Abstract:Detector readout and display and analysis of data are key issues to host computer software. Aiming at features and software demands in deep-sea in-situ nuclear radiation detection, this paper designs a detector software which integrates detector control and readout and display and analysis of data. A protocol based on MODBUS is designed, the work pattern of detector and a unified hardware interface are defined, and drivers for RS-485 and Thunderbolt are implemented in the software for detector readout. User interface is designed by XAML for data display. An algorithm based on wavelet transform and expectation-maximization algorithms for peak finding, fitting, and nuclide identification is designed. By connecting to detector to measure and analyze spectrum of balanced 238U-226Ra source and known radioactive samples from deep sea, the functions of the software are tested. The results shows that the software can stably realize detector configuration and readout, as well as data display and analysis, and has a better analysis result than InterSpec, a spectrum analysis software.

Abstract:The off-grid switching strategy of solar-storage microgrid is an important support to ensure the smooth operation of the grid. Aiming at the problems such as deviation of voltage amplitude, phase and frequency between the microgrid and the large power grid, based on the V/f control of the energy storage inverter, the large power grid voltage is used as the reference value of the controller to simply the structure of the pre-synchronization control link. An improved pre-synchronization method is proposed. Based on the negative sequence current change from off-grid to grid-connected, the islanding detection method is explored to determine the off-grid signal. Considering the output characteristics of photovoltaic power supply and off-grid switching operation, a control mode of two energy storage units is proposed. The switching sequence of V/F and P/Q control modes of different energy storage units is determined. According to the time-sharing conversion, the impact of system switching is reduced. Off-grid switching test, parallel-off-grid switching test, three groups of optical storage and output power grid-connected strategy simulation comparison tests are designed. The results show that off-grid voltage deviations of amplitude, frequency and phase angle are less than 1%, ± 0.1 Hz and 5 ° respectively with improved pre-synchronization control, and the structure is more simple. Based on the switching strategy of the two energy storage units, the switching time of the off/grid mode is shortened and the impact is reduced. The result verifies the effectiveness of the proposed improved pre-synchronization method and the switching strategy based on the two energy storage units.

Abstract:In this paper, a sparse autoencoder (SAE) network is proposed to extract the effective features of vibration signals and apply them to support vector machine (SVM) to detect footstep vibration events, aiming at the problems of low accuracy and manual parameter selection of traditional short-time energy detection method. To alleviate the signal distortion caused by the dispersion effect of vibration signal, the wavelet decomposition method is used, and the decomposition parameters are optimized based on the experimental analysis, and then the location is solved based on the generalized cross-correlation (GCC) and time difference of arrival (TDoA) algorithm. Experimental results show that, compared with manual feature screening, the detection accuracy of an active segment can reach 96.8% by the SAE-SVM algorithm, and the average positioning error of the system is 0.82 m.

Abstract:To solve the problem of high cost and time consuming when the key parameters of micromechanical structure must be encapsulated by wafer flow, an electrostatic driven micromechanical vibration gyro was designed in ConventorMEMS+ and Matlab/Simulink environment, and the model was simulated and tested at the system level. Through modal analysis, DC analysis, AC analysis and other methods to optimize the model structure, combined with virtual body silicon etching process extraction to establish a micromechanical gyro transfer function model, system level simulation analysis in Simulink, obtained the transient response of the system. The key parameters, such as resonant frequency, safe operating voltage and optimal driving voltage frequency, are determined under the condition of open loop, and the relationship between resonant frequency and the length and width of beam is obtained. Under the same working voltage and other conditions, the difference between the resonant frequency and the simulation results is only about 3.8% and 0.4%, indicating that the system-level simulation results can provide theoretical and experimental basis for the subsequent design of the closed-loop measurement and control circuit of the micro-mechanical gyro.

Abstract:The technology of measuring vital signs by millimeter wave radar has great medical value. However, as a kind of noise signal, respiratory harmonics and human random movement signal seriously affect the extraction of heart rate. In order to solve the above problems, this paper proposes a method of measuring dynamic target heart rate based on multi-detection signal separation technology and adaptive noise cancellation algorithm according to the characteristics of different breathing modes. The method comprises the following steps of: simultaneously measuring chest and abdominal micro-motion signals of a person to be measured by using a 77 GHz frequency modulation millimeter wave radar; separating the chest and abdominal baseband signals by using a newly proposed multi-detection point signal separation technology based on the chest and the abdomen; eliminating noise signals by using an adaptive noise cancellation algorithm; and performing frequency spectrum analysis on heartbeat signals to obtain a heartbeat frequency. Experiments show that the method can effectively eliminate the noise signal interference in the state of human random movement, and the error rate of heart rate measurement is only 1.19% in multiple measurement experiments of a single target, which is 0.97% lower than that of multi-channel Kalman smoother method.

