filter and track the ballistic target. Simulation analysis and comparison of the influence of the two coordinate systems on tracking performance are conducted. Simulation analysis compared the effect on tracking performance under two kinds of coordinate system.

In order to solve the problem that the effect of jamming Linear-FM(LFM) fuze decreases in strong noise environment, this paper proposes a WGAN-based LFM fuze signal enhancement technology. Taking advantage of the obvious feature of the LFM signal in the time-frequency domain, we connect two WGANs in series for signal enhancement. The first WGAN learns to reduce the noise in the signal while converting a time-series signal into a spectrogram. The second WGAN learns to enhance the main frequency components of the signal while converting the spectrogram back to a time-series signal. The experimental results show that the algorithm proposed in this paper reduces the noise component and enhances the main frequency component in the signal, which improves the interference effect to the LFM fuze in strong noise environment.

This paper presents a design method of typical damage element simulation generator, and realizes the software based on this method. This method includes characteristic and structure model sub module, parametric design sub module, power parameter setting sub module, power index analysis sub module and so on. This method can provide damage meta related data for damage effectiveness evaluation and fire plan.

Absolutely maximally entangled (abbreviation AME) states are pure multipartite states that, when reduced to half or less of the sites, the maximum entropy mixed state is obtained, which have found significant applications in teleportation and quantum secret sharing. In this paper, we consider the sufficient conditions for the existence of the minimal support AME (n, q) state, and present a systemically explicit construction about minimal support AME states, then several minimal support AME states are drawn by employing GRS codes, Singleton arrays and their deformed arrays.

In view of the fact that the global satellite positioning system cannot be applied to the indoor environment, it is of great practical significance to carry out research on the positioning problem in the complex indoor environment. UWB (Ultrawide Band) positioning technology has a good application prospect in the task of the armed police force, but due to the complexity of the indoor environment, the UWB application environment is usually a NLOS environment (Non-Light of Sight), and there are a large number of obstacles blocking the line of sight of the signal. This paper firstly introduces the mainstream indoor positioning methods according to the required reference points, and introduces the research status, technical principles, TOA (Time of Arrival) estimation algorithm, NLOS identification and compensation algorithm of UWB positioning technology in detail. Using the relatively simple and classic support vector machine learning algorithm to deal with the problem of NLOS channel identification.

A general scheme to sample signals defined on the nodes of a graph is proposed. In this paper, we study the sampling and recovery of graph signals with successive local aggregations under the graph fractional Fourier transform. In the graph fractional Fourier domain, we show that fractional bandlimited graph signals can be perfectly recovered. Our focus is on reconstructing fractional bandlimited graph signals. This sampling strategy, unlike the graph's subset selection scheme, obtains observations from a single node. We give examples of graph signal sampling to demonstrate that fractional bandlimited graph signals can be perfectly recovered and compare it with graph aggregation sampling.

Template matching technique is a common technical tool in underwater acoustic target recognition. The traditional template matching method calculates the distance between the test sample and the template on the basis of feature extraction and compression, and characterizes the similarity by the distance measure, which requires high feature extraction and compression techniques. This paper proposes a Siamese network-based multi-sub-band matching method for the LOFAR spectrum of underwater acoustic targets from the target LOFAR waterfall spectrogram. The method fully considers the limited sample of underwater acoustic target data and the difference of energy distribution of the target in different frequency bands, divides the LOFAR spectrum into multiple sub-bands, and constructs a multi-sub-band Siamese network model under small sample conditions, which is suitable for solving the matching recognition problem of underwater acoustic targets. The actual sea trial data verified the effectiveness of the proposed method.

Communication and sensing are the basic functions of IOV (Internet of Vehicle), it is an efficient and economical way to realize radar sensing function by using communication signal. In this paper, it is proposed a target DOA estimation scheme based on a series of super-resolution algorithms in which the 5G upstream SC-FDMA signal is used as the radar probing signal and a uniform linear array is used to receive the target echo. Modeling analysis and simulation results show that the proposed scheme can accurately find the direction of radar target.

In the process of EEG signal acquisition, it is inevitable to be affected by some artifacts, for the traditional multi-channel EEG signal, researchers have proposed a variety of methods to remove artifacts. However, with the development of wearable EEG acquisition systems, only single-channel EEG signals are collected in many cases. In order to solve the problem that the artifact of single-channel EEG signal is not easy to deal with, this paper proposes an artifact identification and removal method that combines variational mode decomposition (VMD) and second-order blind identification (SOBI). First, the original signal is decomposed into a series of intrinsic modal function (IMF) by the optimized variational modal decomposition. The idea of blind source separation is introduced, and the IMF is used as the input signal of SOBI. Then, the fuzzy entropy and correlation coefficient comprehensive indicators are used to identify extract the components containing artifacts and remove them. Finally, the pure EEG signal is reconstructed by inverse transformation. The semi-simulation experiments show that the proposed method has a better effect of removing EMG artifacts than other existing methods, and can retain useful information to the greatest extent.

Cracks are one of the early manifestations of concrete building diseases. Timely detection and repair of cracks is very important for building safety. In this paper, the holistically-nested edge detection(HED) network is applied to crack detection to realize the preliminary detection of concrete cracks. Compared with the traditional Canny operator, the cracks detected by HED are clearer. In view of the false detection caused by the interference of impurity areas when HED detects cracks, k-means algorithm is used to cluster and divide the HED detection results according to the pixel gray value, so as to obtain the cracks, background and impurity areas. The impurity area is modified into the background area to remove the impurity interference. Compared with the HED detection results, the precision of the detection results after k-means clustering is improved to 86.533%, and the recall rate is slightly reduced, but it is still as high as 88.791%.

The reception effect of celestial electromagnetic wave reflected by reflector is an important indicator to evaluate the performance of radio telescope. In this study, we present a model based on Monte Carlo to evaluate the reception effect of reflected signals. Taking the FAST as an example, firstly, the reflected light of each reflection panel has a certain projection area on the plane where the effective area of the signal received by the feed cabin is located. Considering the different cases where the reflected signal projection overlaps the plane of the feed cabin, Monte Carlo method is used to simulate the case of the reflected signal received by the feed cabin. Then, combined with boundary conditions, the triangular reflector within the target range is traversed and searched. Finally, the reception ratios of the feed cabin and the datum sphere are calculated. By analyzing their electromagnetic wave receiving effects, we find that the reception ratio of the feed cabin is greater than that of the datum sphere. The evaluation model of the reception effect of reflected signals proposed in this paper has certain reference value for realistic modulated radio telescope observation of celestial objects.

