Abstract:Gunshot recognition technology can quickly and accurately provide battlefield information in a military environment, but most gunshot recognition systems are currently deployed on the server side, which is not practical and feasible. To solve this problem, this paper designs a ZYNQbased technology gun sound recognition system. The system takes ZYNQ7020 chip as the core, and makes full use of the ZYNQ chip that integrates ARM and FPGA. First, a multi-channel data transmission link and a sound field feature parameter extraction module are designed in the FPGA part of the chip. Secondly, the ARM part of the chip is deployed after the lightweight network model trained on the PC side processes the feature parameters extracted by the FPGA to identify the types of gunshots. Finally, the three gunshots in the gunshot dataset NIJ Grant 2016-DN-BX-0183 are used. Test in the field. The test results show that the system can accurately classify gunshots, and the average recognition rate of gunshots reaches 91.67%. This achievement has strong application value in the field of gunshot recognition.

Abstract:Static random access memory (SRAM) chips in avionics equipment are prone to single event upset due to high-energy particle radiation in the environment, resulting in the loss of key stored data and seriously affecting aircraft safety. The current system level reinforcement technology has the problems of limited error correction capability and poor practicability. In order to enhance the error correction ability and improve the practicability, this paper proposes a reinforcement met-hod of constructing the optimal solution cyclic shift interleaver combined with (21,16) Hamming code to correct the continuous 4 bit or less inversion of multiple error patterns, and build a fault injection platform using single frame reconstruction technology to replace the particle irradiation experiment, so as to evaluate the effectiveness of reinforcement design losslessly and efficiently. The experimental results show that the reinforcement rate of (21,16) Hamming code combined with the optimal cyclic interleaver against single event adjacent multi bit rollover is increased by 48.54% on average, which enhances the performance of SRAM memory cells against single event rollover and ensures the safety of airborne electronic systems.

Abstract:As a national legal metrology institution, the geometric quantity project team has designed and manufactured a new comprehensive construction quality calibration device to ensure the quality of construction projects. The device achieve multi-dimensional, multi-parameter calibration though the flatness and straightness main calibration structure, the verticality main calibration structure, the thickness main calibration structure, the angle main calibration structure and their auxiliary design. It almost covers all the calibration of current construction quality basic measurement instruments in the field of geometric quantity, meets the 1/3 main measuring instruments accuracy requirements, and realizes reliable traceability and dissemination of value of quantity. It is an innovative, comprehensive and practical high-precision measurement standard device.

Abstract:In order to get the location information of pedestrians in buildings, an indoor personnel location system based on WiFi sniffing is designed. The WiFi fingerprint method is used to realize the positioning function, and the channel characteristics are introduced into the fingerprint method to analyze the influence of different channels on the signal strength. The main control chip of this design selects the FPGA chip of the domestic Anlu company, and uses the FPGA chip to match the fingerprint library. At the same time, relying on the rich expansion interface of FPGA, this design has developed a LAN website and a display interface based on the serial port screen for human-computer interaction. Taking the north area of the third floor of Wende building of NUIST as the experimental site, in the 2 m wide corridor, the positioning accuracy of the system for 14 rooms can reach 90%, and the response time is less than 6 s. All devices in this design are made in China with good popularization and wide application range.

Abstract:In order to improve the accuracy of pipe percussion detection in the wall, this paper uses the fine composite multi-scale dispersion entropy to detect the frequency and amplitude changes of the percussion sound signal, and extract the multi-scale pipe features in the signal. The multi-dimensional pipeline feature matrix was input into the support vector machine, and the sparrow search algorithm was used to determine the optimal value of the parameters of the support vector machine. The classification of buried pipelines in the wall was completed through model training, and a knocking detection method of pipelines in the wall based on fine composite multi-scale dispersion entropy was proposed. Comparing this method with other signal processing methods, the results show that the detection accuracy of the proposed method is up to 97%, which is much higher than the other two methods.

Abstract:Target tracking technology has good application prospect at present, but in an embedded processing platform is facing complicated high real-time requirements, tracking, etc., and restricted by cost and embedded processing platform to calculate force, it is often difficult to meet the demand of reality of its processing effect, so the image processing technology such as target tracking ground implementation is a hotspot of current research content. To solve this problem, this paper implemented an improved Mean Shift target tracking algorithm on FPGA platform. The algorithm first searched for the target by the gradient climbing of the probability density distribution of the target, and then used the prediction mechanism of Kalman filter to predict the location of the next frame search calculation, so as to reduce the number of iterations of Mean Shift. The algorithm makes full use of the parallel and pipelined-processing characteristics of FPGA to realize the real-time target tracking in 1 920×1 080@60 Hz hd video image scene, and the Kalman filtering algorithm enables it to have a certain ability to resist occlusion interference in more complex scenes.

