Abstract:In response to the problem that rolling bearing vibration signal characteristics were difficult to be extracted in the case of strong noise, a method based on composite spectral kurtosis to optimise the variational modal decompositionwas proposed. First, the original fault signal was subjected to variational modaldecomposition, and several intrinsic mode functionswere acquired by optimizing the key parameters of VMD-modal numberand penalty factorrespectively with the principle of the maximum value of composite spectral kurtosis. Then, the kurtosis of each IMFwas calculated, and the component with the maximum kurtosis value was selected as the optimal IMF. Finally, the Hilbert transform was performed on the optimal intrinsic modal function to obtain their envelope spectra, so as to realize the extraction of the fault eigenfrequency. Through the analysis of the public dataset and the relevant data of the homemade test bed, it is shown that the proposed method can effectively extract the fault characteristics of the fault signal under the background of strong noise and realize the discrimination of the fault type.

Abstract:To fully leverage the business application capabilities of standard format radar PUP products, the visualization application system of PUP products of standard format weather radar is designed and developed. The overall architecture of the system, as well as solutions for the collection, decoding, processing, and sharing of standard format radar PUP products, were researched and designed. Considering the large volume of data characteristic of standard format radar PUP products, an improved RLE compression algorithm and an optimized radar image rendering method were proposed to effectively enhance the efficiency of product image display and achieve efficient visualization applications of standard format radar PUP products. Based on the B/S architecture, the system′s functionalities were implemented using mature technologies such as SpringMVC framework, HTML5, CSS, and WebGIS. It provides rich radar PUP product overlay display and sharing applications with GIS data for users across the province. Additionally, it offers layered overlay applications of ground observation data, lightning data, and radar PUP products. Upon deployment, it received positive feedback from meteorological business users, effectively enhancing the forecasting and warning capabilities for severe convective weather and hazardous weather events. The system has achieved early implementation of the standard format radar PUP product in the operational applications of meteorological departments nationwide, providing valuable reference for the visualization standard format radar PUP products in other provinces. It has broad application value for promotion.

Abstract:A proportional resonance selfimmunity control strategy is proposed to solve the problem of a large number of low-frequency harmonics caused by the nonlinearity of the drive motor inverter and the non-sinusoidal waveform of the back electromotive force. This strategy can suppress the current harmonics more comprehensively, while the introduction of resonance control can provide better suppression of specific frequency harmonics. A mathematical model of the motor system is established. Based on the Maxwell tensor method, the analytical equation of the electromagnetic force is deduced. It is analyzed that the 5th and 7th harmonics will deteriorate the performance of the motor in terms of torque pulsation and electromagnetic noise. A multi-physical field co-simulation model using Simulink and Jmag is established. Simulation analysis is conducted to validate the theoretical analysis and the effectiveness of harmonic suppression in reducing torque pulsation and electromagnetic noise.An experimental platform is set up to analyze the current and electromagnetic noise results before and after applying the strategy. The results indicate that the control strategy constructed has a better suppression effect on the harmonic components of the main order of low frequency, and optimizes the low frequency noise characteristics of the motor.

Abstract:In various applications of mobile robotics, such as automated warehousing logistics scenarios, due to the limitation of lidar installation location. The adoption of a single LiDAR for Simultaneous Localization and Mapping (SLAM) introduces challenges pertaining to restricted field of view and complexities in achieving loop closure. In response, we proposes a multi-LiDAR localization and mapping methodology incorporating tight coupling with an Inertial Measurement Unit (IMU), building upon the FAST-LIO2 algorithm. This approach not only expands the perceptual range of the robot but also enhances localization precision and mapping effect. Through rigorous evaluation via offline tests utilizing public datasets and online experiments conducted on the experimental platform, the proposed algorithm demonstrates marked enhancements in localization accuracy and mapping effect compared to the M-LOAM and FAST-LIO2 algorithms, concurrently exhibiting reduced loop closure drift.

Abstract:Electrical stimulation technology has significant application value in clinical rehabilitation of motor function, and the development of advanced electrical stimulation systems is crucial for achieving precise and efficient neuromuscular electrical stimulation. This work combines digital signal synthesis and a constant current source circuit model to design a multi parameter adjustable electrical stimulation system with small output current error. The system can output three waveforms: square wave, triangular wave, and sine wave. The output current error is less than 0.5%, and the maximum output impedance is 4 000 Ω. The frequency error in the range of 10~500 Hz is less than 0.5% for square wave, less than 1% for triangular wave, and less than 3% for sine wave. Compared with KT-90A and PE1-2 medical grade electric stimulators, the system still maintains high current accuracy at a minimum output impedance of 2 000 Ω, and the waveform is not distorted. The system was applied to neuromuscular functional electrical stimulation experiments, and the effectiveness of the system in clinical rehabilitation applications was verified by analyzing the waveform and energy spectrum of EEG waves. This work is expected to provide technical support for the clinical application of electrical stimulation rehabilitation intervention.