For 2D image data target detection, this paper proposes an improved detection algorithm for YoloV5s based on the attention mechanism. By replacing the Focus module in YoloV5s with a CBL module, it facilitates model export. The four attention modules, SELayer, EcaLayer, CBAM and CoordAtt are added to the convolutional network of YoLoV5s. Before the P4 layer, before the P5 layer, before the SPPF layer and before the last layer of the original backbone network, respectively. By experimenting with algorithms that include YoloV5s, It was concluded that among the five algorithms tested for evaluation. The proposed YoloV5s-CoordAtt has the best performance level in terms of accuracy, with a 4.48% improvement compared to the original algorithm.

Situational awareness is one of the main functions of Airborne Early Warning Radar, early warning radar surroundings through active detection awareness, this requires detection distance, wide range of monitoring. Detecting ability affects the decision makers for situational judgment, is of great significance. This paper discusses the influence factors of early warning radar detection performance, the first part of the characteristics of airborne radar internal factors influencing the performances of radar detection. The second part in combination with actual combat environment external factors influencing the performances of radar detection is analyzed. The third part according to various factors put forward ways and means to enhance the capacity of early warning radar to detect. Only fully understand the factors of restricting the detection ability to display the early warning radar performance to the optimal.

3D visualization of medical images is not only a popular research direction in the field of medical image research, but also a powerful auxiliary means for doctors in the surgical process. This paper sorts out the widely used 3D visualization software for medical images, introduces each software, analyzes its characteristics, and concludes that the software can be further developed in combination with 5G technology.

In the analysis of blood component content, spectral method has been widely concerned for its rapid and non-invasive advantages, but there are many interference factors in the process of data collection. This will cause a large deviation in the spectral data of some samples. If it is directly used in modeling analysis, it will introduce bias to model establishment and result in large deviation in prediction. Therefore, it is necessary to screen the samples before establishing the model, in this paper, a method of sample screening using sigma criterion is proposed for noninvasive blood component analysis, The dynamic spectrum (DS) value was extracted by single edge method which is the mean value of the effective slope of the rising edge of the photoplethysmographic (PPG) signal was taken as the DS value. The DS value fluctuates in a certain range due to the change of blood component concentration. The samples beyond the range of 1 standard deviation were excluded and screened. Partial least squares (PLS) linear regression models were established for fasting blood glucose, white blood cell and albumin in all samples before and after screening, and the whole sample was modeled and analyzed. Then the sample set was divided into training set and test set according to the ratio of 10:1, and the three blood components were predicted and analyzed respectively. The experimental results show that the modeling test of three components, respectively, after screening sample of the precision of the sample model, as well as training set and testing set of classified sample sets predict performance after got greatly ascend, improves the correlation coefficient, reduce the root mean square error of the result of the experiment fully demonstrated the effectiveness of the method of sample selection.

SPECT image is a commonly used nuclear medicine imaging instrument and plays a very important role in the imaging diagnosis of bone metastasis disease. As lung cancer is one of the malignant tumors that endanger human life and health, early detection and treatment is the only way to delay the life cycle of this disease. To accurately segment lung cancer bone metastases from SPECT images, different segmentation models oriented to the reliable assessment of bone metastases were studied and constructed in this paper. First, data pre-processing, such as normalization, is performed on SPECT bone metastasis images. Then, improved image segmentation models are constructed based on the U-Net model, including the U-Net model that fuses the residual structure and the U-Net model that fuses both the residual structure and the attention mechanism; finally, experiments were conducted on a set of real clinical SPECT nuclear medicine data, and the multigroup experimental demonstration showed that the U-Net model incorporating residual structure and attention mechanism proposed in this paper could better segment the bone metastases in the thoracic region of SPECT images, and the obtained pixel accuracy PA, category average pixel accuracy MPA, and average cross-merge ratio MIoU were 0.9955, 0.7824, and 0.7291, respectively.

Early screening of the diabetic foot is crucial for diabetic foot prevention and treatment. Since neuropathy and vasculopathy caused by hyperglycemia can lead to abnormal foot temperature, a diabetic foot screening method based on infrared thermography has been recommended. We designed a fully automated method for early screening of the diabetic foot. The automatic segmentation of infrared images of the foot region was achieved by designing an image processing algorithm; the automatic extraction of various temperature features was achieved by combining the angiosome concept. SVM was used to establish a classification model. The study recruited 41 diabetic patients (diabetic group, diabetic foot Wagner grade 0) and 48 healthy individuals (control group) to validate the reliability of the method and showed an accuracy rate of up to 87.5% in the test set.

This paper presents a method that can accurately diagnose Alzheimer's disease. Normalized whole brain volume, normalized gray matter volume, and normalized white matter volume were obtained by segmenting amyloid PET images. Register to a unified MNI spatial template and perform spatial standardization, and select the optimal two-dimensional slice retention level in all volume images as experimental data. Finally, the extracted features of each brain region are classified by EfficientNet-B8. The correct rate of whole brain classification: 0.987017; the correct rate of white matter classification: 0.998665; the correct rate of gray matter: 0.969573; the test results show that this method can more accurately distinguish mild AD patients and normal elderly compared with existing methods. Contribute to the prevention and early diagnosis of AD disease.

Whole-body bone scan is an effective tool for diagnosing bone metastasis from malignant tumors, including lung cancer. However, due to the low spatial resolution of Scintigraphic images, manual analysis of the images is very difficult. In order to accurately diagnose bone metastases, a self-defined convolutional neural network is studied and constructed in this paper for detecting lung cancer-Caused skeletal metastasis in bone scan images Firstly, since thoracic regions such as ribs and spine are the most frequent areas of bone metastases, we preprocessed the collected SPECT whole-body bone imaging data to extract the thoracic regions. Then based on the feature that ResNet can alleviate the vanishing gradient problem caused by increasing depth in deep neural networks and the feature that attention can focus on important information to improve network expression. A deep SPECT image classification network Att-ResNet24 with residual module and hybrid attention mechanism is developed for classification of these images, including bone metastases and non-bone metastases. Experiments on a set of real SPETC whole-body bone images show that self-defined classification networks can be effectively used for the detection of bone metastases, obtaining the best scores 0.737, 0.744, 0.736, and 0.735 for accuracy, precision, recall, and F-1 score, respectively.

