Abstract:The dielectric constant of rock is an important parameter to characterize its electromagnetic radiation characteristics, which is of great significance to remote sensing observation of the Earth. However, the existing methods for measuring dielectric constant usually have specific requirements on the shape and size of samples, and the measured dielectric constant can only represent the dielectric characteristics of local small areas of rocks, but cannot characterize the dielectric characteristics of rocks as a whole. Therefore, based on the theory of microwave interference, a new method for measuring the dielectric constant of rock by microwave radiometer is studied and proposed in this paper. Then, based on this method, the dielectric constants of granite and sandstone are measured by C-band microwave radiometer. In order to verify the correctness of the test results, the AET open coaxial resonator method is used to compare and verify the results. The results show that the dielectric constants of granite and sandstone measured by the new method are 3.89 and 5.71 respectively, and the relative errors are 3.2% and 2.7%, respectively, compared with those measured by AET open coaxial resonant cavity method. The research results show the feasibility of the new method, which is a useful supplement to the current point dielectric constant measurement method.

Abstract:Because of the module's negative-level output performance of the full bridge, the hybrid MMC with an AC modulation ratio greater than 1 is satisfied. The optimal design of hybrid MMC ratio based on high AC modulation ratio, which can reduce the DC failure and increase the AC grid voltage level. The operating area of DC voltage is extended. It is of great significance to the development of flexible HVDC transmission. Therefore, a full-bridge proportional constraint considering DC fault ride-through and half-bridge sub-module capacitor-voltage balance based on the high AC modulation ratio was proposed. The transient characteristics of DC faults and the voltage fluctuations of half-bridge sub-module capacitors were analyzed in ±500kV double-ended hybrid MMC system model. The comprehensive optimization design scheme of the mixed MMC ratio was obtained. The results show that the propor-tion of the full bridge should not be less than 41%. The system voltage recovery time is first reduced and then increased after the fault. The fault self-clearing time is shortened, and the half-bridge sub-module capacitor voltage fluctuation is reduced by 8.15kV. The capacitor voltage fluctuation is positively correlated with the AC modulation ratio. The module fault handling capability and overall performance of MMC with a full bridge ratio of 75% is better. The research results provide a reference for the lightweight design of MMC.

Abstract:Aiming at the requirements of high precision, low noise, high resolution and programmable pulse signals in the fields of radar, communication, electronic metrology and testing, a picosecond-level adjustable pulse width pulse code generation circuit is designed to generate multi-mode and multi-function serialized pulse code signals with precise and controllable pulse width. The pulse code generation circuit is based on the principle of fractional frequency division. By changing the fractional frequency division ratio, the fractional spurious is moved to the high frequency band and filtered by the loop low-pass filter to reduce the noise of the pulse signal. On this basis, the target signal is generated by the parallel-series conversion chip and the clock signal is provided to the FPGA to make up for the shortcomings of low clock frequency and poor accuracy of the FPGA itself. The test results show that the pulse frequency range generated by the pulse generating circuit is 1mHz~400MHz, Minimum pulse width duty cycle step is %~ The output signal type of the pulse signal generation circuit can select the pulse code type signals in the pulse code format of return-to-zero code, non-return-to-zero code, return-to-zero code and pseudo-random code.

Abstract:Fuzzing is the main technology of software vulnerability mining. It can randomly generate test cases and dynamically execute programs that can cover deeper branches. However, there is a certain blindness in mutation in fuzzing technology, and the frequency of random mutation samples executing the same path is very high, resulting in redundancy of mutation samples, thus reducing the test efficiency. This paper proposes and implements a guided grey-box fuzzing method CTM (Control Flow Test Case Generate And Mutation) based on control flow analysis. CTM first statically analyzes the target binary program to obtain the program control flow graph, then analyzes the execution rarity of the program path according to the program control flow, then identifies the sensitive functions on the execution path to calculate the program execution path proportion, and solves and generates test cases; The position of non-format key information is mutated in the testing process; Finally, according to the feedback information of branch coverage, the execution path constraint information is solved by heuristic rules to generate new test case samples. CTM improves the probability of fuzzing to generate test cases that satisfy complex branch conditions through guided test cases and locating mutation methods, thereby improving code coverage and reducing mutation sample redundancy. In order to verify the effectiveness of this method, this paper selects real applications such as readelf and gif2png for testing, and compares it with the mainstream Fuzzing software Driller and AFL in the industry. The test results show that CTM's ability to detect crashes and explore new paths has been improved.