Abstract:In order to quickly identify the leakage fault of heating pipeline, a method of pipeline leakage diagnosis using convolutional neural network (CNN) to identify the pressure data was proposed for the negative pressure wave characteristics of pipeline leakage.By setting up the experimental platform of heating pipeline, the pressure data under normal, leakage and regulating valve conditions are collected as the training set and test set of convolutional neural network.The original data are denoised by wavelet, and the hard threshold processing method is used to effectively eliminate the noise signal. Meanwhile, the enhancement feature appears in the valve condition, which is helpful to enhance the classification ability of the convolutional neural network.The improved AlexNet convolution network model is used to learn and identify the collected data.The results show that the average recognition accuracy of CNN model is 98.3% in laboratory data testing. In the verification of the actual pipe network, the leakage data of the three thermal stations were correctly identified, indicating that the CNN model has good fault diagnosis ability.

Abstract:Aiming at the disadvantages of the terahertz mixer test method, such as single function and low-test efficiency, this paper proposes a terahertz mixer sweep frequency test method based on the sweep test method of microwave mixer. This method can increase the test band from microwave band to terahertz band above 110GHz, and realize the frequency sweep frequency test of conversion loss and return loss of the terahertz mixer. Furthermore, the“signal generator + spectrum analyzer”test method and the test method proposed in this paper are used to verify the 150 GHz-170 GHz fourth-harmonic mixer. The results show that the sweep frequency test method proposed in this paper not only achieves the rapid measurement of conversion loss and return loss of the terahertz mixer, but also agrees with the conversion loss measured by the“signal generator + spectrum analyzer” test method, which is of great significance to the test application of the vector network analyzer in the terahertz band.

Abstract:To address the shortcomings of Rapidly-Exploring Random Tree algorithm (RRT) in the path planning of mobile robots, with many turning points and a large number of redundant points, an improved RRT algorithm based on cell decomposition method is proposed. Firstly, in the initial stage of the algorithm, the cell decomposition method is used to divide the map into feasible regions and obstacle regions; later, according to the neighboring relationship between regions, the selection of random sampling points is fixed in the neighboring regions until it is extended to the region where the target point is located; and the final path searched is optimized to improve the problem of too many path turning points. Simulation results show that the path generated by the improved RRT algorithm is shorter in length and consumes less time. Finally, the improved RRT algorithm is applied to the actual mobile robot to further prove the practicality and effectiveness of the improved algorithm.

Abstract:Accurate estimation of SOC plays an important role in preventing excessive charge and discharge of lithium batteries, improving energy utilization rate of lithium batteries and ensuring safe and stable operation of battery management system. In this paper, a SOC estimation method based on multi-innovation identification theory is proposed for ternary lithium batteries. Adopting forgetting factor multi-innovation least square method for model parameter online identification by building a second order RC equivalent circuit model, multi-information unscented Kalman Filter algorithm was used to estimate the SOC of lithium batteries. Through Verified by UDDS experiment and were compared with EKF, UKF and MIUKF algorithm, the results indicate that FFMILS-MIUKF algorithm to estimate the error of the SOC control at around 1.08%, which has high accuracy and fast convergence.

Abstract:Aiming at the problem that the area of printed circuit board is small and there are many electronic device solder joints on it, which is difficult to detect effectively by traditional detection methods, a surface solder joint detection algorithm of printed circuit board based on GhostNet-YOLOv4 is proposed. First, the backbone network of YOLOv4 is modified to enhance the feature extraction capability. Secondly, adding attention mechanism makes the network pay more attention to defect features. Finally, use GhostNet instead of CSPDarknet53 as the backbone network. Compared with the traditional PCB detection algorithm, this algorithm improves the detection accuracy and speed, and can realize the accurate detection and rapid classification of common defects such as broken circuit, missing welding and short circuit on the surface of PCB. Experiments on PCB data sets show that: the improved algorithm has good practicability, the accuracy on the test set is 86.68%, FPS reached 25.43, can meet the actual detection requirements of printed circuit boards.