SPECT imaging is an effective medical method for diagnosis, treatment, evaluation, and prevention of a range of serious diseases and medical conditions. SPECT bone scans have the potential to provide more accurate assessment of disease staging and severity. To classify medical images more effectively, in this work, we propose an AlexNet-based image classification model. Specifically, the network structure of AlexNet is fine-tuned and attention mechanism is added to construct an improved model. Lastly, experimental results show that the improved AlexNet classification model is feasible for classifying SEPCT bone scan images, with obtaining scores of 0.7377, 0.7677, 0.7723, and 0.7674 for Acc, Pre, Rec and F-1, respectively.

Magneto-Acoustic-Electrical Tomography (MAET) is a multi-physics coupling imaging technology based on the principle of Hall effect, which can be used to detect the electrical conductivity distribution in tissues. Traditional MAET is considered to occur in soft tissue medium, which is regarded as fluid when calculating. But in fact, there are some solid media in human body that will affect our detection, such as skull. Based on the head, this paper analyzes the influence of solid medium on MAET process. Using finite element software, the influence of skull on sound field and electric field is analyzed firstly, and MAET signal of head considering the skull is obtained. The results preliminarily verify the feasibility of applying MAET to the situation with solid medium.

This paper aims to effectively utilize the vast amounts of data generated by data centers, which are used to support fault diagnosis and early warning functions. Due to the large volume and complexity of data, as well as the semantic relationship among data, in this paper, we adopt knowledge graph technology to extract, fuse, process, and update the knowledge in data centers. Then, we provide a feasible method for constructing a knowledge graph for fault diagnosis and early warning in a data center by describing the correlation of data, reasoning and analyzing the data on a reasonable basis. In addition, we also discuss how knowledge is represented.

Age-invariant face recognition (AIFR) can be used in law enforcement, security, and citizenship verification. Face recognition as age changing is huge and tough task for human. In this paper, the features of faces across ages will be studied and cross-age face recognition will be accomplished, based on the Convolutional Neural Network (CNN) algorithm and other modules. This paper aims to reduce the impact of age on face recognition and modify the model based on previous scholars' research. Open-source datasets will be used, and then a deep learning model will be exploited to find results on the ranking 1 accuracy of about 90%. The results and models comparison will be discussed in the paper.

Information hiding refers to embedding secret information in an imperceptible way with the help of redundant information in the multimedia carrier without affecting the sensory characteristics and use value of multimedia, and transmitting it in an open channel. Deep learning is a hot research direction in the field of machine learning in recent years. It allows machines to learn the inherent laws and representation levels of sample data, and ultimately enables machines to have the ability to analyze and learn like humans. Based on the baluja algorithm and the ISGAN algorithm, combined with the Inception module in GoogleNet and the relevant knowledge of the residual network in deep learning, this paper designs an image steganography algorithm suitable for hiding various images. It has been proved , the algorithm solves the problem of color distortion of the stego image in the baluja algorithm, reduces the gap between the stego image and the cover image, and improves the security of the stego image.

The semantic segmentation technology of remote sensing image refers to labeling the semantic information of pixel-level of the image to complete the classification, namely, terrain classification. It is widely used in intelligent maps, smart cities and other aspects. With the increase of satellite image resolution and the development video communication moonlet, its application and scenes are greatly broadened. The traditional remote sensing image semantic segmentation method mainly uses statistical machine learning methods, which cannot take into account the spectral features and the context semantic relationship of pixels, and has a bottleneck in improving the accuracy of classification. In deep learning method, Using the convolutional neural networks to extract features can achieve classification results of remote sensing images with higher classification accuracy. Aiming at the problems of existing deep learning methods in multi-spectral image semantic segmentation, to make full use of the information of multi-spectral images, this paper proposes a semantic segmentation algorithm for multispectral remote sensing images based on deep learning and verified the method on open data sets.

With the development of computed tomography (CT) technology, how to reconstruct high-quality pathological images under low-dose X-ray radiation has become the focus of attention. Compressed sensing theory has become a research hotspot because of its good reconstruction quality under incomplete data. In compressed sensing theory, the design of projection matrix and total variational model is the key factor to determine the quality of reconstruction. In this paper, after the cone beam projection is rearranged into fan beam, a fan beam projection matrix model is designed. Based on the traditional distance driven model, the forward and back projection distance driven model is used to roughly estimate the image, and the multilevel scheme (MLS) is introduced to further improve the reconstruction quality of the image. The experimental results show that under the condition of incomplete projection data, the reconstructed image has less artifacts, complete edge details and clear image.

The test methods of autonomous driving are mainly divided into SIL, MIL, VIL, real vehicle test, etc.; the scenario-based virtual simulation verification method has huge technical advantages in test coverage, test repeatability, test safety, etc., and is an effective method for future autonomous driving function test verification. With the improvement of technology and demand, autonomous driving developers have gradually increased their investment in simulation testing. How to evaluate the comparison between simulation test results and actual test authenticity is a major challenge for autonomous driving. The high-definition map software used in this paper can realize intelligent lane fitting and fast connection, and realize fast and smooth connection in advance while ensuring the consistency between the edited road network and the actual environment. By collecting real scene maps, a high-definition map conforming to the Open Drive format that can be used in real vehicles and simulations is generated, which solves the problem of common use of lane-level high-definition maps in intelligent driving vehicle testing and simulation testing.

Information encryption is an important part of information transmission in modern society. At present, image encryption algorithms face problems such as applicability and security, which need to be solved urgently. This paper proposes an image encryption algorithm based on Arnold-Logistic sequence encryption, which uses the idea of Euler concatenation of fractions to chunk the non-equal-length images in order to perform Arnold transform. Then the random key stream is obtained for the use of Logistic sequences to perform the heteroskedastic operation on the image pixels. After experiments, the algorithm in this paper is proved to be effective in improving the cipher image security, extending the applicability of the algorithm, and also greatly reducing the computing time and space occupation.

Aiming at the complex environment in the inspection of underground heating pipes, including high-temperature flanges, valves, and stale water stains on the surface of heat pipes, a method for detecting water leakage in heating pipelines combining thermal infrared and visible images is designed. Temperature anchor frame detection is performed on thermal infrared images, and deep learning is used for visible image for water stain anchor frame detection. This paper proposes a detection method for determining the degree of correlation and high temperature point to determine the water outlet point based on the position coincidence degree, which excludes the infrared interference caused by the high temperature of the flange and valve, and the visible light interference caused by the old water stains, and has a good detection result for the leakage of the heat pipe.