Abstract:In order to more accurately evaluate the sharpness of the optical system and improve the qualification rate of camera products, this paper designs a camera sharpness detection system based on the Android platform. During the active alignment process of the camera assembly, the image sharpness of the camera is evaluated by measuring the Modulation Transfer Function. Unlike the existing inspection instruments on the market, this inspection system is small, strong universality, easy to operate, low cost and can record the Modulation Transfer Function of camera modules in different states in real time for the purpose of inspecting the image quality of camera modules. The system is more likely to be commercially and industrially viable in the field of camera module detection.

Abstract:At present, glass surface defect detection is mainly manual, which takes a long time and has low accuracy. An improved algorithm model YOLO-M combining YOLOv4 and mobilenetv3 is proposed to solve this problem. Firstly, the mobilenetv3 network is used to replace the original backbone network cspdarknet53 of YOLOv4, and the activation function is modified to improve the running speed on the basis of reducing the model size and parameters. Then, the glass defect samples were photographed and sampled. The defects were divided into wear, bubble and scratch, and the glass defect data set was established. Finally, the glass defect data set is trained by using YOLO-M, YOLOv4 and YOLOv4 tiny algorithms, and the evaluation indexes such as average precision and frame rate under different algorithms are compared. The experimental results show that the frame rate of YOLO-M algorithm in glass defect detection is 57.72 f/s, and the average accuracy is 91.95%. YOLO-M algorithm has obvious effect on the speed and accuracy of glass defect recognition, and can be used as an important reference for subsequent sorting research and other glass product defect recognition.

Abstract:In the Industrial Internet of Things, the massive data generated by the SCADA of wind turbines is not suitable for being directly sent to the cloud for processing due to real-time requirements. This paper designs and build a set of Micro-Wind Turbine equipment condition monitoring system based on Edge Intelligence.Three unsupervised anomaly detection algorithms, including OC-SVM, IForest and HBOS, are analyzed and compared with each other. The experimental results show that OC-SVM have the best real-time anomaly detection effect. The F1 scores in the rotation anomaly test set and vibration anomaly test set are 0.997 and 0.969，respectively. This paper can provide some reference value for the landing verification of edge side training and reasoning scheme.

Abstract:A permanent magnet synchronous linear motor (PMLSM) fixed-time control strategy based on state constraints is proposed for the speed and current double closed-loop control system of permanent magnet synchronous linear motor (PMLSM).The velocity tracking error of the system is constrained by an asymmetric obstacle Lyapunov function,and the fixed-time filter is designed without complex switching terms to overcome the′ differential explosion ′problem in traditional backstepping control.A fixed time disturbance observer is constructed to observe the load disturbance of the motor,and the disturbance is compensated to the control system to enhance robustness.Theoretical analysis proves that the system converges boundedly in a fixed time,and the velocity error can be constrained in a reasonable interval. Based on Matlab simulation experiment,when a sudden load is added at a given speed of 0.5 m/s,the speed tracking accuracy exceeds 98%,which is 2% higher than that of the fixed-time control strategy.The tracking range deviation of the disturbance observer is less than 1%,which has high observation accuracy.The simulation and experimental results verify the effectiveness of the control strategy in this paper.

Abstract:In order to solve the problem of cleaning and testing the glass curtain wall of high-rise buildings, a multi legged vacuum adsorption wall climbing robot is explored. In order to realize the smooth operation of the wall climbing robot, the main thing is to improve the control accuracy of the motor speed. The fuzzy PID algorithm is used for control, and on this basis, the particle swarm algorithm is used for further iterative optimization. Through the iterative optimization ability of particle swarm optimization algorithm, the scale factors of three PID parameters in fuzzy PID control are determined in real time, and a control system with good performance is obtained. The results show that when the scale factors CKP, CKI and CKD are 0.1, 8.23 and 0 respectively, the stability of the system reaches the optimum. Compared with fuzzy PID control, the response speed of the system further optimized by particle swarm optimization algorithm is improved by 0.024 s, and there is also no overshoot, which can meet the requirements of the system.