Abstract:Extreme learning machine auto-encoder (ELM-AE) combines extreme learning machine (ELM) technology with auto-encoder (AE), which can learn data features unsupervised and overcome the expensive time consumption of parameter iterative adjustment. However, ELM-AE, which aims to minimize reconstruction errors, cannot effectively use the data category information in classification problems, resulting in features with poor category separability. In view of this phenomenon, this paper proposes a data classification-oriented feature learning method of extreme learning machine auto-encoder with category information (CELM-AE), which limits the inter class dispersion and intra class similarity of the projected feature vector to the objective function of ELM-AE, and can obtain the optimal data representation with more class resolution through analytical algorithm. The classification experiments of 6 UCI data sets are carried out using the feature representation based on CELM-AE, ELM-AE and AE respectively. The results show that the classification accuracy and stability of the data features obtained by CELM-AE under the two classifiers (ELM/KNN) are better than ELM-AE and AE, and the time cost is very small, which shows the advantages of CELM-AE in extracting the separable feature representation of data.

Abstract:In order to improve the path planning efficiency and security based on traditional Hybrid A* algorithm, an improved Hybrid A* path planning algorithm is proposed and applied to a tractor-trailer mobile robot system. Firstly, the heuristic function is improved to reduce the computation in the path planning process, so as to improve the planning efficiency. Secondly, the obstacle penalty function is designed, which in turn realizes the avoidance of obstacles on the travel path in advance and avoids getting into local optimal solutions in U-shaped obstacles. Finally, Because the tractor-trailer model cannot be considered as a prime point, a collision detection algorithm is used to improve the reasonableness and accuracy of the planned path. The simulation results show that the improved Hybrid A* path planning algorithm can be applied to the tractor-trailer mobile robot system, and has the characteristics of high planning efficiency, high security and smooth path, which provides a theoretical basis for its path planning in practical application.

Abstract:Memristors have the advantages of nanoscale size, low power consumption and similar to neural synapses, etc., and have broad application prospects in neural computing, image classification and other fields. In this paper, a facial expression recognition method based on memristor-based convolutional neural network is proposed. First, a memristor-based ResNet convolutional neural network is constructed and the ResNet network is pruned. Then the weights of all convolutional layers and fully connected layers of the ResNet model are mapped as the memductance values of memristors in the memristive crisscross array. The experimental results show that the recognition accuracy of the memristor-based convolutional neural network model on the FER2013 dataset is 63.82%, and the recognition accuracy on the CK+ dataset is 93.95%. Compared with the original convolutional network, the accuracy loss is only 0.31% and 0.76% respectively. Finally, the influence of the non-ideal characteristics of the memristor on the accuracy is tested, which provides a reference for the actual deployment of the memristor-based neural network.

Abstract:Recently, deep learning networks for image compressive sensing have received a great deal of attention. Deep learning networks can achieve compressed sampling of images and reconstruct the original image from the sampled data. However, the existing compressive sensing algorithms cannot effectively extract original image information in image scenes with uneven information distribution, resulting in low reconstruction accuracy. To address the above problem, this paper proposes an image compressive sensing algorithm based on multi-channel sampling and attention reconstruction. The algorithm includes multiple sampling channels that can apply different sampling rates to different regions of the image according to visual saliency, so that the sampling data can contain more original image information. The reconstruction adopts the residual channel attention structure, which can adaptively adjust the channel features to improve the representation ability of the network. The comparative experiments show that the image compressive sensing algorithm based on multi-channel sampling and attention reconstruction proposed in this paper can achieve better reconstruction quality and visual perception.

Abstract:Aiming at the problem that the existing algorithms are difficult to extract fuzzy target features in the SAR image ship target detection scene, a ship target detection algorithm based on feature reuse pyramid is proposed. The proposed algorithm takes YOLOV4-tiny as the main body. First, a linear factor is introduced into the K-Means algorithm to integrate the initial anchor frame to enhance the adaptability of the network to multi-scale targets. Secondly, an attention mechanism is added to the backbone CSPDarknet53-tiny to suppress interference. information, and weaken the influence of complex background; finally, the feature reuse mechanism is used to strengthen the feature pyramid and improve the network's ability to extract fuzzy target features. The experimental results show that, compared with the YOLOV4-tiny network, the average detection accuracy of the improved algorithm on the SSDD dataset is improved by 11.79%, which proves the effectiveness of the improved algorithm in ship detection.