In order to improve the texture of product packaging and solve the problem of poor color constof product packaging, it is necessary to study and design the packaging visual optimization system based on graphic elements, improve the texture of product packaging from material, aesthetic and context dimensions, and realize the optimization of visual texture of graphic elements. And the use of the characteristics of color constancy, in enhancing the color saturation, to ensure that the product packaging color is inconvenient. Its graphical elements can also be searched in the database of packaging visual optimization system. Finally, after the system test, it is also proved that the system can effectively optimize the texture of the product packaging. Therefore, it is necessary to further strengthen the optimization of product visual system.

Brain-computer interface (BCI) technology is a human-computer interaction technology that realizes the connection and communication between the human brain and external equipment by collecting and decoding electroencephalograph (EEG) signals. It is a major direction for the development of BCI to analyze and decode the incoming EEG signal in real time through online BCI systems to enable the control of external devices. In order to improve the efficiency of online BCI, it is important to further improve the recognition accuracy and reduce the classification time. Long Short-Term Memory (LSTM) network can improve the recognition accuracy by using the temporal characteristics of EEG signals through its temporal recursive structure. Therefore, the optimized Bi-directional Long Short-Term Memory (BiLSTM) can well express the long-term dependent information in the input. In the study, a new model was proposed, the Attention mechanism was introduced to the BiLSTM while integrating Convolutional Neural Networks (CNN). In the case of avoiding gradient disappearance or gradient explosion, the problem of ignoring correlation information between channels is solved, and learning high-dimensional features in EEG data. This study improves the accuracy of motor image (MI) EEG signal classification under the condition of saving the steps of feature extraction, which is of great significance for the development of online BCI systems.

Aiming at the problem of model instability and overfitting of deep neural networks with the deepening of the number of network layers, the current mainstream method is to use batch normalization (BN) to alleviate them. However, since the BN method is more sensitive to batch size when the batch size is small, the model performance will be poor. For a relatively large model, due to the limitation of video memory, the batch size cannot take a large value, limiting the model's performance. Because of the dependence of BN on batch size, this paper adopts group normalization (GN) instead of batch normalization (BN) in the UNet network to alleviate the impact of the model on batch size. Then experiments are carried out on the WHUBuilding dataset. The experimental results show that the improved model (UNet-GN) improves the mean intersection over union (MIoU) and mean pixel accuracy (MPA) by 10.66% and 1.65% respectively compared with the original model (UNet-BN).

In line structured light vision measurement, an adaptive image enhancement algorithm is proposed to address the problem that the conventional image enhancement algorithm is ineffective in enhancing the laser stripe features in the unevenly illuminated images acquired by the system. The algorithm separates the channels in HSV color space, processes the separated S and V channels separately using an iterative adaptive histogram equalization algorithm to enhance the image saturation and brightness, and merges the equalized S₁ and V₁ channels with the original H channel to convert the image to RGB color space to complete the image enhancement. The experimental result shows that the information entropy index value S of the proposed algorithm is higher than those of other comparative algorithms, and the algorithm performs better in laser stripe image enhancement with better robustness.

For visual perception of indoor mobile robots, real-time segmentation of 3D objects is a very challenging problem. Due to the complexity and disorder of indoor point cloud data, many methods have been proposed to improve the segmentation accuracy, which most of them cannot meet the requirement of rapidity. Therefore, this paper proposes a method for fast segmentation of 3D point clouds based on supervoxel density clustering. First, the ground under the indoor scene is removed by using plane fitting algorithm and the point cloud data after ground removal is denoised at the same time. Then, the target object is over-segmented to obtain the supervoxels with local geometric features. The core points of the nearest neighbors at the center of the supervoxels are searched out using the kdtree indexing acceleration. Finally, the density clustering of the supervoxels is performed based on the core point density up to. Experimental results on publicly available datasets show that our proposed method can effectively improve the speed of the algorithm while ensuring accuracy.

Since December 2019, some hospitals in Wuhan, Hubei Province have found many cases of unidentified pneumonia, which were later named COVID-19. In recent years, the disease has spread widely all over the world. Many researchers are trying many methods to make rapid discovery and diagnosis for COVID-19, which is important to facilitate the treatment of the disease and to reduce its transmission. In this study, we propose the comparison of methods for COVID-19 pneumonia diagnosis based on chest X-ray (CXR) images using Transformer and CNN. Both of them are two of the most commonly studied deep learning method. In this paper, all four models will be trained with the COVID-19 Chest X-ray Database and tested for the performance on the four-class classification. The results showed that Swin Transformer achieved the highest score in the comparison of accuracy as 0.916. Additionally, VGG as one of CNN architectures performs well on the recognition of COVID-19 positive. This study can help doctors reduce their workload and effectively improve the speed and accuracy of detection of COVID-19 and other viral pneumonia.

Conventionally image translation is used to convert synthetic aperture radar (SAR) images to optical ones to increase interpretability. Due to the different imaging natures of SAR range sensing and optical directional projection, it is within the expectation that some kinds of SAR images could be reasonably translated into optical ones, while others not. Arguably, less good patches in the translated image could act as a potential indicator of salient differences of the two imaging mechanisms. Following this line of thought, considering in urban SAR images, the layover is mainly caused by buildings, we proposed a method to detect building layovers in SAR images by evaluating the correlation of the original SAR image and the translated one by the cycle-translation, that is, from SAR to optical to SAR translation. Because the presence of building is the main factor of the layover phenomenon in urban SAR images, building layover areas in SAR images are expected to be less correlated, or they should have a low correlation coefficient. Preliminary experiments validate our underlying principle and method.

In recent years, with the rapid development of science and technology in my country, people's requirements for computer information technology have become higher and higher. In order to better explore new things, engineers and technicians need to continuously optimize computer information processing systems to help people better solve practical problems. The article mainly analyzes the development process of computer intelligent image recognition algorithms, expounds the computer intelligent image recognition algorithm, and finally puts forward the technical application of computer intelligent image recognition algorithm, which provides a certain research foundation for related staff.

With the growth of image classification models and the development of deep learning, the level of computer image classification has basically surpassed that of humans, but the level of fine-grained computer image classification is still weaker than that of humans, so fine-grained image classification is currently given a lot of attention. This paper carries out visualized analysis for the fine-grained image classification in terms of cooperation, scientific research, research trend and other aspects by CiteSpace software, and also points out that fine-grained image classification relies on CNN while looking into Transformer. However, fine-grained image classification has the problems of model reliance on prior knowledge (CNN), lack of sufficient data (Transformer), and lack of local relevance (CNN/Transformer), which can be solved by reducing the complexity of the model, transfer learning data augmentation, combination of CNN with Transformer, etc. On this basis, this paper indicates that Swin-Transformer has good generalization, hierarchy and translation invariance, and will become the main research direction for fine-grained image classification in the future.