Abstract:In view of the output uncertainty and intermittent problems of distributed power generation equipment in microgrid, and the shortcomings of traditional deep deterministic policy gradient algorithm, such as slow convergence speed, poor robustness, and easy to fall into local optimum. In this paper, a deep deterministic policy gradient algorithm based on prioritized experience replay is proposed, aiming at the lowest operating cost of the microgrid system, to realize the energy optimal scheduling of the microgrid. First, the Markov decision process is used to model the microgrid optimization problem; secondly, the prioritized experience replay pool with Sumtree structure is used to improve the efficiency of sample utilization, and importance sampling is applied to improve the influence of state distribution on the convergence results. Finally, this paper uses real power data for simulation verification. The results show that the proposed optimal scheduling algorithm can effectively learn the operation strategy that minimizes the economic cost of the microgrid system. At the same time, the introduction of prioritized experience replay and importance sampling improves the performance of the algorithm.

Abstract:This paper proposes an indoor fingerprint localization algorithm based on KNN and XGBoost algorithm to address the problems that the localization accuracy of KNN algorithm needs to be improved and the stability of localization is poor.The algorithm first divides the sample set into a training set and a test set, RSSI data of AP in training set was used as features and coordinates were used as labels, and XGBoost algorithm was used for modeling. Secondly, the KNN model is integrated, the nearest neighbor set found by KNN algorithm is introduced into XGBoost model, and combined with the prediction results of individual XGBoost algorithm to achieve coordinate positioning.Finally, the effects of the algorithm′s K-value, number of regression trees, decision tree depth and learning rate on the error are investigated in a practical setting to determine the relevant parameters of the algorithm.The experimental results show that the average localization error of the proposed algorithm is 1.55 m, which is 24.76% and 11.93% less than that of the KNN algorithm and XGBoost algorithm, respectively, and the cumulative distribution function curve converges faster and has better localization performance.

Abstract:For chamber pressure tester for environment factor of the calibration data processing using the least square method for linear fitting, ignore the influence of abnormal value problem, the author uses the minimum distance square method and the weighted least squares method respectively to fit the calibration data of the two methods, from F significance test and sample analysis of simulation results of determination coefficient two aspects, and with the least squares method, To find a convenient method for fitting the data collected in the harsh environment of high temperature, high pressure and high impact to replace the least square method, so as to obtain the sensitivity coefficient with high confidence, which can provide data basis for the design of weapon barrel. The simulation results show that the minimum distance sum of squares method has the best fitting effect, but its accuracy is very small compared with the least square method. Considering the least square method is more suitable for linear fitting of the calibration data of chamber pressure because of its simple use.

Abstract:Surface Electromyography (sEMG) signal is a kind of weak physiological signal that effectively represent muscle activities; however, it is susceptible to many noise interferences in the acquisition process. In order to adaptively set key parameters of Variational Mode Decomposition (VMD) and further eliminate the noises in the sEMG signal, a sEMG signal denoising method based on Improved Sparrow Search Algorithm (ISSA) optimized VMD and second-generation wavelets threshold is proposed in this paper. Firstly, The VMD parameters setting was optimized by adopting ISSA based on improved tent chaotic mapping, adaptive weight and dynamic change of the population number of sparrows, and quality factors were used as objective function. The optimized VMD was used to decompose the pre-treated sEMG signal, and the signal and noise components were distinguished by the spectrum correlation analysis. Finally, the signal component was denoised by the second-generation wavelet threshold to obtain the denoising signal. The results are shown that: ISSA can effectively improve parameter optimization ability for VMD compared with SSA, the denoising method for sEMG signal based on ISSA-VMD and second-generation wavelet hard threshold has better denoising performance than other methods under different noise levels. For actual sEMG signals, the method based on ISSA-VMD and the second-generation wavelet hard threshold can effectively remove noise.

Abstract:Since the popularity of UAVs has brought about many privacy and security problems, the solution should comply with dynamic and effective detection of UAV signals, so as to achieve effective control. This paper proposes a software radio intrusion detection method for UAV with hardware implementation. It uses adaptive denoising and cubic clustering to recognize and classify UAV signals. Simulation results show that the detection probability can reach 100% when the SNR is above -16.2 dB. Based on the software radio platform AD9361+FPGA+STM32 as the core, it carries out the relevant engineering implementation. The measured results show that it is practical and effective, which can effectively identify UAV signals and types in both indoor and outdoor complex environments with strong application prospect.