Abstract:In order to accurately predict rolling bearing degradation trend, a combined prediction method of KPCA and TCN-Attention was proposed. Firstly, KPCA was used to extract nonlinear features from the high-dimensional feature sets of bearings, and the first principal component was used as the performance degradation index of bearings to normalize and smooth the first principal component. then, an attention mechanism is added to the temporal convolutional network TCN, the weight coefficients of the key features of the hidden layer are given, the part that contributes the most to the local features extracted by the TCN at each time step is found, and then the key information is extracted; finally, the Cincinnati IMS is used. The life cycle data of the bearing outer and inner rings verify the feasibility of the method. The experimental results show that compared with TCN, Gru and LSTM without attention mechanism, the predicted values of RMSE and MAE of the outer loop are reduced to 0.00299 and 0.00217, respectively, and the predicted values of RMSE and MAE of the inner loop are reduced to 0.03401 and 0.02490, respectively , with higher prediction accuracy.

Abstract:A VLF/UHF band power amplifier with low operating frequency and wide bandwidth is designed. In order to meet the index requirements, the design adopts three-stage transistor cascade. Selected crystal after the transistor, the input, output and inter-stage matching networks are designed according to the structure and performance of transistors at each stage. The multi-stage L-shaped matching network is really due to the power amplifier’s output, In the impedance matching circuit, the output impedance matching circuit is realized by multi-stage L-shaped matching network combined with 1:1 coaxial balun impedance matching network. The output impedance of the gain stage transistor and the input impedance of the driver stage transistor are matched to 50Ω respectively, and then the inter-stage matching circuit design between them is realized in cascade. The output impedance of the driver stage transistor and the input impedance of the output stage transistor are directly matched together through the multi-stage L-shaped matching network, and the inter-stage matching circuit design between the driver stage transistor and the output stage transistor is realized. Process a 500kHz~1GHz power amplifier is manufactured. The test results show that the output power can reach more 20W, the gain flatness is better than ±3.0dB, and the working efficiency is about 20%.

Abstract:In order to improve the control performance of the intelligent pneumatic control valve, this article was based on the modeling analysis of the pneumatic actuator, A optimal control strategy of control valve based on model parameter learning was proposed. Firstly, the dynamic model of the pneumatic actuator was established, and the five-step switch control algorithm was analyzed. Secondly, the control parameter self-learning strategy required for optimal control was designed based on the model. Finally, according to the control parameters obtained by parameter self-learning,the five-step control method was improved to give an optimized control strategy and implementation steps. The experimental results show that there is no obvious overshoot in the control process of the proposed optimal control algorithm, and the oscillation is significantly weakened. The control accuracy is significantly improved, and the adjustment time is shortened. The average adjustment time of small strokes is shortened by 38.1%, and the average error is reduced by 61.4%.The average adjustment time of large stroke is shortened by 38.7%, and the control accuracy is improved by 39.4%.

Abstract:The randomly deployed nodes will lead to the insufficient coverage in Wireless Sensor Networks (WSNs). To solve this problem, an improved sparrow search algorithm - increment of coverage ratio (ISSA-ICR) was proposed. Firstly, to solve the problem that the producer converging to the origin, ISSA modified the location update method of the producer to avoid the algorithm falling into the local optimal solution; Secondly, to balance the global and local search ability of the algorithm, t-distribution disturbance with the number of iterations as the degree of freedom parameter and the dynamic adjustment strategy of the number of producers- scroungers were proposed; Thirdly, random regression cross-border processing strategy was adopted to solve the problem of individual cross-border relocation, and the candidate location of nodes to be deployed was determined; Finally, the node scheduling optimization model was constructed based on ICR strategy to determine the final deployment location. The simulation results show that compared with sparrow search algorithm, standard particle swarm optimization and adaptive virtual force disturbance sparrow search algorithm, ISSA-ICR can respectively improve 4.96%, 8.81% and 3.84% coverage ratio compared with the three algorithms, meanwhile reducing the nodes moving distance.

