Abstract:Aiming at the problems of a large amount of communication data and high communication costs existing in the current power information acquisition system, a power information acquisition system based on a uC/OS embedded real-time operating system is designed. The system includes a smart electricity meter, power information acquisition terminal, host computer software, and master station data center. According to the parameters set by the upper computer, the system collects the data in the electricity meter downlink. After data compression processing in the acquisition terminal, the data is transmitted to the master station through remote communication technology, which is convenient for the staff to monitor and manage the power data. In this paper, the Lempel-Ziv-Oberhumer lossless compression technology and run-length coding technology are combined to improve the data compression scheme, and the feasibility of the data compression scheme is verified by the comparison experiment of data compression. The experimental results show that the data compression technology used in the system effectively improves the communication efficiency of data transmission, and verifies the practicability and effectiveness of the system.

Abstract:According to engineering needs, this paper proposes a new type of high-power microwave active phased array antenna unit. The power capacity and standing wave characteristics are improved by modifying the structure of the coaxial waveguide converter. A square metal block was introduced between the coaxial probe and the stepped waveguide to achieve impedance matching, a transition cavity and matching slot structure were added at the end of the waveguide, and a single ridge structure was added to the horn section to change the waveguide impedance. The single ridge extends out of the aperture surface to reduce the reflection coefficient and further improve the power capacity. In X-band, the VSWR is less than 2, the bandwidth is improved to 1.25GHz, and the power capacity is 33.24kW, which is 230% higher than that of the horn antenna under the conventional coaxial waveguide converter. The 7×7 phased array can achieve MW-level transmission power and has the ability to transmit high-power microwaves. The array can achieve ±30° conical scanning to meet the needs of high-power wide-angle scanning.

Abstract:In order to maximize energy efficiency in HetNets (Heterogeneous Networks), the problem of energy efficiency optimization is designed as a multistage decision problem in this paper. According to the resource allocation of the optimization goal, the initial problem is decomposed into two sub-problems that optimize the parameters ABS (Almost Blank Subframe) ratio and CRE (Cell Range Expansion). The MAAC (Multi-agent Actor-Critic) algorithm is used to solve the sub-problem, and then the initial optimization problem is solved by iterating over the solution of each optimization sub-problem. In the process of parameter optimization, a single small base station is used as an agent, and the MAAC algorithm is used to find the optimal solution for each CRE, so as to realize the asynchronous CRE optimization between cells. Experimental results show that compared with the Turbo Q-learning algorithm, the convergence speed of the proposed method is increased by 40%, and the load between small base stations is more balanced through asynchronous optimization of CRE.

Abstract:Aiming at the problems of high loss and low consistency of the coupling structure of the existing liquid-leakage monitoring fiber sensor with side coupling structure, a liquid-leakage sensor with cylinder defect coupling structure was proposed. The optical fiber coupling simulation is carried out by Zemax software, and the cylindrical, cone and semi-sphere structures are compared and analyzed. The cylindrical coupling structure has higher coupling rate when the insertion loss difference is small. Polymer fiber (POF) was processed to obtain a cylindrical coupling structure with insertion loss of 0.31dB.The LED lamp belt is used as the scanning light source to verify the detection ability of the optical fiber with coupling structure. When the coupling medium changes from air to water, the optical power intensity changes up to 32.8% or more.The experimental results show that the loss of the cylindrical coupling structure is significantly lower than that of the existing coupling structure, and the optical fiber sensor can effectively detect the liquid leakage.

Abstract:At present, the low measurement accuracy of a single MEMS gyroscope mainly restricts the wide application of MEMS gyroscopes in aerospace, automotive, weapon guidance and other fields. However, it is difficult to rapidly improve the accuracy of a single gyroscope based on the current process and technical level. This paper studies how to use low accuracy MEMS gyroscope to obtain high accuracy navigation information through array technology and data fusion algorithm. Firstly, several similar gyroscopes with matching key parameters are combined to form a virtual gyroscope by array technology. Then, the multi-input system is fused by filtering method to give an optimal estimation value, so as to achieve the purpose of improving the sensor accuracy with the help of quantity. Finally, experimental verification is carried out. The verification results of simulation data and measured data show that the redundant gyroscope based on array technology can reduce the slow drift in the gyroscope and improve the angular velocity measurement accuracy after data fusion through the designed filtering method.