Extracting feature parameters is the most important part of the current acoustic sediment classification methods. Aiming at the problems of multiple selectivity and instability in the sediment classification after feature parameters extracted from the sonar image, it was proposed that inputting sonar images directly into the improved convolutional neural network(CNN) for seabed sediment classification without extracting image features. This paper compared the classification results of 12 types of sediment classification models based on feature extraction and four classical CNN models. It showed that the classification accuracy of the former ranges from 39% to 92.6%, which was greatly influenced by the types of feature parameters and classifiers; the classification accuracy of the latter ranged from 84.2% to 89.6%, among which AlexNet presents the best effects in CNN. Then the particle swarm optimization(PSO) algorithm was used to optimize the five training parameters in the AlexNet model. The same training set and test set were used for comparative experiments. And the results showed that the PSO-AlexNet model can obtain 93.8% of the seabed sediment classification accuracy. And the classification effects are better than the sediment classification model based on feature extraction.

Texture has a wide range of applications in the field of image processing. The application of existing texture research methods in DEM is mostly texture feature extraction and index calculation. However, the formation of landforms is caused by terrain relief. The existing texture analysis methods generally have a deficiency in the reflection on the human visual system. In this paper, the visual boundary lines, visual outline lines, and skeleton lines were used as the DEM relief textures that can visually reflect terrain relief. Then the extraction method is refined in conjunction with digital terrain analysis theory. Finally, a multi-dimensional visualization model of DEM relief texture was built, which achieves a better visual effect and provides new ideas for digital terrain analysis, and also has a wide range of popularization and application value.

In the post-harvest supply chain of fruits and vegetables, the spoilage rate during storage is as high as 10% due to improper management measures, which has seriously restricted the development of related industries. In response to this problem, this paper proposes a method for predicting the storage shelf life of fruits and vegetables. In this paper, we develops a micro-environment monitoring system based on the Internet of Things (LOT) technology, integrating the sensors of temperature and humidity, oxygen and carbon dioxide, using the LoRa wireless communication module to realize real-time data transmission; With the help of MATLAB platform, we build a multi-sensor data fusion model based on artificial neural network (ANN), and we convert the environmental parameters with multi-source coupling relationship into shelf life by this model. The results show that the model has good prediction performance and meets the requirements of shelf life prediction. The method can effectively reduce the economic losses caused by fruits and vegetables spoilage, when being applied to the post-harvest supply chain management of fruits and vegetables.

Breast cancer is the most prevalent malignant disease among cancers, and its mammography and examination reports are an important basis for clinicians to judge the nature of breast tumours. Current research based on mammography reports has focused on the natural language classification models, ignoring the practical problems of the semantic hierarchy in examination reports and the imbalance of benign and malignant categories. To address this issue, we propose an interpretable classification model of breast tumors with tabular mammography data (STaNet). The model is based on a semantic tree to structure the image presentation part of the mammography report, allowing it to focus on the disease information carried at the semantic level of the report better. Meanwhile, we dynamically conbine the focus loss function parameters with the category loss, allowing the model to dynamically adjust its focus category as it learns. We evaluated our model using real-world mammogram report data and compared the model with other excellent models. The experimental results show that the accuracy of our proposed model is better than other models and its interpretable results are in line with the diagnostic criteria of pathologists.

In recent years, human motion simulation and virtual reality have developed rapidly. The computer field mainly takes the natural cooperation of any computer as the main form, and carries out relevant research. This motion capture system can accurately and real-time record the motion of human body, and further obtain the different characteristics of human motion at different times by analyzing the relevant three-dimensional data. Nowadays, this technology is widely used in animation video production, dance art, sports training, competitive gymnastics and other fields. Using this technology, the connection between three-dimensional technology and practical technology is realized. According to the investigation and analysis, the capture of dance limb movement form is fully combined with computer application technology. At present, it is not comprehensive and in-depth in dance education and teaching. This paper verifies the operability of this point, mainly studies how to apply this technology to actual dance art education and teaching, and puts forward the analysis path of dance limb movement based on feature vector matching, It can be reasonably developed and used in dance art education and teaching to effectively improve the efficiency of dance art education and teaching.

At present, the wireless gateway is most be used to achieve wireless communication between Profibus-DP master and slave stations. Due to the long time delay of wireless gateways, it can only be used for low baud rate communication. As such, a communication module using wireless repeater solution is designed in this paper, to realize Profibus-DP wireless communication. This article describes in detail the communication timing requirements of Profibus-DP and the software and hardware design of the wireless communication modules, especially in accordance with the Profibus protocol and timing requirements to realize the wireless communication of Profibus-DP at a high-speed communication rate.

To get the shape information of the freshwater fish in the factory environment, a freshwater fish contour detection algorithm that can resist image distortion was proposed. This method consists of three parts: distortion correction, background subtraction, and ellipse fitting. Firstly, a projection matrix of pixel points between the test image and the template image was built to correct the distortion of the image. Secondly, the ROI of the test image and template image was matched to remove the background to get the fish body image. Finally, several times of ellipse fitting method was used to ensure that the main part of the fish body was accurately fitted. The experimental results showed that the proposed algorithm can ensure the error variance within a range of 15%, greatly improving the detection accuracy in comparison to the traditional background difference and ellipse fitting algorithms.

Using Unmanned aerial vehicles (UAVs) to assist mobile edge computing (MEC) network is a promising solution to provide flexible and low-latency computing service for resource limited user equipments (UEs). However, the computing resources that one UAV can provide are often limited and cannot provide computing services for massive mobile users, thus limiting the application of this technology. So, a novel multi-UAV-assisted MEC network is proposed in this paper. The optimization problem is formulated with the aim to make full use of network’s computing resources to reduce the number of failed tasks and energy consumption. To solve this challenging problem, a novel solution is proposed. First, we use a swarm intelligence algorithm to investigate the deployment of UAVs. Second, we offer an efficient matching algorithm to explore the tasks assignments under the given deployment of UAVs. The simulation results show that the proposed solution can effectively reduce the number of failed tasks and energy consumption.