Abstract:In order to improve the accuracy and efficiency of existing registration algorithms, a point cloud registration algorithm based on point cloud feature vector extraction was proposed. The point curvature and the number of points in the neighborhood are used as comprehensive criterion to filter feature points, and then feature vectors are extracted by principal component analysis of feature points. The transformation relationship of feature vectors is used to solve the transformation matrix between the point clouds to achieve rough registration of point clouds. In the precise registration, the point cloud k-dimensional binary tree is created, and the nearest neighbor search by the k-dimensional binary tree is used to improve the precision registration efficiency of the ICP algorithm. The proposed algorithm was compared with a variety of algorithms in the public data sets Bunny and Horse and the measured environmental point cloud data to verify the effectiveness. The results show that computation time is reduced by 60% compared with ICP algorithm, and the proposed algorithm has good accuracy and registration efficiency.

Abstract:Missing data reduces data availability. Prediction of missing data becomes very important. A prediction algorithm named ISSA-DELM(Improved Sparrow Search Algorithm optimized Deep Extreme Learning) was proposed to solve the problem of missing data. First of all, singer chaotic map, Cauchy-Gaussian mutation strategy and cosine weight factor combined with sparrow search algorithm. Secondly, the input weights and biases of the autoencoders in each extreme learning machine in the deep extreme learning machine are optimized by ISSA. Then ISSA-DELM is applied to predict missing data. The experimental results show that, ISSA-DELM has strong stability and the highest prediction accuracy compared with SSA-DELM、Particle Swarm Optimization DELM( PSO-DELM)、DELM in the case of small data volume and low miss rate. The evaluation indexes, such as RMSE, MAE and the coefficient of determination are better than the compared algorithms.

Abstract:The meteorological data is multi-element time series. In order to solve the problems of large prediction error and insufficient time feature extraction of traditional temperature prediction algorithm, a GRA-Conv-BiLSTM temperature prediction method is proposed by integrating grey correlation analysis, ConvLSTM and BiLSTM together. The grey correlation analysis method is used to solve the problem of difficult parameter selection in the traditional methods, and the time window is set. The grey correlation analysis method solves the problem of difficult parameter selection in traditional methods. The time window is set, combined with the historical temperature as the input of the model, the prediction model of ConvLSTM and BiLSTM dynamic weighted fusion is established to enhance the spatio-temporal feature extraction ability of the model, and the experiment is carried out with the historical data of a meteorological station in Sichuan Province as a sample. The results show that for the multivariate meteorological time series with a large amount of data, the model shows stronger advantages, can adapt to dynamic nonlinear changes and has higher prediction accuracy.

Abstract:Traditional extractive and abstractive methods lack readability and accuracy in the summary auto-generated task, so a HRAGS (Hybrid Guided Summarization with Redundancy-Aware) model-based hybrid summary generation method was proposed. First, the method used the BERT pre-trained language model to obtain a contextual representation and combined with redundancy-aware method to construct an extractive model. Then a couple of trained BERT encoders were united with a randomly-initialized Transformer decoder contained two encoder-decoder attention modules to construct an abstractive model. The abstractive model adopted a two-staged fine-tuning approach to resolve the training imbalance problem between encoders and decoders. Finally, an Oracle greedy algorithm chose key sentences as external guidance and source document with guidance were put into the abstractive model to acquire a summary, which was verified on the LCSTS evaluation dataset. Experimental results shows that the HRAGS model can generate a more readable, accurate and high ROUGE score summary compared with other benchmark models.

Abstract:Aiming at the problem of array correction in the anti-interference processing of satellite navigation array, an array antenna adaptive correction technology based on carrier phase measurement is adopted in this paper. Each array element of the array antenna is used to receive the same satellite signal, capture, track, observe and extract the carrier phase of each channel of the specified satellite, calculate the carrier phase difference, calculate the error between each channel, and correct the amplitude and phase inconsistency of the array antenna, so as to reduce the impact of amplitude and phase inconsistency on the performance of anti-jamming algorithm. Compared with the traditional array antenna adaptive correction algorithm, it has the following advantages: 1) it can be corrected separately for each satellite; 2) It can comprehensively correct the channel phase inconsistency caused by RF channel inconsistency, antenna array installation error and AD sampling inconsistency; 3) The channel error of the satellite can be corrected in real time. Simulation results show that this method has good amplitude and phase error suppression effect.