Abstract:Aiming at the problem that the sparrow search algorithm is easy to fall into the local optimum and the con-vergence accuracy is insufficient, this paper proposes an improved sparrow search algorithm based on tabu search algorithm. Firstly, Latin hypercube sampling is used to initialize the population to ensure a more uniform distribution of initial spatial solutions; secondly, the time-varying Cauchy mutation operator is introduced to optimize the discoverer search strategy when R2

Abstract:To improve the efficiency of obstacle avoidance path planning of collaborative robot in complex environments, improved Artificial Potential Field & Rapidly exploring Random Tree (APF-RRT) algorithm for ellipsoid obstacles. First of all, enveloped obstacle with ellipsoid bounding box,then the collision detection method was analyzed; In addition, to speed up path exploration Random point effect and random point selection mechanism were introduced to improve the RRT algorithm, then the hybrid method of improved RRT and artificial potential field was uesd for path planning; Finally, deleted redundant points on the planned path, and four-time B-spline curve was used for path smoothness, improved path quality.The simulation results showed that, in the ellipsoid obstacle environment, the algorithm can complete the obstacle avoidance path planning quickly and the planned path was shorter.

Abstract:Noise feedback active control systems usually have the disadvantage of slow convergence and high control residuals. This paper has designed a Nearest Neighbors-Multifrequency Notch-Filter (NNR-MNF) model for dealing with line spectral noise. It calculates the approximate solution of the optimal filter coefficients in the time domain and then iterates the filter parameters from a position close to the optimal solution. This procedure allows the system to control complex noise by using a small learning rate. As a result, it can reach rapid convergence while avoiding the problem of system divergence. The experiment results based on the destroyer engine noise dataset show that the NNR-MNF algorithm reduces the convergence time by about 70% compared with the traditional control method. The results show that the use of neural network methods based on machine learning can effectively improve the control speed of noise active control systems, and provide a new optimization solution for active noise control problems.

Abstract:Aiming at the problem of low accuracy in non-contact measurement of respiratory rate (HR) and heart rate (BR) in indoor complex environment, this paper proposes a vital sign detection method based on 77GHz mmwave radar perception. Through multiple experiments at different distances from the radar and taking the measured data of Mindray ECG monitor as a reference, the accuracy of this method is verified. Firstly, the multi pulse if signal is processed by incoherent accumulation technology between pulses to improve the signal-to-noise ratio (SNR) and signal amplitude. Then, the target detection method based on distance dimension FFT spectrum and CA-CFAR adaptive threshold is used to extract the human target phase information from the indoor complex environment, and the respiratory and heartbeat signals are separated by FIR band-pass filter. The improved smoothing filter combined with the second-order phase difference method is used to remove the noise and reduce the impact of respiratory harmonics on the heartbeat signal. Then, the root MUSIC algorithm is used to obtain HR and Br. Finally, the measurement results of Mindray ECG monitor are used as a reference to verify the measurement accuracy of this method. The experimental results show that the average absolute error rates of HR and Br obtained by this method are less than 5.08% and 2.61% respectively, and the average absolute error is 0.94bpm and 1.97bpm. Therefore, this method can improve the measurement accuracy.

Abstract:In order to make full use of forest resources and effectively improve the utilization rate of wood, a dual-manipulator air-coupled ultrasonic testing system was constructed based on industrial manipulator technology and air-coupled ultrasonic testing technology, and the verification experiments were designed and carried out for the system defects inspection resolution and defects reproduction ability. The experimental results show that when the 100 kHz focusing probe is used for defect inspection of wood samples, the dual-manipulator air-coupled ultrasonic testing system has a good capacity to detect defects such as knots larger than 5 mm in wood with a thickness of 15 mm, and has a strong capacity to reproduce the shape of the defects of complex shapes and to inspect defects in parts with complex-curved surfaces. The successful implementation of the testing method can provide a scientific basis for the integrity and usability assessment of wood.