The research of cross-domain mission planning and data security is an important component of developing multi-level cross-domain collaboration. The drawbacks and problems of traditional cross-domain mission planning include geographical physical isolation, very different authentication models, and transmission hierarchy. In order to achieve efficient cross-domain mission planning and meet the collaboration needs of different domains, a cross-domain dynamic authorization mission planning scheme is proposed based on blockchain technology. It first constructs a cross-domain mission collaboration organization architecture that enables each mission domain to join the consortium chain network with a unified authentication, then we designed a method process of dynamic authorization and a smart contract of cross-domain mission collaboration, which can transfer and control the authority of cross-domain mission planning and ensure the supervisibility and traceability of each process of mission collaboration. The experiments show that the scheme can achieve effective control of the cross-domain mission implementation process, and meet the timely requirements of cross-domain mission planning by running efficiently.

Space-Air-Ground Integrated network(SAGIN) is a heterogeneous network which combines ground network, air network and space network. In SAGIN, a sub-network access algorithm based on reinforcement learning is proposed to make full use of multi-dimensional network resources, improve network delay and reduce packet loss rate. The algorithm learns from the environment iteratively and revises the model constantly, so as to make the optimal access choice. In the simulation experiment, the ASA-RL algorithm has obvious improvement over the comparison algorithm in terms of communication delay and packet loss rate.

Radar Embedded Communication (REC) is a covert communication method that embeds communication waveforms in radar backscattered echoes. In order to solve the problem that the conventional REC waveforms are not orthogonal, this paper proposes a design strategy of orthogonal REC waveforms, which can achieve complete orthogonality between REC waveforms. The simulation results show that the designed waveform can reduce the symbol error rate (SER) of the receiver and improve the communication reliability without sacrificing the performance of low probability of interception (LPI).

As the basis of pedestrian and traffic statistics, multi-object tracking (MOT) is widely used in smart cities and smart shops. Due to the high cost of cloud analysis, MOT algorithm is mainly deployed on the end-side device. However, limited by computing resources and capital cost, it is difficult to deploy large models in end-side device. As tracking based on detection has become the most effective MOT method, a lightweight detection network model YOLOv5- Concise with only 0.302M parameters is proposed for end-side deployment in this paper. Then, the application of knowledge distillation technology based on outputs and feature relation in object detection is studied. By knowledge distillation, the YOLOv5-Concise model's detection precision mAP0.5 is increased by 7.26%, mAP0.5:0.95 is increased by 16%, and the detection speed remained unchanged. Finally, on the camera equipped with T40 chip, the detection speed of the model is measured to 28 FPS, and the accuracy rate of 98.1% is obtained in the pedestrian statistics test.

According to the needs of engineering support simulation field, this paper proposes a design method of engineering support simulation support software, and realizes the corresponding software based on this method. The method includes engineering support simulation initialization and data receiving module, engineering support equipment simulation module, obstacle setting simulation module, obstacle removal simulation module, motion support simulation module, engineering support simulation data output and recording module. Based on this method, the problem of engineering support simulation can be solved.

This paper presents a design method of power field reconstruction system based on empirical formula, and realizes the software system based on this method. This method includes five parts: warhead structure parametric modeling subsystem, warhead power field calculation and simulation subsystem, warhead power field three-dimensional presentation subsystem, model setting and management subsystem, database management subsystem and so on. There is data exchange between various parts. Based on this method, the power field can be reconstructed.

It is an urgent requirement for the development of smart grid to allocate resources effectively to satisfy the changing power service environment and the needs of power users. 5G network slices provide the solutions. For dynamically demands of resources in smart grids, reinforcement learning is introduced to address the resource allocation problem of RAN slices in this paper. The training process in reinforcement learning is divided into two stages: slice layer and user layer. The slice layer performs the action of resource allocation, and the user layer calculates and feeds back the reward of the action. This paper focuses on how resources in slices are allocated to users, that is, resource reallocation in the user layer stage. The evaluation function of quality of experience (QoE) is defined, and a fair allocation algorithm that maximizes the utility of QoE to improve the experience quality of user is proposed. The simulation comparison results indicate that our algorithm combined with reinforcement learning not only improves the resource utilization, but also maximizes the user's QoE, which has obvious advantages over other methods. Moreover, the training cost of reinforcement learning is reduced and the convergence speed is accelerated.

Extreme wind event is one of the major natural disasters in complex terrain areas, and it is also an important factor to be considered in construction projects in different regions. In order to analyze the simulation ability of numerical model during a synoptic process under the complex terrain in the Tibetan Plateau, the wind data observed from the wind lidar observation system and the high-resolution numerical model data simulated by WRF model were be used in this study. The statistical analysis methods was used to investigate the horizontal distribution characteristics of the wind around the observation point (OP) region in detail. The results show that the time series of wind observation data between the different heights in the near-surface boundary layer have well consistency, the correlation coefficients are 0.9777, 0.9856, and 0.9436. The simulated wind is flowed from the mountain ridge to the mountain valley around the OP region. The strong wind may occur when the southerly air flow prevails around the OP region, on the contrary, the strong wind may not occur when the westerly air flow prevails. The differences of the maximum wind speed between simulation and observation are large at 40 m height and small at 160 m height, while the differences of the minimum wind speed are small at 40 m height and large at 160 m height. There is a significantly correlated between observed and simulated wind speed, and the correlation coefficient are 0.3089 at 40 m height, 0.3063 at 100 m height, and 0.2522 at 160 m height. The model can be used to study the wind characteristics under the complex terrain where the lack of observation stations. The observation data and analysis results are expected to be useful for the wind-resistant design of construction projects under the complex terrain regions.

The space-ground integrated network (SGIN) is a future communication network that integrates different dimensions networks to form a new network application scenario based on the terrestrial network and extended by satellite network. SGIN can cover natural spaces such as space, ground, ocean, etc., and can provide users with more diverse network services. As a new type of network, SGIN also faces the problem of network ossification in traditional network. In traditional network, network virtualization can overcome this problem effectively. However, most of the current researches are based on terrestrial network models or satellite network models, which cannot adapt to the dynamic and heterogeneous nature of SGIN. In order to meet the demand of SGIN for network virtualization, we establish a SGIN time evolution graph model and propose a virtual network embedding algorithm (CD-VNE). The algorithm selects the network sub community with the most abundant physical resources through community detection technology, which can reduce the consumption of link bandwidth resources and the risk of satellite link failure. The simulation results show that CD-VNE can increase the Revenue/Cost ratio and acceptance ratio.