Abstract:This paper presents a novel approach to the design of a 4×4 slot array antenna operating at 5.8 GHz, utilizing a slot element based on the Koch Snowflake hexagonal fractal structure. The antenna is fed by a parallel feeding network, resulting in characteristics of high directivity. The proposed design was experimentally validated through fabrication and testing. Measurement results indicate a 5.7% impedance bandwidth (5.56~5.89 GHz) with well-matched impedance. At the operating frequency, the antenna exhibits commendable directional radiation characteristics and stable gain, achieving a peak gain of 19.85 dBi and a corresponding aperture efficiency of 80.69%. Moreover, the 3.dB gain bandwidth extends to 18.40% (5.13~6.20 GHz), demonstrating the efficacy of the Koch Snowflake fractal structure in enhancing the performance of slot array antennas.

Abstract:Stacked plate are counted by hand, which takes long time and has poor accuracy. Hence, the paper proposes a plate counting instrument based on embedded platform with a lightweight model. The instrument can detect in real time the number of stacked plate at production and logistics site, which deploys the improved Faster R-CNN network to the Industrial Personal Computer.In order to alleviate the difficulty of small object detection, the network algorithm by using lightweight network MobileNetv2 to integrate the efficient channel attention as the backbone network, using spatial attention and inverted residual structure module to reconstruct the FPN structure, proposing an HIOU_Loc algorithm based on on Height intersection over union to remove redundant prediction boxes. The plate counting experiment on a IPC equipped with N4100 CPU. The results show that the accuracy of the plate counting algorithm proposed in this paper reaches 98.51%, and it only takes 0.31 s to detect a high-resolution plate image. A quantitative calibration module is designed for the instrument. The instrument can reach 100% accuracy in counting stacked plate after the manual calibration module, which meets the requirements of stacked plate real-time counting in practical scenarios.

Abstract:To address the issue of low vehicle detection accuracy in road surveillance, this paper proposes an improved vehicle detection method based on YOLOv7. Firstly, we introduce the Efficient Multi-Scale Attention Mechanism (EMA) for cross-space learning to enhance attention to feature information. Secondly, we replace the SPPCSPC module in the neck network with the SPPFCSPC module, trim the CBS layer, and introduce the EMA attention mechanism to strengthen attention to small target areas, thereby obtaining more accurate vehicle features. Additionally, we incorporate the EMA attention into the MP module to fuse more important feature information. Finally, employing the MPDIoU loss function accelerates model convergence and enhances detection accuracy. Experimental results show that the improved YOLOv7 achieves a detection accuracy of 86.69%, which is a 2.83% improvement over the original YOLOv7 network. This enhancement effectively boosts the accuracy of vehicle object detection, providing assurance for applications such as road video surveillance.

Abstract:The accurate prediction of PV power is very important for the safe and stable operation and real-time control of the integrated energy system. In order to solve the problems of noise interference in photovoltaic power prediction and poor prediction accuracy of traditional single prediction model, a short-term photovoltaic power prediction model based on ICEEMDAN and TCN-AM-BiGRU is proposed. Firstly, the Pearson correlation coefficient was used to screen the key meteorological factors, and the historical PV power data were divided into three similar days: sunny, cloudy and rainy by fuzzy C-means clustering. Secondly, ICEEMDAN is used to decompose the historical training set into several regular subsequences and reconstruct them according to the permutation entropy. Finally, the sequence features are extracted by TCN, the attention mechanism is introduced to assign different weights, and then the prediction is made by BiGRU to output the final prediction result. Taking the actual data of a photovoltaic power station as an example, the prediction model and other models were verified and analyzed. The results showed that in sunny, cloudy and rainy weather, compared with other comparison models, the accuracy of the proposed model increased by 1.69%, 3.58% and 4.40% on average, the MAE decreased by 57.61%, 36.83% and 40.94% on average, and the RMSE decreased by 56.90%, 34.30% and 36.63% on average, which verified the effectiveness and superiority of the proposed model.