Abstract:In response to the problem that the traditional logistics distribution operation process is cumbersome and vulnerable to human factors, the method of introducing drones into logistics for distribution is proposed. Firstly, the Petri net theory is used to model the traditional logistics and distribution system and the UAV logistics and distribution system respectively. Second, a Markov chain is constructed based on stochastic Petri nets, and the performance of the model is analyzed using the Markov chain; then the reduction rules of stochastic Petri nets are used to equivalently transform the two models and calculate the average operating time of one distribution business process. Finally, a comparative study is carried out on the analysis results. The research and comparison results show that, compared with traditional logistics distribution, the use of drones to re-plan logistics routes, while satisfying the rationalization of the entire route, improves the time-limited performance by 25.6%. It can be seen that the use of drones for logistics distribution can be significantly Shortening the delivery time and effectively improving the delivery efficiency has certain theoretical reference significance for actual logistics problems.

Abstract:Aiming at the problem of distortion in the edge region of the existing fish-eye image distortion correction algorithm, a fast correction algorithm is designed by introducing the stretching factor based on the longitude correction algorithm. Firstly, the effective area of the image is extracted by the fast scanning method to obtain the radius and center of the effective area of the distorted image. Secondly, based on the fish-eye image distortion model and the radial distortion principle, the distortion correction stretching factor is determined, and then the stretching factor is added in the longitude and latitude directions for correction. Finally, the correction results are compared with the longitude correction algorithm, longitude and latitude correction algorithm , double longitude algorithm and the recentering correction algorithm. The results show that the algorithm can effectively improve the distortion of image edge, and the correction efficiency is significantly improved compared with other algorithms. In conclusion, the algorithm proposed in this paper can quickly and effectively correct the fish-eye image with high quality, and provide reference for the research on improving the quality of distortion correction.

Abstract:In order to simulate heterogeneous distribution in microscopic displacement experiment, it is necessary to splice several different pore binary images into a complete image. However, these spliced images do not have overlapping areas, so image splicing and restoration should be carried out according to the texture information of the image to be spliced. This paper studies the correlation of image neighborhood and proposes an image Mosaic method based on block matching and multi-level sampling. This method combines five decision criteria to repair curved contours and irregular textures between pore image boundaries. Through the image Mosaic process from coarse to fine scale, The final spliced thin section pore image can reflect the core characteristics more truly. In order to verify the effectiveness of this research method, the proposed algorithm is compared with existing traditional image repair algorithms and image repair methods based on deep learning, and the results of image Mosaic examples are evaluated by subjective visual and objective indicators. The results show that the proposed algorithm is superior to the existing image restoration algorithms in PSNR and SSIM, improving 6.08dB and 0.015 respectively, and has better performance in natural texture transition and overall structure consistency.

Abstract:The mainstream target detection algorithms in the field of automatic driving have poor detection effect on small targets, which poses a threat to driving safety. The one-stage anchor-free CenterNet algorithm is improved to solve this problem. Firstly, the original backbone network is replaced by ResNeSt50 network with split-attention mechanism, and the ReLU activation function is upgraded to FReLU, which strengthens the effect of feature extraction with little additional computational overhead; Then, a lightweight network PASN is proposed to fuse semantic features of different scales, and Spatial Pooling Pyramid (SPP) module is introduced into the shallow feature input to enhance the expression of small target information; Finally, random multi-scale input training is carried out on Kitti data set. The verification set results shows that the FPS of the improved algorithm reaches 37.7, meets the real-time requirements, the average precision of small targets is improved by 12.9% and the mean average precision is improved by 13.9%， At the same time, the detection speed and average precision are higher than the mainstream algorithm Yolov4；It can detect 31 images per second on the real vehicle, which provides strong support for the development of automatic driving technology and has engineering application value.

