Abstract:Cerebral palsy is the most common cause of walking dysfunction in children. Lower limb exoskeleton has been maturely used in the rehabilitation of adult spinal cord injury or stroke patients. However, there are relatively few studies on the improvement of gait and walking efficiency in children with cerebral palsy. To promote researchers′ understanding of the lower limb exoskeleton for children with cerebral palsy from the technical and clinical perspective, this article firstly analyzes the special problems faced by the research of exoskeleton for children with cerebral palsy. Secondly, according to the number of lower limb driving joints, this article classifies the lower limb exoskeleton of children with cerebral palsy and reviews its research status, focusing on its mechanical characteristics, control strategies and clinical manifestations. The advantages and disadvantages of children exoskeletons for cerebral palsy with single and multi-joints were analyzed. Finally, the development trend and challenges of lower limb exoskeleton for children with cerebral palsy are discussed.

Abstract:Search and rescue robots are important detection instrument platforms. To address the poor environmental adaptability of existing rigid-structured search and rescue robots, a design method for the search and rescue robot with crab-inspired rigid-flexible coupling mechanisms is proposed in this article. Firstly, the robot mechanisms are designed. Secondly, the lateral motion gaits of the robot are designed based on crab motion experiments, and the feasibility of the gait design is verified through simulations. Thirdly, the software and hardware systems of the robot are designed. Finally, experiments are implemented on the robot′s walking ability on multiple terrains. Moreover, tests of functions, including dark light environment adaptability, human detection, and obstacle avoidance capabilities, are carried out. The results show that this small-sized robot with simple control can realize basic motion functions, the maximum velocity and the minimum power consumption of the robot during the lateral motion is about 3. 4 cm/ s and 8. 6 W, and the angle range of the turning motion is ±90°. Furthermore, the robot adapts well to different terrains and environments and can complete search and rescue functions like environment perception and human detection.

Abstract:A kinematic modeling and control method for the cable-driven supernumerary robotic limbs is proposed to address the problems of complex motion modeling and low control accuracy caused by its wiring form and drive coupling. The kinematic model of the rigidflexible supernumerary robotic limbs is formulated, which is based on the ( denavit-hartenberg, D-H) method and the Euler transform principle. The active decoupling model between adjacent joints is derived based on the joint coupling mechanism, and a supernumerary robotic limbs control strategy based on the drive decoupling kinematics model is proposed. Finally, an experimental prototype of a supernumerary robotic limbs is established, and its motion model and control method are evaluated. The results show that the maximum multi-point positioning error of the robot end is 7. 45 mm, the maximum moving path error is 7. 24 mm, and the average value of the overall error is 6. 08 mm, which verify the correctness of the decoupling kinematics model and control strategy of the proposed rigid and flexible supernumerary robotic limbs with good control accuracy and motion quality.

Abstract:A control strategy based on 3D vision to guide the robot for spraying parts hanging is proposed because the hanging chain transporting spraying parts is difficult to be strictly positioned, which results in its low hanging efficiency. The MASK of the sprayed part is obtained by training and inference through the instance segmentation network Mask-RCNN, and the color and depth maps are obtained after pixel alignment and instance segmentation. After the hand-eye calibration of the robot vision system, the robot joints are further designed to move smoothly by means of five polynomial interpolation, and the robot is controlled and guided to hang the sprayed parts. The experimental results show that the average error of the position angle of the sprayed parts is not more than 10°, and the average error value is 6. 83 mm in the Z-movement direction, with a minimum of 0. 02 mm, and the robot can be guided to realize the autonomous hanging of the sprayed parts in the simulation environment and on-site environment.

Abstract:Industrial robots play an essential role in intelligent manufacturing. The repeatability of the end effector is an important indicator to measure the robot′s ability to complete precision operation. In this article, a theoretical model of the robot end effector′s pose repeatability accuracy measurement is proposed, which is based on directional cosine. A pose deviation detection method is designed, which is based on the improved Gaussian curve fitting (IGCF) algorithm and cubic spline fitting-Pareto particle swarm optimization-echo state network (CSF-PPSO-ESN) algorithm. The measurement of the robot end effector′s pose repeatability accuracy is realized by acquiring the cross-laser pattern′s offset angle and center point position. The experimental results show that the position measurement error after compensation is ±1. 5 μm, and the angle measurement error after compensation is ±2 arc-sec. The proposed pose deviation detection method provides a reference for the online real-time measuring of the robot end effector′s pose repeatability accuracy.