Abstract:In the communication process of various types of instrumentation bus networks, the system often needs to process the received high-speed data stream in real time on the application layer. And how to extract data frame data from continuous data stream is the main problem discussed. In this regard, analyzes the protocol processing methods of common instrument buses, and designs a set of frame extraction algorithms, including frame extraction state machine, improved Sunday frame header matching algorithm and intra-frame subdomain search algorithm. Then this paper tests the algorithm in two environments: direct sending and sending via TCP network. Experiments show that the performance of the algorithm is better than the frame length domain decoding under the Netty framework. Finally, in order to actually test and apply the algorithm, this paper uses the algorithm to extract and store the data frame in real time for the 64-channel, 100 kS/s analog acquisition card, and display the waveform of the acquired analog quantity. This algorithm can be used for data separation, frame header identification and frame data extraction at the application layer of the instrument bus.

Abstract:Ensuring the safe operation of power cables is the basis of building a new intelligent power system. In order to realize the digital early warning of external force damage events, an online identification method of external force damage vibration signals based on VMD-WOA-ELM is proposed. Firstly, the collected abnormal vibration signal is decomposed into several intrinsic modulus function components (IMF) by VMD, then the time and frequency domain eigenvalues of each IMF component are extracted to form the eigenvector, and finally the extreme learning machine (ELM) is used to identify the type of vibration signal. In order to solve the problem of poor classification stability caused by the random selection of initial weights and thresholds of ELM model, whale optimization algorithm (WOA) is used to optimize the parameters of ELM to obtain the optimal classification model. This method is applied to the identification experiment of construction vibration signal type. The vibration signals of four typical breaking events were collected, and each signal has 100 groups. 80% of them were used as the training set and 20% as the test set to test the recognition performance of the algorithm. The algorithm is compared with traditional ELM, PSO-ELM and GA-ELM. The results show that under the same computer operating conditions, the classification accuracy of WOA-ELM is 98.75%, which is 5% higher than that of traditional ELM, and the overall running time is only 4.10 s. Compared with the other two algorithms, this algorithm has the highest recognition accuracy, the fastest convergence speed and the best comprehensive performance.

Abstract:Ultrasonic guided waves use highfrequency signals in the detection of small cracks in rails, which will lead to significant signal attenuation and reduced detection sensitivity for small cracks. It is difficult to identify small cracks. In order to improve the recognition of defect signals, this paper uses barker code as the encoding method, and use BPSK-like as a decoding method, which is used in the signal processing of guided wave detection of small cracks in rails. The artificial crack with a depth of 6 mm and a width of 0.5 mm on the rail bottom will be verified through experiments. In order to measure the effect of the algorithm, the original transceiver guided wave signal without any signal processing and the guided wave signal processed by the barker code-matched filter codec are used respectively. Compare with the experimental method. The results show that the guided wave signal processed by the barker code-like BPSK codec has an obvious enhancement effect on the identification of small cracks at the bottom of the rail, and the effect is better than the other two methods, which can be used for the detection of small cracks in the guided rail in the future. provide support.

Abstract:As an important component of mechanical system, rolling bearing is prone to failure due to harsh working environment. The vibration signal of faulty bearing includes transient impact components, harmonic components, background noise and other components. In order to extract fault features accurately, based on sparse representation theory, a bearing fault diagnosis method based on Laplace wavelet dictionary is proposed. First, a number of vibration signal fragments are intercepted, and the correlation filtering method is used to find the signal fragment with the largest correlation coefficient, and the basis function is determined accordingly, and Laplace wavelet atoms are constructed and expanded into a sparse dictionary. Then, the OMP algorithm is used to complete the sparse reconstruction of the signal under the dictionary to achieve noise reduction. Finally, the envelope analysis is performed on the noise reduction signal to extract the fault features and determine the fault type. Both simulation and experiment verify the effectiveness and feasibility of the proposed method, and it has certain application value.

Abstract:Aiming at the issue of low accuracy of existing visual defect detection algorithms for magnetic tile appearance recognition, an algorithm based on S-ASPP and dual attention mechanism is proposed. Firstly, in order to solve the problems of low detection efficiency and high hardware cost for algorithm deployment caused by large amount of model parameters and complex network structure, a lightweight GhostNet backbone network is proposed to extract low-level visual features. Secondly, the S-ASPP module is designed to introduce dense connections and depthwise separable convolutions, so as to reduce the amount of model parameters and improve the prediction speed of the model. Next, in order to solve the problem that semantic information may be lost during feature extraction, a dual-attention module is designed. The middle-level features of GhostNet are feed into the dual-attention module and concatinate with the high-level features after ASPP processing, so that the network can acquire more semantic features and improve the segmentation accuracy. Finally, in order to verify the effectiveness of the proposed method, the magnetic tile defect detection algorithm is compared with DeepLabV3+, PSPNet and U-NET algorithms on the magnetic tile data set. Experimental results show that the magnetic tile defect detection algorithm based on separable ASPP and dual attention mechanism has high recognition accuracy, and the average crossover ratio reaches 82.43%. The average pixel accuracy of the category reached 93.08%. The algorithm balances the relationship between the recognition accuracy and the number of parameters and has the best comprehensive performance.