Abstract:Human action recognition is one of the research hot-spots in the field of computer vision. It has far-reaching theoretical research significance in human-computer interaction, video surveillance and so on. In order to solve the problem that 2D CNN can not effectively obtain time relationship, based on the advantages of Transformer in modeling long-term dependency, Transformer structure is introduced and combined with 2D CNN for human action recognition to better capture context time information. Firstly, 2D CNN integrating channel-spatial attention module is used to capture the inter spatial features. Then, Transformer is used to capture the temporal feature between frames. Finally, MLP head is used for action classification. The experimental results show that the recognition accuracy of HMDB-51 datasets and UCF-101 datasets is 69.4% and 95.5% respectively.

Abstract:Defect detection on outer packaging of wine bottles is of importance. A improved YOLOv3 algorithm is proposed to deal with that problem. The final result shows the improved YOLOv3 meets the production line’s requirements for accuracy and speed well. First, The SENet Module is introduced into YOLOv3 backbone network’s residual block, applying the attention mechanism to enhance the feature extraction. Second, The Adaptive Feature Fusion Network (ASFF) is introduced into the Feature Pyramid Network to fuse the feature information in different scales, which enhance the predictive ability of the model. Third, The Focus Loss function is used to solve the problem of unbalanced positive and negative samples, which will help accelerate the convergence speed of the loss function. The improved YOLOv3-ASFL achieves a mAP up to 92.33% in the self-made wine bottle cap dataset, which is 6.59% higher than the original YOLOv3, and the single image detection time is only 0.085s. The improved YOLOv3 model has a better performance and meets the needs of the wine bottle packaging production line for defect detection.

Abstract:In order to reduce the incidence of accidents caused by fatigue driving, a method is proposed to build a fatigue driving detection model by integrating convolutional neural network with face feature points and fatigue indicators. Firstly, the driver's eyes and mouth areas are located by the SSD network, and the VGG16 network learns the fatigue features contained in the eye and mouth areas. At the same time, 68 feature points of face, eye aspect ratio and mouth aspect ratio are combined to determine the driving fatigue state. Finally, the mean average precision of SSD algorithm and Faster-RCNN algorithm is calculated under the same test set. The model is applied to YawDD dataset. And the feasibility of this model is verified by simulating driving environment. The experimental results show that SSD algorithm is better than Faster-RCNN algorithm, the detection accuracy of this model on YawDD dataset is about 97.2%, and the camera can also detect the driver's state in real-time. The model is effective in detecting fatigue state and reducing the accident rate caused by fatigue driving to a certain extent.

Abstract:Aiming at the problems of low accuracy and slow speed in the detection of surface defects of lithium batteries by traditional methods, an improved YOLOv4 algorithm is proposed. Firstly, a dilated convolution is used to replace the conventional convolution in the CSPDarknet-53 backbone network, which improves detection of defects of different scales. Secondly, an efficient channel attention is inserted into the neck network to adaptively select the size of the one-dimensional convolution kernel to reduce the complexity and computations of the model. Finally, a conditional convolution is fused in classification and bounding box regression to improve the network performance, and the data set is expanded to solve the problem of network training overfitting caused by too few defective samples. The experimental results show that the improved YOLOv4 algorithm can effectively detect the surface defects of lithium batteries and improve the ability to identify and locate surface defects of lithium batteries. The mean average precision of the improved algorithm is 93.46%, which is 3.03% higher than the original algorithm.

Abstract:Aiming at the problems of low segmentation accuracy and over-segmentation in the current automatic segmentation of prostate tissue regions on magnetic resonance images, a U-Net segmentation algorithm combining dense connections and Inception modules was proposed. Firstly, the contrast-limited adaptive histogram equalization method was used to process the prostate image to enhance the detectability of the information. In addition, the algorithm introduces the idea of ​​dense connection into the U-Net model, improves the connection method of the original encoder and decoder, and realizes the fusion and dissemination of multi-scale semantic information. Meanwhile, the Inception module driven by atrous convolution is used to replace the original concatenated convolution operation to increase the width of the network and enhance the feature extraction and expression capabilities for objects of different sizes. Finally, for the over-segmentation problem of non-organized objects, a corrector with classification-guided function is designed to reduce false positive predictions. By testing on the public dataset of NCI-ISBI 2013 Challenge, using Dice similarity coefficient, accuracy rate and false positive rate as evaluation criteria, the mean values ​​can reach 86.12%, 97.96% and 1.11%, respectively. The experimental results show that this algorithm has better segmentation effect than other segmentation algorithms.