Abstract:To enhance the detection ability of unmanned aerial vehicles (UAV) in ground camouflage targets, this article proposes a multi-scale cross attention improved single machine ground camouflage target detection and localization method. Firstly, a multi-scale cross attention module is designed to enhance cross attention based on the original multi-scale pyramid. The ability to distinguish the boundaries of camouflaged targets is enhanced. Secondly, an open-source drone target detection and positioning system is established, which integrates data such as drone carrier positioning modules, inertial navigation sensors, and optoelectronic pods to calculate the spatial position of the target image after obtaining its position. Finally, a jungle camouflage dataset is constructed and validated through relevant experiments. The results show that the method has a ground target detection accuracy mAP of 70. 2% in typical camouflage scenarios, which is 5. 7% higher than before improvement. It can effectively output the azimuth distance between the target and the UAV, and the average operating efficiency of the algorithm can reach 29. 4 fps, which can meet the real-time requirement of UAV ground target detection and positioning.

Abstract:A long-term tracking method based on the weighted online sample updating is proposed to address the problem of tracking failure caused by target loss during long-term tracking. First, the ResNet50 network is used to extract the deep features of the target and enhance the initial frame sample to optimize the target model, which could improve the influence of the initial frame sample weight. Then, the target model is used to classify the test frame sample, and the confidence score is used to weight the online learning samples to enhance their quality and improve the classification performance of the model. Secondly, the target state is determined by the confidence score, and the target is tracked and located. When the target is lost, a spatiotemporal constraint search is used to adaptively expand the search area at the loss point and randomly search for the target, while utilizing online learning to quickly optimize the target model and enhance its search ability. Finally, an adaptive threshold discrimination method is proposed for the search process, fully utilizing the image background information, using the background confidence score when the target is lost as the discrimination threshold, reducing the influence of similar backgrounds in the search process to accurately retrieve the target. Experiments on the LTB50 dataset show a success rate of 66. 1% and a tracking F-score of 64. 4% , outperforming other methods. Real-world experiments on a quadruped robot platform achieved success rates of 87. 8% and 85. 8% under full occlusion and out-of-view scenarios, respectively. The effectiveness of the proposed method is evaluated.

Abstract:To address space complexity and time difference of human action, an adaptive recognition method of human skeleton action with spatial-temporal tensor fusion is proposed. Firstly, the spatial-temporal feature tensors of adjacent frames are established by making full use of the intra-frame spatial relationship and the inter-frame temporal relationship of the human action skeleton sequences. Secondly, the difference of spatial-temporal feature tensors of adjacent frames is calculated to achieve the spatial-temporal feature tensors of key adjacent frames and compose the behavior spatial-temporal feature tensors. Then, the spatial feature difference of the action spatial-temporal feature tensors and the multi-scale temporal convolution is used to construct the adaptive attention mechanism of the behavior spatio-temporal feature tensors to complete the fusion of action spatial-temporal features. Finally, a deep stochastic configuration network is used to recognize human action according to the spatial-temporal feature fusion tensor of action. The NTU RGB-D data set was used for experimental simulation, and the recognition accuracy reached 84. 57%. The corresponding system is designed for practical application verification. The results show that the proposed method is suitable for dealing with the space complexity and time difference of human action recognition.

Abstract:One of the main contributors to water pollution is heavy metal ions, which is dangerous for both human health and the environment even in tiny amounts. Heavy metal ions have been discovered to be biomarkers for several diseases and pathological situations in bodily fluids like sweat and serum. Therefore, in ecological environmental supervision and precision medicine, the quick and accurate detection of heavy metal ions plays a critical early warning role. This article firstly compares the widely used detection methods at present and then summarizes the heavy metal ion sensors developed based on MEMS technology, which can be divided into UV-visible spectrophotometry, colorimetry, fluorescence detection, luminescent chemistry, electrochemical method, and cantilever beam detection method according to the detection principle. Based on the aforementioned detection methods, the present portable instruments and systems developed are summarized, and the development trend and application prospect of heavy metal ion detection sensors and systems. Keywords:h