Abstract:During the operation of high-speed trains, due to the impact of different degrees of foreign bodies on the bottom plate of the equipment cabin under the train, the decline in the performance of composite materials affects the safety of train operation. Therefore, it is particularly important to find a reliable and accurate method to detect the damage location. Due to the change of structural parameters and modal characteristics of composite materials, a modal analysis method is proposed to predict the stress position of composite materials, and the relationship between modal parameters ( natural frequency and relative amplitude ratio ) and stress position is studied. The plane model of the composite was established by the modal analysis software m+p SO Analyzer, and the vibration characteristics of the composite were tested and analyzed. The contour maps of the natural frequency and relative amplitude ratio of the composite under different positions were established. Furthermore, the distance from the stress position to the center of the composite material is determined by the natural frequency ; finally, the stress position of the composite was determined by combining the relative amplitude ratio. The results show that the force position of carbon fiber epoxy resin composite specimen can be successfully predicted based on the change of natural frequency and amplitude, and the error can be controlled within 5 %.

Abstract:In view of the difficulty in detecting hot spots of photovoltaic panels in power stations in China, combined with UAV inspection technology, a fast detection method of hot spots of photovoltaic panels based on deep convolutional neural network was proposed. Firstly, a photovoltaic panel recognition model was designed. The Yolov4 backbone feature extraction network was replaced by the lightweight MobileNetV2 network, and the standard 3×3 convolution in PANet was replaced by the deeply separable convolution, which could realize the rapid recognition of photovoltaic panels from infrared images. In order to quickly identify hot spots and solve the problem of reflective noise of photovoltaic panels, MobileNetV2 network is introduced into deeplabv3 + model, improve the target loss caused by sampling and the cross entropy loss function is modified to dice loss function to further improve the segmentation accuracy. The experimental results show that the method can accurately identify hot spots of photovoltaic panels, with an accuracy of 99. 56% and a detection speed of 22. 1 frames per second. The hot spot segmentation accuracy of photovoltaic panel recognition reaches 95. 99%, MIoU reaches 85. 58%, and the detection speed is 24. 5 frames per second. This method can meet the needs of photovoltaic panel fault detection.

Abstract:In order to measure the nacelle internal flow resistance accurately, It is necessary to study the structural design of thenacelle internal flow resistance balance. On the basis of summarizing and analyzing the usage of conventional balance, we adopt the method of great stiffness, force amplification function elements as basic sensitive unit, combining multiple balance elements,development of five–component force balance successfully.The stiffness of the balance is seven times higher than that of the conventional column-beam balance, the veracity index of the balance is up to the standard of the GJB.In the wind tunnel test, each element of the balance returns to zero well, the dynamic performance is stable, and the law of measuring load is reasonable,The precision of resistance measurement reaches 0.0001 level. It provides a reference for the development of the force balance for measuringthenacelle internal flow resistance of civil aircraft.

Abstract:The drone platform has small memory and limited computing resources.Aiming at the problems of complex network structure and slow detection speed of classical detection methods, a real-time detection method based on lightweight is proposed. Firstly, the lightweight model MobileNetv3 replaces CSPDarknet53 as the backbone network and the effective channel attention mechanism is fused to reduce the memory occupation of the model. Secondly, the residual structure fusion module RFM is introduced to enhance the feature extraction capability of the network. To further improve the generalization ability of obstacle detection and the convergence speed of the algorithm, the Control Distance-IOU loss function is replaced by the CIOU loss function for network training.The experimental results show that memory occupation of the improved model is reduced by 80% to only 39.5M and the FPS is improved by 168% to 49.21 frames/s under the same basic detection effect as the original model.