Abstract:The Auto Associative Kernel Regression (AAKR) algorithm does not consider the contribution of each element in the state vector to the Euclidean distance when calculating the similarity, and the model parameter is often calibrated based on subjective experience. As a result, the accuracy of the model is relatively low. A non parametric modeling method for establishing the normal behavior model of gearbox is proposed based on the SFO algorithm and the modified AAKR algorithm. Firstly, the memory matrix is constructed by full parameter equal interval partition method; Secondly, the distance weight coefficient is introduced into the AAKR model, and the width coefficient and distance weight coefficient in the AAKR model are optimized by SFO algorithm; Finally, the health index is constructed based on sliding window and residual data to realize the condition monitoring of wind turbine gearbox. Taking the measured data of a 2MW wind turbine as an example, the results show that compared with the traditional AAKR, weighted AAKR and robust state estimation model, the average accuracy of the proposed algorithm is improved by 1.55%, 0.6% and 0.76% respectively. In fault early warning, the constructed health index can more sensitively and accurately reflect the early fault and development trend of gearbox.

Abstract:A modified MUSIC algorithm based on fractional low order moments is proposed to estimate the direction of arrival of coherent signal sources in the background of impulse noise. Firstly, the fractional low-order moment matrix is calculated according to the received data of the array, so as to suppress the impact of the impact noise on the received data of the array. Then, the basic principle of modified MUSIC algorithm is adopted to solve the rank loss of array covariance matrix caused by the existence of coherent signals by conjugate rearrangement of matrices, and the estimation performance of the algorithm is improved. The effectiveness of the proposed method is verified by computer simulation experiments. Experimental results show that: The proposed algorithm can effectively suppress the impact of the impact noise and de-coherently process the coherent signal, and has higher estimation accuracy and probability of success than the fractional low-order singular value decomposition algorithm and the adaptive reconstruction method after data preprocessing.

Abstract:In the measurement process of thermal mass flowmeter, due to the systematic error of analog -to-digital converter (ADC) itself, and the influence of input signal and external circuit noise, the digital quantity signal transformed by ADC will deviate from the theoretical value and constantly fluctuate. Therefore, we propose a software filtering method based on the improved Kalman algorithm, which can improve the convergence speed of kalman algorithm while eliminating the error. In addition, the probability distribution model of data error is established by probability statistics method, which provides theoretical basis for kalman parameter selection. Finally, the algorithm is implemented on FPGA and verified by experiments. Compared with the data before filtering, the average error is reduced from 1.81% to 0.82%, and the data fluctuation is less than 2LSB.The results show that the method can effectively eliminate errors and improve the stability of the measurement system.[ 基金项目：国家自然科学基金资助项目（12074354）]

Abstract:In view of miniaturization of multi band transmission antenna in wireless communication system, this paper presents a miniaturized slot antenna, which is fed by L-shaped microstrip line, a slot with an opening at one end formed by three steps on the ground plane to cut off the current path, so as to realize dual-band coverage of WLAN band. The electromagnetic simulation software is used to simulate and optimize the parameters of the antenna, by changing the length and width of the slot, the antenna can work at 2.45GHz and 5.8GHz at the same time, the results show that the antenna port return loss S11≤-10dB, and the working bandwidth are 2.4~2.49GHz and 5.48~6.1GHz respectively, which meet the requirements of IEEE802 11b/g and IEEE802 11a/n standards. Meanwhile, the antenna is fabricated, and the port return loss, gain and radiation pattern are tested, the simulation results are in good agreement with the test results. Compared with the traditional planar slot antenna, the antenna proposed in this paper is compact in size, simple in structure, easy to implement and well to integrate with the circuit, which hopes a wide range of communication applications in the future.