Abstract:In order to improve the accuracy of the traditional SSD algorithm for small target detection, an improved SSD target detection algorithm is proposed: ResNet50 based on deformable convolution is used as the feature extraction network of the SSD algorithm to improve the processing ability of the target; the feature pyramid (FPN) to fuse feature maps of different layers and enrich the semantic information of shallow feature maps; introduce channel attention mechanism during feature fusion, extract corresponding channel weights, increase the proportion of important information, and improve the detection effect. Finally, the PASCAL-VOC2007 open source data set was used for simulation experiments, and compared with the traditional SSD target detection algorithm, the accuracy is significantly improved, which verifies the effectiveness of the algorithm for small target detection.

Abstract:Aiming at the problems of low efficiency and low precision in manual measurement of honing Angle of engine cylinder wall in practical industrial testing, a measuring method based on machine vision was proposed. Firstly, Gabor optimal filter channel algorithm is used to process the sample image to obtain the enhanced linear feature pattern image. Then, DFT transformation is performed on the pattern image to obtain the Fourier spectrum image. Then, the peak lines in the spectrum image are obtained based on the digital differential analysis algorithm and the included Angle of the two peak lines is calculated as the calculation result. Meanwhile, the results were compared with the manual measurement results based on Camera Measure software. The test results show that the error of this method is only 0.33% compared with manual measurement, and the mean difference of repeated measurement is within ±1°. In terms of detection time, the average detection time of a workpiece is 0.53s. This method has the advantages of high precision and fast speed, which can effectively replace manual measurement in industrial testing.

Abstract:In order to improve the recognition accuracy and recognition speed of face recognition in embedded devices, a Ghostnet-based lightweight face recognition algorithm called Ghostfacenet is proposed. Firstly, a fixed number of intrinsic features are generated by pre-determined convolution. To address the problem of computationally intensive convolutional operations, linear operations with low computational cost are used instead of convolutional operations to generate a series of feature information associated with intrinsic features. Secondly, the Ghostfacenet-Bottleneck is designed based on the Ghost module in Ghostnet and the depthwise separable convolution. And the Ghostfacenet lightweight convolutional neural network is constructed from Ghostfacenet-Bottleneck. Finally, the Arcface loss function and the Airface loss function are combined to further increase the intra-class compactness of faces as well as inter-class differences. It also allows for better convergence and generalization capabilities of lightweight models. The experimental results show that Ghostfacenet is 11.08 times, 8.57 times, 2.75 times and 2.82 times faster than Resnet50, Efficientnet, MobilenetV2 and Mobilefacenet respectively in embedded devices. This is a significant improvement in operational efficiency without a significant reduction in recognition performance and is ideal for embedded devices with limited resources.

Abstract:Safety problems in all walks of life are particularly important. Abnormal behaviors of personnel must be detected in time and corresponding measures must be taken to effectively prevent safety accidents. Therefore, this paper proposes an abnormal behavior recognition algorithm based on the improved yolov5 network, which can ensure the safe operation of the enterprise by dealing with the abnormal behavior of personnel in video monitoring in real time. Firstly, feature processing is carried out on the input data set. In this paper, the backbone feature extraction network of yolov5 is used to extract video features, which can aggregate and form image features on different image granularity; Secondly, it is sent to the time attention block. Because the contribution values of the features at different times are different, this module is added to give different contribution values to the features; Finally, it is sent to the feature prediction network, which is built by LSTM to decode the historical feature sequence to predict the current feature. Taking playing mobile phone and smoking as examples, the accuracy of the proposed network is as high as 96.42% in the training set and 95.21% in the test set.

Abstract:In order to solve the problems of structurally complex, numerous parameters, high configuration required for training, low transmission frames of real-time detection pictures and difficult to achieve industrial application popularization of YOLOv4 target detection network, a lightweight target detection network SL-YOLO based on YOLOv4 is proposed. It improve and optimizes the original YOLOv4 network, and replaces the original backbone network of YOLOv4 with ShuffleNetv2 lightweight network, integrates SENet module into ShuffleNetv2, reduce the network computing complexity, add Swish activation function to the network layer to make the model convergence effect better; at the same time, the simplified weighted bidirectional feature pyramid structure is used to replace the feature fusion network of YOLOv4, aims to optimize the target detection accuracy; the importance of each channel was determined, thus the redundant pruning was performed, and the model was compressed. The result of a comparative experiment on open data set PASCAL VOC and MS COCO shows that the memory of the model is compressed by 89.4%, the amount of floating-point operations of the model is reduced by 88.4%, and the detection speed of the model is increased by nearly two times, which indicates the SL-YOLO lightweight network can effectively reduce the amount of model reasoning calculation and improve the model detection speed simultaneously, and greatly improve the speed of model detection.