By using the B3LYP method in density functional theory, the structural optimization of pelargonidin, cyanidin, delphinidin, malvidin, petunidin and peonidin were performed under the vacuum model. Under the same calculation level, the signal processing calculation was carried out through computer software, the calculation results were displayed in the form of images, and the infrared vibrational spectroscopy, ultraviolet-visible absorption spectroscopy, and fluorescence spectroscopy were obtained. At the same time, Mullinken charge distribution and frontier molecular orbital energy levels were predicted and analyzed. Under vacuum conditions, the molecular configuration of each anthocyanin is in a stable structure. The molecular stability of malvidin is the weakest, and the molecular stability of pelargonidin is the strongest, followed by peonidin, cyanidin, delphinidin and petunidin, respectively.

In order to solve the problem of time-consuming and laborious manual labeling of garment tags during the garment shelving process, an improved Inception-v4 garment tag recognition network is proposed. The introduction of residual connections to the Inception module in the recognition network speeds up convergence in the case of deeper networks. And the performance of the model is improved by using Soft Label and a loss function based on inter-class distance. Subsequently trained and tested on the Fashion AI dataset, the experimental results show that the method described in this paper achieves good performance with faster convergence and a 4.1% improvement in accuracy.

Using the B3LYP method in density functional theory (DFT), theoretical calculations and molecular simulations of vitamin B2 were carried out under the -311G (d,p) basis set, and the structure was optimized. The charge distribution, infrared spectrum, ultraviolet-visible absorption spectrum, fluorescence spectrum, frontier molecular orbital energy level and nuclear magnetic resonance spectrum of the fully optimized vitamin B2 were simulated and predicted and analyzed.

With the rapid development of Internet and IoT, the demand for system capacity and spectrum efficiency is increasing, and Non-Orthogonal Multiple Access (NOMA) is being discussed as an evolutionary technology standard for 5G, which can meet the requirements of large-scale connectivity and high throughput. Firstly, the basic principles, technical proposals, performance simulations and potential research directions of NOMA technology are described. The development and principles of NOMA technology are outlined, and the advantages of NOMA over Orthogonal Multiple Access (OMA) technology are compared. Next, the potential research directions of NOMA in the future are presented, including the combination with Multiple-Input Multiple-Output (MIMO), the combination with cognitive radio, and fullduplex NOMA.

Achieving the rapid and accurate detection of winter jujube in complex scenes is essential for the precise picking of the robotic arm. Intelligent object detection methods have been applied in a variety of fruit detection fields. However, intelligent winter jujube detection remains a challenging task due to its small shape, similar to leaf color and position overlap. To address these issues, an improved YOLOv4 architecture was proposed. First, we chose MobileNetv3 framework as the backbone feature extraction network of YOLOv4 to extract features, then, Feature Pyramid Network (FPN) was utilized to maximize the use of features. Finally, the predicted result was outputted by Yolo Head. To verify the effectiveness of the proposed method, we compared the state-of-the-art object detection network including SSD, Faster R-CNN. Comparative experimental results showed that our model outperforms other methods with higher accuracy and can meet the requirement of real-time detection. Moreover, our research also provided technical references for the detection of other small target fruits.

In the process of telemetry data distribution from the satellite to each telemetry station, due to the influence of weather, magnetic field interference and other accidental factors, the observed data will seriously deviate from the target true value. We call this part of data outliers. The accuracy of telemetry data is very important. If there are outliers in telemetry data, it will directly affect the subsequent fault detection and prediction, produce abnormal misjudgment, and even cause false alarm, which will perplex the subsequent staff to confirm the satellite state and deal with satellite abnormalities. In this paper, an improved arithmetic of 53h is proposed to eliminate outliers from the telemetry data of a navigation satellite. The elimination rate of isolated outliers can reach 100%, and it can avoid the wrong elimination of correct data by continuous outliers. The model is simple and fast, and can be well applied to non real-time and real-time telemetry data preprocessing.

Resource scheduling technology as an important means to achieve the optimal allocation of SAGIN resources, plays a vital role in Earth observation, emergency communications and other fields, through the in-depth understanding and analysis of the research status of the integrated network resource scheduling field, the current resource scheduling technology is not yet perfect, it is difficult to meet the multi-satellite collaboration, network topology real-time changes, massive users as the characteristics of the scheduling needs. Starting from the research point of multi-dimensional resources, this paper considers the synergy between resources, proposes a multi-dimensional resource vector model, and on this basis proposes a multi-dimensional resource collaborative optimization scheduling algorithm to achieve unified scheduling from the global perspective. In the simulation experiment, it is proved that the indicators such as the number of task completions, total tasks revenues, average revenue and scheduling decision-making time can be effectively improved, among which the number of task completions is increased by 10%-30%, the total tasks revenue is increased by 30%-60%, and the scheduling decision-making time is saved by about 30%.

There are many types of operations in electric power field, which require high safety skills and safety awareness of electric power workers. In order to find out the weak links of power operation safety and improve the comprehensive quality of workers, this paper proposes a safety portrait model of power workers based on global vectors for word representation (GloVe) and bi-directional gate recurrent unit (BiGRU). Firstly, the proposed model adopts the Delphi method to set multidimensional labels for power workers and forms a text data set of power workers. Then, GloVe is used to construct word vector for the text data set of power workers. Finally, the constructed word vector is input into BiGRU network for training, and the labels of power workers to be analyzed are output. The results of the proposed model can help electric power corporations improve the comprehensive ability of workers in a targeted manner, and provide reference and guidance for personnel assessment, work order management, production control and other needs.

For the expansion problem of quasi-cylinder labels with unequal radii, this paper proposes a cylindrical-like label surface expansion algorithm based on structured light vision. First, a structured light vision system is built by using an area scan camera and a cross laser. Second, the cross-laser point cloud of the quasi-cylinder label is obtained, and all the label point clouds are reconstructed according to the obtained laser point cloud. Finally, using the camera imaging model and the acquired images, the surface unfolding of the quasi-cylinder label is realized. The experimental results show that the algorithm takes only 150ms to realize the surface expansion of the quasi-cylinder label. And this algorithm does not require complex pose optimization, so the efficiency of the algorithm is further improved.

It is more and more important to study the characteristics of underwater scattering field. Underwater targets can be simplified into specific shapes. This paper takes the hemispherical and cylindrical combination model as the research object to analyze the echo elements of different azimuth signals, so as to provide theoretical support for underwater nearfield confrontation.