Abstract:Graphene with atomic thickness has advantages of excellent electromechanical properties and super-large specific surface area. Combined with the significant piezoresistive effect, graphene opens up window for next generation of the ultra-sensitive pressure sensor. The main cause of the low-yield for graphene sensor made by the conventional fabrication process is analyzed to be the overloaded stress during its release process from the liquid. To solve this problem and improve the yield, a new pressure sensor scheme by using polymethylmethacrylate (PMMA) / graphene composite heterogeneous film (hetero-film) is proposed to replace the atomic graphene film. The corresponding fabrication process compatible with the traditional COMS process is designed and applied to sensor fabrication, which reaches a much higher yield and holds the hope for large-scale fabrication. The pressure sensing test shows that the sensitivity of the new graphene pressure sensor is up to 7. 42 × 10 -5 / kPa, which is comparable to the existing research. However, the extracted sensing precision is low to 2. 6% ~ 3. 5% in the full-scale range (FS), which is nearly 2 orders worse than the silicon-based high-performance pressure sensor (0. 05% ~ 0. 01% FS). The electrical noise of the measurement system and the intrinsic resistance noise of the polluted graphene may be the key root. This work suggests that more research on graphene pressure sensor should focus on the improvement of sensing precision indicator rather than the pursuit of sensitivity.

Abstract:The key functional units — displacement sensors of digital high-precision measuring instruments and equipment rely heavily on high-precision machining. To address this issue, a novel displacement sensing principle based on a combined measurement method is proposed. The alternating magnetic field produced using the uniform distribution of the excitation winding on the plane is utilized to establish a moving reference system. The mapping relation is obtained between the measured displacement and the time reference. Magnetic field is restricted by precisely controlling the shape of the induction winding to suppress harmonic errors in principle. The differential arrangement of the induction winding and a combined measurement method are employed to improve its anti-interference ability and the measurement precision. The measurement error is analyzed by using the electromagnetic simulation to optimize structural parameters. The sensor prototype is produced and tested in experiment. The results show that the error is ±2. 25 μm within the range of 144 mm, and the resolution is 0. 15 μm. Differing from traditional high-precision displacement sensors relying heavily on high-precision lithographic machining, the proposed method makes full use of precisely restricted and innovation in sensing principles to perform highprecision displacement measurement with the feature of simple in structure and low cost, which has the significant academic value and the high application value.

Abstract:It is difficult to balance the high signal-to-noise ratio and high time interpolation resolution of the electromagnetic linear time gate displacement sensor developed in the early stage. To address this issue, a new sensor structure for improving the signal to noise ratio of the sensor is designed, and a new measurement method with high signal-to-noise ratio and high time interpolation resolution is proposed. A linear time gate displacement sensor based on stratified coupling of air gap magnetic field is developed. The mathematical model of the sensor air gap magnetic field is formulated, the spatial distribution characteristics of the air gap magnetic field are analyzed, and the principle of hierarchical coupling of the air gap magnetic field of the plane coil is studied. According to the principle of stratified coupling of air gap magnetic field, a stratified coupling displacement measurement model of the sensor air gap magnetic field is established. Electromagnetic field is simulated and error of sensor measurement model is analyzed. Finally, an experimental platform is built to test the performance of the sensor. The experimental results show that the structure of layered coupling of air gap magnetic field improves the signal-to-noise ratio of the sensor, and the measurement accuracy of the sensor is improved by 31. 4% on the original basis. The high signal-to-noise ratio and high time interpolation resolution measurement method are adopted. The measurement accuracy of the sensor is improved by 37. 3% on the original basis.

Abstract:The pedestrian dead reckoning (PDR) algorithm based on the mobile phone inertial sensors is one of the core methods for pedestrian navigation. However, due to factors such as the sensor noise, the positioning error of dead reckoning accumulates over time, which may lead to the divergence of the pedestrian positioning. To compensate the positioning error and improve the positioning accuracy of the PDR method, the GNSS positioning is generally introduced and combined with the conventional PDR method via the Kalman filter. In this article, a pedestrian collaborative positioning method based on the factor graph optimization is proposed. The state transition, measurements and collaborative ranging information are all used as state constraints. In addition, the optimal estimation is performed uniformly. To evaluate the performance of the method, experiments are implemented in both open-sky area and GNSS denied environment. The experimental analysis results show that the pedestrian collaborative positioning method based on the factor graph optimization can effectively improve the positioning accuracy both in open-sky and GNSS degraded area. Compared with the cooperative method based on Kalman filter, the maximum horizontal positioning error is reduced by more than 30% .