Abstract:Aiming at the problem of large error of temperature field in thermal field detection of internal combustion engine cylinder head, combined with the difference of thermal field of fire surface at different positions, a multi-scale division method of the fire surface based on gradient variation law was proposed, which laid a foundation for the accurate detection of the thermal field distribution of the flame face. Firstly, according to the heat transfer law of the cylinder head gas side and the temperature of the thermal field, the fire surface was divided into inward-exhaust, inward-intake, exhaust-valve nose bridge area and peripheral area with the fuel injection hole as the center. Secondly, the variation characteristics of temperature gradient in each region are analyzed, and each region is divided into different scales based on different gradient variations. Finally, the number and location of temperature measurement points in different scales were determined, and the temperature of each point was measured by combining thermocouple temperature measurement technology. The experimental results show that compared with the temperature measurement values under a single scale, the temperature measurement method based on multi-scale division idea can highlight the temperature variation in the nose bridge area, which is prone to thermal fatigue damage, and provides a more reliable data source for the detection and evaluation of thermal fatigue damage in the inner wall of cylinder head.

Abstract:Non-orthogonal multiple access (NOMA ) technology, which enables multiple users to share a single time-frequency resource block, has become a research hotspot of 5G multiple access technology. In order to maximize the energy efficiency of NOMA system, power allocation in non-orthogonal multiple access technology is studied, and a new NOMA power allocation scheme based on chaotic particle swarm optimization is proposed. The energy efficiency optimization model of NOMA is established, and the energy efficiency of the NOMA system is optimized by using the chaotic particle swarm optimization to distribute the power of the system. The simulation results show that the energy efficiency of the system is maximum when the power is 36dBm, and the iteration times are less and the energy efficiency is better than traditional particle swarm optimization algorithm.

Abstract:This paper studies the moving UAV to suppress the interference by dynamically sensing and learning the interference wave direction, and adjusting the beamforming strategy in real time. Aiming at the problem that the UAV can not obtain all the actions of the jammer for strategy training in the actual scene, a method of using the cooperation within the cluster to collect the action data of the jammer to supplement the training data is proposed to improve the anti-jamming of the cluster. The beamforming decision is modeled as a Markov decision process. Based on the deep reinforcement learning architecture, a data Aided Cooperative spatial anti-jamming algorithm for UAV cluster is proposed. The simulation results show that when the auxiliary data reaches 40%, 60% and 80%, the system throughput is improved by 33%, 55% and 70% respectively. It is verified that the method proposed in this paper can effectively improve the cooperative anti-jamming ability of UAV.

Abstract:In this paper, for the purpose of transmitting digital images via light communication, a visible light communication digital image transmission system basing on FPGA is designed. First of all, in terms of the digital image encoding technique, the content of an image is converted into BMP file format at the transmitting end of the optical communication system. Then, the dataset of a digital image is transmitted to the FPGA modulation system for OOK modulating and in this procedure, transmission is based on UART protocol. After modulation, the modulated signal is power amplified before being sent out via optical wireless communication. Secondly, the received dataset of the image is demodulated non-coherently at the receiving end of the optical communication system so that it can be displayed or further processed. Furthermore, basing on the FPGA, the modulation and demodulation circuits of optical communication and violet LED communication circuits are designed and software algorithms are implemented via System Verilog. Finally, the whole system was tested and the experimental results proved that in the condition of satisfying communication speed and BER, the feasibility of effective transmission of image information could be verified.

Abstract:To focus on low drift of sea ice and to meet the accuracy requirement of radar monitoring system, a video accumulation and optimized mean filtering algorithm is proposed to detect sea ice concentration by data filtering in both time and space dimensions. The deconvolution algorithm is adopted for radar echo processing, which can reduce the angular error due to the horizontal beam width (θH) and the radial error due to the radar pulse width (τ) effectively. Based on the statistical characteristics of noise, simulation is used to verify the performance of video accumulation and optimized mean filtering algorithm on the Signal-to-Noise Rate (SNR) of sea ice echo images. Taking the measured data of vehicle-mounted mobile radar FAR-2117 in Xiajiahe for algorithm validation, the results show that the proposed method has high accuracy in detecting sea ice concentration and the error is satisfied to meet the engineering requirement.