Abstract:Lidar point cloud segmentation technology plays an important role in intelligent vehicle environment recognition. Due to the problems of near dense and far sparse point clouds, uneven distribution, and the presence of noise in LiDAR, inaccurate point cloud segmentation occurs. A self adaptation DBSCAN with Euclidean joint clustering algorithm is proposed to address the above issues. This method first preprocesses the point cloud data, using through filtering, voxel filtering, and cube filtering to extract, sparse, and denoise the point cloud. Then, it combines the adaptive DBSCAN algorithm and an improved variable threshold Euclidean clustering algorithm to cluster and segment the point cloud. Real scene data was collected for testing, and the results showed improvements in evaluation indicators such as C-H coefficient, contour coefficient, D-B coefficient, and contour coefficient. This indicates that the variable threshold joint clustering algorithm significantly improves the accuracy of point cloud segmentation, effectively improves the intra class consistency and inter class differences of clustering results, and provides a more reliable foundation for object detection and recognition.

Abstract:With regard to the problem of tracking telemetry and command instruments in accuracy evaluation, put forward an adoption double the satellite navigate a measuring of receiver to control the project of equiping the accuracy acceptance, give accuracy acceptance of general process, deduced in detail a kind of differ from traditional of the true value compute a new method, and provided the concrete calculate way and model of this method.The foundation analyzed in the theory up, adoption calculate the example carry on calculation and analysis and express as a result according to the double the satellite navigates a receiver of measure to control to equip an accuracy acceptance method, can in advance according to measuring the accuracy index sign of controling the material, analysis need the satellite navigates a receiver of index sign, can also provide a metered data to prop up for route design at the same time, save to measure to control the manpower and material resources of equiping the accuracy acceptance, can be measuring to control the related realm expansion application of instruments accuracy evaluation.

Abstract:With the in-depth application of mobile robot in production and life, its path planning ability also needs to develop to both rapidity and environmental adaptability. In order to solve the problems existing in the existing mobile robot path planning using reinforcement learning methods, which are easy to fall into local optimization in the early stage of exploration, repeatedly search the same area, and explore the late convergence rate and slow convergence rate, an improved Q-Learning algorithm is proposed in this study. The algorithm improves the Q matrix assignment method to make the exploration process directional in the early iteration and reduces the collision situation; the Q matrix iterative method is improved to make the Q matrix update forward-looking and avoid repeated exploration in a small area; the random exploration strategy is improved to make full use of environmental information in the early iteration and close to the target point in the later stage. The simulation results of different raster maps show that the algorithm in this paper has higher computational efficiency by reducing the path length, reducing jitter and improving the speed of convergence based on the Q-Learning algorithm.

Abstract:The rapid expansion of the application range of unmanned systems makes the visual perception environment more complex and changeable, which makes it difficult for traditional visual control algorithms to effectively control visual sensors to obtain accurate visual perception images, thus affecting the stable operation of unmanned systems. Therefore, the research on intelligent visual control algorithms based on unmanned systems is proposed. The gray value of the visual perception image of unmanned system is transformed by Gamma curve nonlinear, and the contrast of the image is enhanced by the gray world method. Based on the processed image, the image moment is calculated, namely the space moment, the central moment and the normalized central moment, to describe the global and local characteristics of the image. According to the obtained visual perception information of the unmanned system, the intelligent visual control framework is built. Obtain the desired image feature matrix, extract the current moment image feature matrix, and nonlinear map the camera angle through the extreme learning machine based on the improved firefly algorithm, so as to obtain the intelligent vision control law, so as to eliminate the visual perception image error and realize the effective control of intelligent vision. The experimental results show that under the background of different experimental groups, the minimum average time of visual control obtained by the proposed algorithm reaches 1 s, and the minimum average error of visual control reaches 0.12%, which fully confirms the better application performance of the proposed algorithm.

Abstract:Neural networks have been extensively utilized in various fields, steganography for neural network is a research emerging direction in academia in recent years. Embedding capacity and robustness are important indicators for steganography. But balancing embedding capacity and robustness is challenging. This paper proposes a robust steganography for neural network models. Embedding secret data into neural network without visibly reducing the performance of the original task. This is achieved by embedding secret data during the training process instead of modifying the network parameters after training. Receivers can obtain the secret data from data decoding networks, the parameters of data decoding networks are generated using the embedding keys. In this way, it is unnecessary to transmit the decoding networks secretly. Additionally, introducing reed-solomon codes to improve data extraction robustness. Experimental results reveal that the robust steganography for neural models improves robustness while maintaining superior embedding capacity.