Abstract:The traditional detection methods based on a single laser displacement sensor are low in efficiency and greatly affected by axial shaking. To address this issue and solve the problems of high-precision and rapid detection of the roundness of the inner cavity section of ultra-long and deep-hole pipelines, this article proposes a point-based detection scheme based on two laser displacement sensors. By formulating mathematical models and numerical simulation, the eccentricity parameters of the rotating shaft of the detection device and the installation deviation parameters of two laser displacement sensors are simulated. The influence of each parameter on the evaluation results of the roundness of the deep hole pipeline is analyzed. On this basis, a mathematical correction model of two laser displacement sensors with installation errors is proposed, and an experimental system for pipeline roundness detection is established to evaluate the effectiveness of the model. Compared with the direct roundness evaluation of the data collected by the two laser displacement sensors, the results show that the roundness evaluation of the corrected data decreases from 0. 30~ 0. 50 mm to 0. 05~ 0. 15 mm, and the measurement time is shortened from 18. 7 s to 9. 8 s.

Abstract:High temperature strain gauge is an important research assurance for aero-engines. To meet its higher use requirements, the structural parameters of high temperature strain gauge sensitive grid are optimized. Firstly, the influence of wire length, wire spacing and wire bending number on the measurement error and fatigue life of sensitive grid is analyzed by the finite element method considering the temperature error. Secondly, the response surface model of measurement error and fatigue life is formulated, which is based on response surface methodology. The multi-objective gray wolf algorithm is used to perform the structural parameter optimization, and the Pareto optimal solution set is obtained. Finally, according to the optimization results, a high temperature strain gauge is prepared for high temperature vibration fatigue test. The simulation results show that the sensitive grid structure with different parameter combinations has different effects on the performance of the high temperature strain gauge. The Pareto optimal solution set is evaluated by comparing the production requirements and performance of the high temperature strain gauge, and an optimal sensitive grid structure is obtained. The fatigue life of the optimized strain gauge at high temperature is increased by 30. 4% compared with that before optimization, and the optimization effect is remarkable.

Abstract:The open service capability of 5G has changed the industrial internet data service process. When migrating industrial businesses to 5G network, the coordination of cloud-based data processes and the conflict avoidance of networked services should be considered. Taking power industry for example, to ensure data reachability and availability, a data service model based on Petri Net is proposed for unified modeling and analysis of business temporal logic and network functions. Then, based on reachability graph, VIKOR orchestration scheme is adopted to ensure the deterministic process and execution time of multiple data from space-time dimension. Simulation results show that it can avoid data conflicts after migration and improve deployment efficiency. Field test of power 5G experimental network verifies that this method can guarantee transmission delay of 12. 03~ 18. 35 ms and delay jitter of 2. 75~ 5. 62 ms of the bearer layer, the completion time of 5. 96~ 6. 68 s for power remote control business, and the data quality has been improved to meet the operational performance requirements of the power telecommunication network.

Abstract:To address problems of large distribution difference in rolling bearing vibration data under different working conditions, difficulty in obtaining labeled vibration data under certain working conditions, the non-sharing of data among different users and the small amount of single user data, which lead to the low accuracy of the established diagnosis model, a federated feature transfer learning framework and a fault diagnosis method of rolling bearing under different working conditions based on the federated multi-representation adaptation are proposed. The time domain vibration data of rolling bearings are transformed by wavelet transform and the time-frequency spectrum can be obtained. The priori labeled public data are used as the source domain and the multi-user unlabeled privacy silos data are used as the target domain. A multi-representation feature extraction architecture is introduced to improve the original residual network, multi-representation features of source domain and target domain are extracted, and multi-user local models are constructed respectively. To enhance the security of the federated framework and reduce the communication overhead, the deep neural network model compression idea is used to improve the parameter transfer strategy in the federated transfer learning framework. A federated global model for rolling bearing fault diagnosis under different working conditions is formulated on the server side. On two bearing datasets, experimental results show that the proposed method can integrate soils data knowledge without multi-user sharing data, and establish an effective fault diagnosis model of rolling bearings under different working conditions, the average fault diagnosis accuracy canreach 97. 6% , which is at least 3. 2% higher than the single user modeling. which has high accuracy and strong generalization.