Abstract:Aiming at the problems of low efficiency and high error detection rate in the traditional manual sampling measurement of automobile pipe flange, a measurement system of automobile pipe flange size based on machine vision was developed to improve the detection efficiency and measurement accuracy. Use software HALCON of camera calibration, image using median filter to test to deal with the noise, then method of gray-scale image enhancement, complete the image preprocessing, image using Canny operator after the subpixel edge detection, using the least squares fitting to the extraction of subpixel edge information get continuous aperture outline, the flange end face size was obtained, and the measurement results were displayed through HALCON and C# joint programming. Finally, the measurement results were compared with those of image measuring instrument. The experimental results show that the measurement accuracy of the system can reach 0.08 mm, which meets the requirements of measurement accuracy. Flange detection time is 1.2 s, compared with the traditional manual detection, detection efficiency is increased by 70%, can be applied to the actual production site.

Abstract:As the basic problem of image enhancement, image defogging has received extensive attention. It has become a challenging area of research. For the problems of color distortion, fog residue of defogging images in prior method and deep learning method, this paper proposed an image defogging algorithm based on attention mechanism for detail recovery. Firstly, the improved CBAM (convolutional block attention module) module is introduced to design the attention basic block and encapsulate the basic block into blocks. Secondly, to strengthen the information interaction ability within the block, dense connection residual blocks are introduced between the blocks. Finally, the detail recovery module is designed to recover the detail of the fog image to further reduce the impact of fog residue. Numerical simulation experiments show that the proposed algorithm achieves higher peak signal-to-noise ratio and structural similarity compared with the mainstream defogging algorithm on the RESIDE (realistic single image defogging) data set and better visual effects on real images on the RESIDE (realistic single image defogging) data set.

Abstract:Aiming at the problems of many seals, shallow impressions and low accuracy of seal identification, a modified YOLOv5 seal identification algorithm is proposed. The algorithm improvement is divided into two aspects. First, CBAM attention module is introduced to improve the feature extraction ability of the model. Secondly, and EIoU Loss is introduced to replace the CIOU Loss boundary box regression loss function in the algorithm, which effectively solves the aspect ratio described as the relative value, which is a certain fuzzy problem. Experiments show that the improved algorithm′s seal recognition F1 score has reached 0.95, which is a 2% improvement compared to the original algorithm. Finally, to use the seal to verify the effectiveness of the model, the improved YOLOv5 model is called in the digital archive processing system, and the results show that the improved algorithm can run stably in the system.

Abstract:In the application of Intelligent Transportation, target detection algorithm based on deep learning has the problems of large number of model parameters, slow calculation speed and low accuracy of simple network for vehicle detection. This paper presents an efficient lightweight vehicle detection model, which is improved by using YOLOv4 network as a reference model. First, this paper uses CSPMobileViT network to replace the original backbone network, then replaces PANet with BiFPN, replaces 3×3 standard convolution in BiFPN with deep detachable convolution, and finally adds ECA module before BiFPN and YOLO-Head. In the loss function section, the Border Regression Loss CIoU is improved to Focal EIoU to solve the problem of difficult sample imbalance. The experimental results show that the mAP value of the improved network is 96.77%, the detection speed reaches 0.023 4 s per picture, the model size is only 32.76 MB, and the parameter amount is 8 587 541. Compared with the original algorithm, the mAP is improved by 1.54%, while the model size and number of parameters are only about 1/8 of the original model, and the FPS is improved by 7.5, so the improved algorithm has better detection effect.

Abstract:The tooth flank region that reflects the tooth flank shape information in the interferogram is surrounded by background noise, and its processing accuracy has a great impact on the subsequent processing steps such as phase unwrapping and the interferogram registration, which is directly related to the accuracy of the final measurement result. Aiming at the phenomenon that there are spurious fringes in the boundary of tooth flank interferogram, a tooth flank region segmentation method combining Butterworth filter and Haar wavelet is proposed by analyzing the distribution of tooth flank interferogram fringes. Finally, under the experimental conditions of this paper, it shows that the proposed method can exclude and correct the wrong phase values of the unwrapping in the boundary spurious fringe regions of the tooth flank interferogram from the valid measuring region by comparing with the time-domain segmentation method, which proves that the method based on the frequency domain can effectively improve the accuracy of the segmentation of the tooth flank region of the interferogram.