Abstract:At present, the detection of pipeline, especially high altitude pipeline, is mainly carried out by manual hand-held equipment, and the testing conditions are very limited, which is easy to cause low detection efficiency and safety risks. In order to improve the efficiency of pipeline inspection and protect the life safety of inspectors, this paper proposes a path information recognition algorithm based on machine vision for pipeline crawling robot. Firstly, region of interest (ROI) setting and perspective transformation are performed on the obtained images, and binary images of RGB and HLS color space are fused. Then, the contour extraction method not only realizes the extraction of obstacles and damages, but also compensates the non-pipeline pixels, which improves the stability and accuracy of polynomial fitting. Finally, the sliding pane method is used to extract the pixel of the fused binary image, and the centerline of the pipeline is fitted by polynomial. Experiments show that the algorithm presented can identify the centerline, obstacles and damage in complex ambient light accurately, and has strong anti-interference ability, which can meet the needs of pipe climbing robot to travel along the pipeline independently.

Abstract:In view of the limited depth of field of the image acquisition sensor, which leads to the out-of-focus phenomenon in the local area of the acquired image, this paper proposes a new multi-focus image fusion algorithm. Under the framework of NSST, the fusion rules based on discrete cosine transform (DCT) and local spatial frequency (LSF) are used for the low-frequency sub-band decomposition coefficients, and the fusion rules based on the maximum and minimum filtering combined with average filtering and median filtering (MMAM) are used for the high-frequency sub-band decomposition coefficients; and then perform INSST reconstruction to obtain fused images. The experimental results show that, compared with the classical image fusion algorithm, the proposed algorithm can effectively fuse the high and low frequency sub-band information of the image, and achieves better results in both subjective and objective evaluation.

Abstract:In order to overcome the problem of vulnerable disturbance of target detection in complex and variable remote sensing situations, this paper suggests SD-Centernet, which is an optional orientation target detection model combining self-Calibrated module and D_Triplet Attention. This new method introduces rotation angle in the network structure, which provides angular information to the detection box. Self-Calibrated module is introduced into the Dlanet feature extraction network to increase the perceptual field of the output features by fusing information from two different spatial scales through an adaptive calibration operation. Meanwhile, D_Triplet Attention is introduced to enhance the focus of image-based local information, which better solves the cross-dimensional interaction problem. 86.25% detection accuracy and 14.9 fps detection speed have been achieved on SD-Centernet in HRSC-2016 Dataset, that effectively improves the multi-directional target detection in remote sensing aerial photography.

Abstract:The DeblurGAN method uses Conditional Generative Adversarial Networks (cGANs) to solve the end-to-end image deblurring problem, but there are problems of insufficient image edge detail recovery and low robustness. Aiming at this problem, a blind restoration method of motion blurred images based on DeblurGAN is proposed. In the generative network, a multi-scale convolution kernel neural network is used to extract features, and cascaded atrous convolution is used to expand the receptive field of neurons; an adaptive normalization method is used to replace the instance normalization method used in the original generator. Second, the gradient image L1 loss is introduced, combined with adversarial loss and perceptual loss, as a regular constraint for image deblurring, making the edge features of the generated image clearer. The experimental results show that the peak signal-to-noise ratio of the image restored by the proposed method is 5.4% higher than that of the DeblurGAN algorithm, and the structural similarity index is 1% higher; the subjective clearing effect is better, and the grid effect is eliminated.

Abstract:In recent years, traffic accidents caused by electric vehicle drivers driving without helmets have occurred frequently, resulting in great personal injury and loss. The investigation shows that most accidents occur at traffic intersections. Therefore, it is necessary to carry out the monitoring and control of helmet wearing behavior of electric vehicle drivers at traffic intersections. In this paper, a large number of target data of electric vehicles and drivers are collected by machine vision sensors, and the corresponding data sets are made. The processed data sets are trained on the pytoch framework by using the improved yolov5 neural network to obtain the optimal weight parameters; Compared with the original neural network, the improved Yolov5 algorithm has a detection accuracy of 92% and 98% for electric vehicles and helmets, which is 1% to 2% higher than that of the original neural network. Finally, the training improved yolov5 model and sort algorithm are used together to track and label electric vehicles while detecting their wearing helmets, so as to realize the effective control of illegal electric vehicle driving behavior at traffic intersections.