Abstract:Pressure vessel gas leakage intelligent detection and identification techniques are susceptible to interference from a variety of factors, and intelligent detection models require a large amount of monitoring data training. In the actual industrial environment, available data, especially data labels, are very scarce. To address the problems such as interference from multiple working conditions and lack of labeling information of data, this article proposes an unsupervised variable working condition intelligent detection technique by using transfer learning. Firstly, samples of multiple leaks are collected in laboratory environment and select three different pressure working conditions to divide the data into labeled source domain and unlabeled target domain. Secondly, a convolutional feature extractor is designed to propose an improved joint distribution adaptation mechanism for the edge distribution and conditional distribution of the two domains, and further improve the distribution difference metric to enhance the neighborhood confusion. Experimental results on six transfer learning tasks validate the effectiveness of the method, with higher accuracy than the classical domain adaptive algorithm.

Abstract:To address the problem of micro-cracks detection on the surface of cold-rolled rolls, this article proposes an alternating current magnetic flux leakage detection method excited by double-layer flexible parallel cables, which is used to profile the working surface of the rolls, and uses the high-precision tunnel magnetoresistance ( TMR) sensor array as the detection sensor. It has the ability to detect micro-cracks on large curvature surfaces. Meanwhile, by adjusting the excitation current of the double-layer PCs to offset the tangential component of the background magnetic field, the excitation current is increased further and the TMR overrange saturation problem is avoided. The tangential component of the magnetic field is used to characterize defects, enabling it to characterize multidirectional cracks in a single scanning direction. Based on the analysis above, this article establishes a three-dimensional finite element simulation model of double-layer PCs and roll, and studies the influence of roll radius, defect angles and depth on AC MFL signal through simulation. Finally, a corresponding verification experimental platform is established. Experiment shows that the probe can effectively magnetize the surface with large curvature, detect and distinguish multidirectional cracks, and detect micro-cracks at a depth of 15 μm.

Abstract:Ultrasonic defect detections are easily disturbed by complex noise in ultrasonic echo signals. To improve the accuracy of ultrasonic defect detection, an ultrasonic echo signal noise elimination method based on the hybrid decomposition is proposed. The fusion of empirical mode decomposition and correlation coefficient is used to preprocess the ultrasonic echo signal, and the preprocessing signal is obtained to eliminate the low-frequency noise components. Based on variational mode decomposition, the preprocessed signal is decomposed into several band-limited intrinsic mode functions ( BLIMFs), and the mutual information is introduced to estimate the optimal mode number. The useful modes are extracted according to the correlation coefficient of the BLIMFs to the preprocessed signal, and the denoising results are reconstructed. The denoising performance of the proposed method is evaluated by simulation and measured ultrasonic echo signals. Compared with two existing methods, experimental results show that the proposed method can simultaneously eliminate high-frequency and low-frequency noises in ultrasonic echo signals. Under different SNR conditions, the mean SNR of the EMD, VMD and the proposed method are 10. 01 dB, 9. 48 dB, and 16. 09 dB, respectively, which proves the superiority of the proposed method for noise elimination of the ultrasonic echo signals.

Abstract:To address the problems of weak feature extraction ability, lower positioning accuracy and poor robustness of the feature pointbased visual simultaneous localization and mapping ( SLAM) algorithm in low-texture environment, this article proposes a featureenhanced visual SLAM algorithm based on the sparse direct method. Firstly, the image sequence is preprocessed to improve the feature extraction ability of the algorithm. Then, the pose is solved by combining the sparse direct method based on graph optimization and the feature point method. The operation efficiency and robustness of the algorithm are improved under the premise of ensuring the positioning accuracy of the algorithm. The experimental results of the TUM data set show that the positioning accuracy of the proposed algorithm is better than those of the current SLAM algorithms. In the scenario with sparse texture in the TUM data set, the number of feature points extracted by the algorithm is 9. 6 times more than that of the ORB-SLAM2 algorithm, and the average number of points per frame tracking time is reduced by 58% .