At present, spinal disease diagnosis mainly relies on manual inspection, which has much problems such as low recognition efficiency and human resources. This paper proposes a spinal disease recognition based on YOLOv5 algorithm analysis target detection model, can help doctors to identify patients more fast and efficient type of illness. While traditional YOLOv5 algorithm adopted by the box testing cannot effectively spinal lesion recognition, so the selection based on the data of the key position, which forecasts task cleverly into the key tasks. And the Head end of YOLOv5 algorithm, loss function and the proportion of the NMS, further improvement makes the model can accurately identify patients with spine illness situation and its type. The experimental results show that YOLOv5 algorithm based on key points regression in spinal MRI image validation test dataset, it concluded that model can correctly identify the disease location and disease types of the average accuracy of 52%, and the key position on the patients correctly identify whether the average accuracy of 87.8%.

In order to realize the fast and accurate detection of Albas cashmere goat in natural scene, a recognition method of Albas cashmere goat in natural scene based on improved ylov4 is proposed. Attention mechanism can enhance the feature extraction ability of the network and enhance the performance of the model. After analyzing the structure of yolov4 special network, attention mechanism is added to the three positions of yolov4 model to further improve the feature extraction ability of the network. For the improved model training, P, R, F1 and mAP are selected as model performance indicators. Experiments on the data set show that the accuracy rate of the improved yolov4 model is 85.83%, the recall rate is 85.92%, F1 is 85.87%, and mAP is 82.32%. Among them, the recall rate of the improved yolov4 model is 1.5% higher than that of yolov3, the average accuracy is 1.53% higher than that of the original yolov4 network, 3.31% higher than that of yolov3 network and 5.2% higher than that of yolov3-cbam. To some extent, this model can realize the rapid and accurate detection of Albas cashmere goat, which provides technical support for the development of intelligent animal husbandry.

In this paper, a method for grading Parkinson's disease patients based on the analysis of the complexity of electrostatic gait signals is proposed. The non-contact electrostatic sensing device is used to obtain the electrostatic gait signals generated by the natural stepping process of Parkinson's patients and healthy people, and extract their features respectively. The LZ complexity, C0 complexity, multi-scale entropy (MSE) and permutation entropy (PE) were obtained, and the differences in gait characteristics of the two groups were compared and analyzed. Then, the support vector machine (SVM) algorithm was used to establish the Parkinson's disease grading model. The experimental results show that several complex features proposed in this paper can effectively distinguish Parkinson's patients with different severity levels, and the classification accuracy rate is 91.77%.

The development of mobile communication technology promotes the intercommunication of heterogeneous networks, and the identity authentication of wireless devices is essential to protect communication security of heterogeneous networks. The uniqueness and non-forgery of radio frequency fingerprints (RFFs) originate from the hardware tolerance of wireless devices used by physical layer-based authentication schemes to realize authentication between devices during communication. At this stage, demands for long-distance communication between devices are rising, such as walkie-talkies and drones. In addition, due to the complex and changeable characteristics of the wireless communication environment, noises easily lead to problems of signal feature offset and feature ambiguity, becoming essential factors that affect the accuracy results of device authentication. This paper takes data enhancement as a starting point to study individual identity authentication of wireless devices in the case of low signal to noise ratios (SNRs). Based on adversarial autoencoder (AAE), we propose our low-SNRs signal recognition generative adversarial network (LSRGAN), which reconstructs disturbed and poor-quality wireless signals, and improves recognition performances of signals with low SNRs. Experimental results show that when the SNR is -5dB, the average recognition rate of devices can still reach 95.00%.

The main objective of this thesis is to make the traditional SHA-256 algorithm computationally efficient and secure. By improving the iterative structure of the SHA-256 algorithm, we change the output of the algorithm from eight 32-bit groupings to four 64-bit groupings, so that the 16 iterations per round of the main loop of the algorithm become eight iterations, thus reducing a total of 32 iterations in four rounds, which in turn improves the computational efficiency of the algorithm, and further improves the avalanche property of the LHF by constructing a logistic chaos The avalanche property of the LHF is further improved by constructing logistic chaos mappings with variable parameters during the message computation. Finally, it is proved through experiments that the probability of bad points in LHF is reduced by 1% compared with MD5, and the probability of bad points in LHF is reduced by 0.3% compared with SHA-256, so it has excellent avalanche and diffusion effect, and also in the performance analysis, the computation efficiency of LHF is accelerated by 53.13% compared with SHA-256 and 14.77% compared with MD5, so it achieves the purpose of speeding up the computational efficiency.

Most schools rely on human observation for student learning evaluation and teachers to observe and evaluate students' classroom behaviors, giving each student a usual grade, which is inaccurate and too subjective, and there is no public dataset of students' classroom behaviors yet, so we constructed a classroom teaching video library and a library of students' classroom behaviors. First, the classroom monitoring video is extracted regularly in combination with the cv2 library to get a real-time picture stream of each student; then the spatiotemporal features of each student's behavior are learned by using convolutional neural networks, so as to achieve real-time recognition of classroom behavior in classroom teaching scenarios oriented to multi-student targets. In addition, an intelligent teaching assessment model is constructed and an intelligent teaching assessment system based on student classroom behavior recognition is designed and implemented to help improve teaching quality in order to realize smart education. Through experimental comparison and analysis on classroom teaching video dataset, it is verified that the proposed real-time multi-person student classroom behavior recognition model in classroom teaching video can achieve 70% accuracy rate, and the constructed intelligent teaching assessment system based on classroom behavior recognition has also achieved better operation results on classroom teaching video dataset.

Nowadays, as the electromagnetic environment becomes more complex, there is a greater chance of coupling effects between devices on the same platform. This is undoubtedly an obvious negative impact on equipment performance, and how to reduce the impact has become the focus of research. Therefore, based on the actual case of a combat vehicle model research and the coupling effect of related theory, the model of vehicle antenna systems under coupling effect is discussed in this paper. The coupling coefficient is less than - 50 db as an index. Preliminary antenna system is worked out the optimized layout, providing certain reference for subsequent related work.

The advent of 5G communication era provides new opportunities for the development of major operators and also brings new challenges. 5G communication needs more and more tower resources to support the large-scale construction, but the tower companies themselves invest in the construction of towers face many difficulties such as limited land resources, long construction cycle, and financial pressure. The antennas of communication base stations are mounted on electric towers, which can reduce the need for new communication towers to achieve wider coverage of communication signals. Three types of shared business models are proposed for power and communication tower sharing: direct leasing, indirect leasing and joint venture construction. In order to compare the economic differences between shared towers and unshared towers for all parties involved, an analysis of the revenue obtained by grid companies, tower companies and third-party companies in the different business models is carried out by combining the differential net present value method.