Abstract:Aiming at the difficulty of target matching between two kinds of sensors in the fusion of radar and video data in the process of highway vehicle tracking, a highway vehicle tracking method based on target trajectory similarity matching was proposed. Firstly, the radar data is converted to the dimension of video data by projection transformation. Secondly, curve fitting algorithm is used to interpolate discrete trajectory points into continuous trajectory curves. Finally, the similarity between the trajectory curve projected on the image of the radar detection target and the trajectory curve of the video detection target is calculated, and the matching relationship between the radar detection target and the video detection target is obtained by screening the similarity matrix. Comparative experiments were carried out with vehicle data collected in real scenarios on highways. The results show that the average success rate of target matching in the expressways is 94.71%, which is 3.01% and 3.69% higher than that of other similar methods. The proposed method can effectively filter false targets and is more suitable for vehicle tracking in highway scenarios.

Abstract:Addressing the issues of image blurring and uneven light distribution encountered when capturing images in adverse weather conditions, which lead to decreased scene contrast and subsequently increase the difficulty of distinguishing detection targets from the background in images, this paper proposes an improved YOLOv8s algorithm to enhance the detection capability of vehicles and pedestrians in harsh weather environments. Firstly, based on the YOLOv8s algorithm, this paper optimizes the C2F module in the backbone network with an expandable residual structure, enhancing the model′s adaptability to environmental changes. At the same time, an efficient multi-scale attention mechanism is introduced before the SPPF module in the backbone network, which can more effectively capture the rich and varied multi-scale features in images. Secondly, the detection head of the YOLOv8s algorithm is redesigned to reduce the model′s complexity while maintaining accuracy. Finally, the introduction of Wise-IoU improves the regression loss function of the YOLOv8s algorithm, enhancing the algorithm′s convergence speed and detection accuracy. Experimental results show that the improved YOLOv8s algorithm achieves an mean average precision of 91.41% on datasets for vehicle and pedestrian detection under adverse weather conditions, which is a 2.56% improvement over the original algorithm, with a model parameter reduction of 8% and a computational reduction of 4.9 GFLOPs. Compared to other mainstream object detection algorithms, the significantly improved YOLOv8s algorithm not only ensures real-time performance but also effectively meets the challenging requirements for vehicle and pedestrian detection under adverse weather conditions.

Abstract:The wire rope used in coal mine plays an important application value in mine operation, and its reliability is directly related to the operation efficiency of the mine and the life safety of the staff. Aiming at the problems of low detection accuracy and insufficient detection efficiency of existing wire rope surface defects. This paper proposes an improved YOLOv8 detection algorithm YOLO_BF. Firstly, an improved double-layer link attention mechanism (BiFormer) is introduced into the backbone network to enhance the model ′s ability to analyze images and information fusion, which significantly improves the accuracy of the model. Secondly, the repeated weighted bidirectional feature pyramid network (BiFPN) is embedded to improve the ability of network defect feature extraction. On this basis, WIoU is used to improve the convergence speed of the model. Finally, GhostConv is used to replace the traditional convolution to realize the lightweight of the model. Compared with the original basic network YOLOv8n, the accuracy, recall and average accuracy are increased by 2.3%, 3.3% and 5.2% respectively.It is more in line with the practical application requirements of wire rope damage detection.

Abstract:Although object detection can provide the location, size and category of nearby targets for autonomous vehicles, there are still problems of missed detection and false detection in multi-object detection in dense scenes, so an AD-YOLOv5 vehicle detection model is proposed. Firstly, the C3 module in the feature extraction network is optimized to obtain the C-C3 module using the lightweight structure CBAM attention mechanism, which improves the ability to acquire feature information and reduces the attention to other features; secondly, in the detection head section, the classification and regression tasks are decoupled in order to achieve stronger feature representation; then, the generalized power transform is used to perform the transformation operation on the IoU, and the Alpha-IoU loss function with better robustness is proposed, which improves the detection accuracy of the model and accelerates the convergence speed of the model; finally, to add to the complexity of the sample, the GridMask data enhancement technique was used and experiments were carried out on the processed dataset. The experimental results show that the mean average accuracy of the improved target detection model reaches 72.72%, which is 2.25% higher than the original YOLOv5 model, and the model has a high convergence speed, and the visual comparison experiments intuitively show that the model of this paper can effectively avoid the phenomenon of misdetection and omission detection in dense scenes.