Abstract:The review of aero-engine compressor blades is an indispensable part of its development cycle. To improve the traditional manual blade review process, which is time-consuming, laborious, and highly uncertain, an engine blade profile detection system based on computer vision is designed. Firstly, to achieve the blade measurement of three-coordinate detector commonly used in industrial production, the batch image extraction for portable document format is completed. The out-of-tolerance judgment of blade image is completed by using color matching and Hough transform. Secondly, for the blade image within tolerance, color matching and morphological operators are used to enhance the blade image, which improves the ratio of valuable information. A residual network is trained to complete the task of morphological anomaly detection of the blade edges. Finally, to facilitate the labeling task on a massive image dataset, a universal image classification and labeling program is designed, and a blade quality detection program is designed to verify the effectiveness of the system for blade out-of-tolerance judgment and anomaly recognition on the blade image dataset. The experiment shows that the accuracy of the system for identifying anomalies in blades with or without out-of-tolerance reaches 100% and 92. 9% , respectively, which could satisfy actual needs of industrial production.

Abstract:Traction wire rope (TWR) plays an important application value in large-scale industrial lifting equipment. While using the traction wire rope for operation, it is necessary to regularly diagnose the defects of the traction wire rope to avoid safety accidents. The traditional method is manual visual inspection, but this method has long detection time and low efficiency. Therefore, this article proposes a network algorithm for detecting traction wire rope defect ( TWRD) based on YOLOv5 improved network, abbreviated as TWRD-Net. In order to facilitate deployment on industrial equipment with low computational power, a lightweight LW-C3 module is first designed to reduce the model′s parameter count and computational overhead. Secondly, the PAN structure is improved by designing a CLW-FPN structure to enhance the model′s sensitivity to defect semantic information extraction and defect location information. Finally, this article designs β-CIoU loss function. Compared with CIoU Loss function, β-CIoU reduces the loss of bounding box regression and further improves detection accuracy. This article establishes a TWRD dataset and conducts experiments by using TWRD-Net. The experimental results show that the accuracy of the proposed TWRD-Net can reach 98% , mAP can reach 99. 4% , and frame rate can reach 151 fps. Compared with other mainstream detection model experimental results, it has the advantages of high accuracy, small size, and fast detection speed, which can provide reference for industrial equipment quality inspectors.

Abstract:Due to the impact of high-speed airflow impact under complex operating conditions in the wind tunnel, the wing produces dynamic bending and twisting deformation. This article proposes a fast measurement method that uses a single-double mixed stereo vision method, which can achieve the bending and twisting deformation measurement of the wing in the wind tunnel by using fluorescent markers. Firstly, the layout and computing model of the wing′s bending and twisting deformation is proposed. Then, a camera coordinate system is established to calibrate cameras based on the relative relationship of the wind tunnel coordinate system and the aircraft model′ s own coordinate system. Finally, a single camera 3D reconstruction model is constructed according to the wing points with known Y-axial constraints. Therefore, the dynamic bending and twisting deformation can be calculated under blowing condition. This method can realize the fast dynamic measurement of the wing bending and twisting deformation by establishing a special layout of dynamic wing deformation system. Based on the initial position of the aircraft model, the average error of the obtained twisting angle is 0. 228 °. It can effectively solve the problems of reconstruction with high frame rate in binocular cameras and provide data support for optimizing the design parameters in aircraft.

Abstract:For single photon counting imaging technology to detect weak target signal, low signal-to-noise ratio, and unclear target area of the obtained image, serious background noise and other problems, a single-photon detection system based on solar-blind ultraviolet is designed by using 270±5 nm solar-bind UV filter, microchannel plate image intensifiers (MCP) with irnage gain >10 5 and scientificgrade complementary metal oxide semiconductors (sCMOS) with a mazimum resolution of 1 504×1 504. The system uses time sequence control to obtain single-photon spot images. To highlight the target area in the image, the improved morphological top-hat transformation algorithm is used to enhance the spot target area. Then, the image is binarized by the triangular threshold method, and the coordinates of the target area are extracted by using the connected domain. Finally, the area extremum algorithm is used to count single photons in the target area of the original image. The article conducts a series of imaging with a single exposure time of 80~ 100 ns and data processing experiments on ultraviolet light sources. The results show the feasibility of the designed single-photon imaging detection system and photon counting algorithm.