Abstract:In view of the problems of the existing fish classification network models, such as poor anti-interference ability, high computational resource consumption, and difficulty in field deployment, this study proposed a lightweight fish intelligent classification and identification model based on the improved EfficientNetV2 model. By introducing Hybrid Dilated Convolution and Coordinate Attention modules, this model improves the model structure of EfficientNetV2, increases the receptive field, improves the global attention of the model to the fine-grained features of the target, and enhances the anti-interference ability of the model. After training, the model was evaluated by comparative ablation experiments, and the results showed that the accuracy of the EfficientNetV2 - HDCA model proposed in this study on the verification set was 97.01 %, which was 3.8 percentage points higher than that the accuracy before improvement. The number of parameters in the improved EfficientNetV2 - HDCA model is 22.06MB, which is 0.45MB higher than that before improvement. In order to visually demonstrate the effectiveness of the EfficientNetV2-HDCA model proposed in this study, the Grad-CAM thermal experiment was also passed. The experimental results show that the model can extract the features of key parts of fish more comprehensively than before, and has a certain anti-interference ability.

Abstract:Aiming at the problem that single parameter characterization performance of aero-engine is not comprehensive, and is easily affected by external environment and flight conditions, a health indicator construction method based on Spectral regression and Gaussian mixture model for multi-source information fusion is proposed. Select the parameters that have complete records and are closely related to the engine health performance, take the difference value of the performance parameters of the left and right engines of the same aircraft as the data source, reduce the dimension of the features through Spectral regression, build a normal state model using the Gaussian mixture model, and then use the distance based on Bayesian inference to characterize the test data and the global distance of the Gaussian mixture model to identify the abnormal state of the engine. Verified by real QAR data from two aero-engine abnormal event cases, the results show that the proposed method can evaluate the health status of aero-engine more effectively and identify engine abnormalities in advance than that of airlines, reserve enough time to make reliable maintenance plans for the engine mechanism, and improve the safety and economy of the aircraft.

Abstract:The automobile paint film may shield the body from corrosion caused by the outside environment, but because of the effect of the manufacturing process, there will be a variety of flaws when the paint is applied. Therefore, the progress of automated automotive production depends on the identification of small flaws in paint film. Based on the actual needs of automotive paint film inspection, this paper proposes a visual image-based detection method for small defects in car paint film. This method first superimposes multiple frames of automobile paint film images to suppress noise; then uses two-dimensional wavelets to extract image trend items and characteristics of minor defects; finally, through the local adaptive threshold and the optimized segmentation coefficient of field curvature, the binary segmentation of defects is realized. The experimental results show that the method in this paper can effectively detect defects larger than 0.1mm on large-area images, and the recall rate of small defects reaches 97.6%, and the false detection rate is 1.3%, which can achieve the expected detection results.

Abstract:To achieve high-precision, full-coverage, all-weather satellite navigation, the performance of the system's services must be continuously and reliably monitored. A method of satellite navigation system service performance monitoring and analysis based on Long Short Term Memory(LSTM) neural network is proposed and implemented by using the monitoring data of monitoring stations around the world to solve and analyze position dilution of precision(PDOP) and positioning accuracy in order to further improve the accuracy and stability of satellite navigation system service monitoring and analysis. The experimental results demonstrate that, for PDOP, the mean accuracy of the results predicted by BDS and GPS based on LSTM is 5.15 and 3.89 percent higher than that predicted by ephemeris data, respectively; for positioning accuracy, the mean accuracy of the results predicted by BDS and GPS based on LSTM is 73.77% and 79.64% higher than that predicted by PDOP and user equivalent ranging error, respectively. It can be seen that the predictions made using the LSTM network outperform those made using ephemeris data in terms of prediction quality and localization accuracy. The method can effectively predict the future trend of data based on the historical data of PDOP and positioning accuracy, track the system service performance, and provide a reference basis for system service performance warning.