Abstract:3D object detection based on point cloud is essential in many applications, such as robotics, autonomous driving. LiDAR point clouds contain reliable geometric information for 3D scene understanding. However, due to sparsity and occlusion, point clouds depict only partial surfaces of objects, which severely degrades the detection performance. To handle this challenge, we propose a novel twostage detector based on symmetric shape generation ( SSG-RCNN). The shapes of 3D interested objects are roughly symmetric. In the first stage, SSG-RCNN predicts a symmetric point for each foreground point to complete objects shapes while generating 3D proposals. In the second stage, SSG-RCNN utilizes self-attention pooling module to aggregate proposal-wise features from raw points and symmetric points. Finally, proposal-wise features are used to refine 3D proposals. Extensive experiments on KITTI benchmark show that SSGRCNN has remarkable detection performance. Especially for hard difficulty level objects, SSG-RCNN achieves 77. 64% AP on the KITTI test set, which is better than previous state-of-the art methods.

Abstract:To address the problems of single capacitance feature extraction scale and low utilization of intermediate layer features in the image reconstruction process of electrical capacitance tomography based on deep convolution neural network, a multi-scale adaptive feature aggregation network model is proposed for electrical capacitance tomography image reconstruction. Firstly, a feature enhancement module (FEM) is designed by using stacked enhanced selection kernel convolutional module, which adaptively extracts feature information from multiple scales of the capacitance vector by concatenating multiple FEM. The artifacts caused by using ordinary convolution is reduced. Secondly, a feature aggregation mechanism is introduced, which uses long and short residual connections to enhance the correlation of far and near feature information. The problem of insufficient utilization of middle layer features in the network is solved. Compared with traditional algorithms and CNN algorithm, the experimental results show that the proposed method has better performance in subjective visual effects and objective evaluation indicators, with the highest image correlation coefficient reaching 0. 962 9 and the relative error of the image reduced to 0. 053 0.

Abstract:Electrochemical impedance spectroscopy (EIS) can reflect the electrochemical parameters inside the energy storage cell more accurately. To solve the time-consuming problem of traditional measurement methods, a method of measuring impedance information of energy storage battery with broadband excitation signal is adopted and validated by simulation based on MATLAB Simulink platform and EIS measurement test platform. This method injects a broadband AC signal into the battery. Fast Fourier transform decomposition of broadband AC signals is utilized to obtain battery impedance information at each frequency point to significantly reduce the measurement time of electrochemical impedance spectroscopy. In this article, two broadband excitation methods are used to measure the battery equivalent circuit model and the energy storage battery. Compared with electrochemical workstation sweep measurement methods, measurement time can be saved by 60% and 78% , respectively, and the measurement result error is small. The rapid measurement of battery EIS with broadband excitation measurement method can provide technical support for broader application scenarios such as realtime diagnosis and on-line inspection of battery management systems.

Abstract:The existing magnetic field energy collector standard energy management circuit has advantages of low efficiency, high power consumption, and high input threshold power. When the transformer collects weak magnetic field energy, it is difficult to drive the multicore cable monitoring system. To address the above issues, this article proposes a low-power and multi-core cable weak magnetic field energy collection management circuit principle of intermittently charging discharge. A low-power intermittent control circuit for selfpowered power supply is designed to greatly reduce power consumption. This circuit is comparable to the energy conversion efficiency of the management circuit through self-power supply variable frequency matching. The experimental results show that when the three-core cable passes the AC power of 40 A and 50 Hz, the output power of the magnetic field energy collector is 3. 3 mW, the maximum output power of the management circuit can reach 2. 45 mW, the maximum efficiency is 74. 24% , which is higher than the standard energy collection management circuit 6. 9 times. The proposed high-power self-power supply control circuit to meet the minimum average power consumption of the startup threshold is only 1. 52 μW, which is only 5. 93% of the control circuit power consumption with auxiliary power supply. This high-efficiency management circuit can be used not only for the weak energy collection of multi-core cables, but also for weak emotional magnetic field energy management such as the weak cable peripheral magnetic field, electromagnetic vibration energy collector, and reverse scattering sensor network caused by armor shielding.