Abstract:In order to solve the problem of difficulty in estimating the powder output of the coal mill in thermal power plants, the soft measurement method is used to establish a BP neural network model combining the system parameters of the coal mill and the powder output of the coal mill, and the relationship between the parameters and the powder output is established. The non-linear mapping relationship is used to estimate the powder output of the coal mill. In order to reduce the error of the model, the WOA-BP algorithm model was established by using the Whale Algorithm (WOA) to optimize the weights and thresholds of the BP neural network. In order to verify the reliability of the WOA-BP algorithm model, the WOA-BP and PSO-BP of the coal mill's powder output were established respectively by the whale algorithm (WOA), particle swarm algorithm (PSO), genetic algorithm (GA) and BP neural network. , GA-BP, BP neural network algorithm model. The research results show that among the four algorithm models, the WOA-BP algorithm estimation model has the best prediction ability for the powder output of the coal mill, and the average absolute error is only 0.94.

Abstract:Based on wireless infrared temperature sensor and data acquisition terminal, a distributed temperature monitoring system for offshore electrical equipment was constructed to meet the actual requirements of temperature monitoring for offshore electrical equipment. The system application software was developed to realize continuous temperature monitoring for offshore electrical equipment. Aiming at the problem that traditional temperature prediction is difficult to deal with a large amount of fluctuating data and has limited ability to deal with time series, a combined prediction method of Bayesian optimization and LSTM was proposed. The temperature characteristics of the transformer during operation were analyzed by taking the transformer monitored offshore platform as the research object. The LSTM network prediction model with strong timing was adopted and the Bayesian optimization algorithm was introduced to train and update THE LSTM parameters. The practice shows that the LSTM model based on Bayesian optimization has good prediction effect on transformer temperature of offshore platform, and its root mean square error is 0.139 and prediction accuracy is 98.56%. Through the comparative analysis of four prediction models including support vector machine, BP neural network, LSTM and Bayesian LSTM, the advantages of Bayesian optimized LSTM model for offshore transformer temperature prediction are confirmed.

Abstract:To address the problems of poor real-time performance and insufficient small target detection capability of existing methods for quartz crucible bubble detection, a modified YOLOv5 algorithm for quartz crucible bubble detection, YOLOv5-QCB, is proposed. Firstly, a self-built quartz crucible bubble dataset is constructed, and based on the characteristics of small bubble size and dense distribution, the depth of network down-sampling is reduced to retain rich detailed feature information; meanwhile, the neck using dilated convolution to increase the feature map perceptual filed to achieve global semantic feature extraction; finally, the effective channel attention mechanism is added before the detection layer to enhance the expression of important channel features. The results show that compared with original model, the improved YOLOv5-QCB can effectively reduce the missed detection rate of small bubbles, improve the average accuracy from 96.27% to 98.76%, and reduce the weight by one-half, which can achieve fast and accurate detection of quartz crucible bubble targets.

Abstract:Blood count is a common clinical test. Aiming at the low accuracy of the existing blood cell detection methods due to the uneven, dense and mutual occlusion of blood cells in blood microscope images, an improved YOLOX blood cell detection algorithm is proposed. The algorithm firstly addes Focal loss into the loss function to improve the imbalance of positive and negative samples and uneven cell types in the single-stage target detection algorithm; then, a mixed attention mechanism is added into the residual module, which reduces the probability of missed detection and false detection caused by the mutual occlusion of blood cells; then an adaptive spatial feature fusion module is added at the end of the feature fusion to improve the feature expression ability; finally, an inverse depth separable convolution module is added into the residual module to reduce model parameters and slightly improve detection precision. The proposed algorithm is tested on the BCCD blood cell data set, and the detection accuracy of the improved YOLOX algorithm on the blood cell data set is 92.5%, which is 2.4% higher than that of the YOLOX algorithm, and the model parameters are reduced by 8%; The detection accuracy of the algorithm on the COCO2017 general data set is 41.7%, which is 1.2% higher than that of the original YOLOX algorithm.