Abstract:In recent years, the widespread deployment of high-definition and ultra-high-definition surveillance cameras has led to a significant increase in the volume of fixed-scene video data, such as surveillance videos. This sharp rise in data has imposed tremendous pressure on video storage and transmission. To further eliminate redundancy in fixed-scene videos, this paper proposes a novel compression and reconstruction method. By employing background extraction and an inter-frame foreground difference detection-based foreground extraction and compression approach, a substantial amount of data redundancy is removed from the videos. Experimental results show that, compared to MPEG-4, the proposed method achieves higher video reconstruction performance at a higher compression ratio. Compared to H.264, H.265, and DCVC-DC, the proposed method improves compression performance by 82.75%, 76.19%, and 59.56% respectively, while maintaining a high level of video reconstruction quality. This effectively alleviates the storage and transmission pressure of fixed-scene videos.

Abstract:Driver distraction behaviour detection is of crucial significance for the development of driver-centered human-vehicle co-driving systems. Aiming at the existing convolutional neural network-based driver distraction detection models that lack global feature extraction capability, have weak generalisation performance and neglect of the importance of different regions in the driving scene, a driver distraction detection model based on deep learning is constructed to achieve accurate prediction of driver distraction behaviour. First, a residual structure based on HorNet is developed to enhance the feature representation capability through higher-order spatial interactions; second, inspired by the human attention mechanism and the existing attention mechanisms, an adaptive weighted attention strategy is designed to extract the features most relevant to the driving behaviour; and then, the model in this paper is trained on the existing categorical dataset, and the a priori knowledge is used as the initial weights to improve the training results which in turn improves the generalisation ability of the model; finally, the driving behaviour features are visualised to improve the trust in this paper′s model. The experimental results show that the model in this paper can accurately detect driver distraction behaviour, which is significantly better than existing methods in terms of accuracy, and reliability.

Abstract:In order to solve the problem that a large number of detail features are missing after the simplified three-dimensional grid model of transmission towers, a lightweight algorithm for the three-dimensional grid model of transmission towers is proposed based on Quadric Error Metrics algorithm. The algorithm firstly defines the detail features in the 3D grid model of the transmission tower, then proposes the detail feature extraction strategy of the transmission tower, and introduces the detail feature significance factor and vertex approximate curvature factor to optimize the folding cost in the QEM algorithm. The experimental results show that the improved algorithm can effectively retain the important geometric features and detailed features of the three-dimensional grid model of the transmission tower, avoiding the problem of large-area feature loss in the simplified model, and compared with the ordinary QEM algorithm, the maximum error, mean error and mean square error of the simplified model are reduced by at least 39.77%, 10.64% and 64.99% respectively, which realizes the high quality and lightweight of the three-dimensional grid model of the transmission tower.

Abstract:Accurate plant counting is crucial in precision agriculture, forming a critical foundation for monitoring crop growth and predicting yield. To address challenges such as densely packed, overlapping, and aerial small targets of tobacco plants during the maturity stage, a lightweight GEW-YOLOv8 tobacco plant counting algorithm was proposed. The algorithm utilizes the GhostC2f module to reduce the parameters and computational workload of the model and employs an efficient multi-scale attention mechanism to discern occluded tobacco plants. Additionally, the WIoU loss function is introduced to accelerate model convergence and improve accuracy. Experimental results show a significant improvement in efficiency and accuracy compared to the original model, with a 24.7% reduction in FLOPs and a 26.7% decrease in model size. The improved model tobacco plant detection accuracy AP0.5 and AP0.5~0.95 reached 99.1% and 86.2% respectively, which were increased by 0.8% and 3.6% respectively compared with the original YOLOv8n model. The improved model can more swiftly and accurately identify field tobacco plants, providing technical support for intelligent tobacco agriculture.

Abstract:With the development and popularity of smartphone products, a large number of bowed-head tribes have emerged who play mobile phones at any time regardless of the occasion; for the frequent occurrence of traffic accidents caused by bowed-head tribes′ dependence on mobile phones, a multimodal bowed-head tribes′ hazard perception and warning system based on mobile phones is proposed. First, gravity acceleration on the mobile phone side is used to monitor behaviors in real time based on fuzzy control rules, including Walking and looking at the mobile phone, Walking up and down stairs, Looking at the mobile phone at rest, Walking with the mobile phone in hand, Walking with the mobile phone in pocket; and then the user′s environment is described in real time using the mobile phone′s rear view camera images based on the grouping of fast spatial pyramids pooled in the lightweight YOLO network, including: stairs, crosswalks, low-light environments, puddles, and normal road surfaces. Finally, a state-environment-multimodal hazard detection model is constructed for the Android system; and based on the detection results, audible, visual, and tactile three-dimensional warning signals are given to the bowed tribe by using sound, image, and vibration signals to reduce the potential hazards of the bowed tribe such as fall injury and collision. Online experiments show that the proposed multimodal threat perception model for mobile phones is highly accurate, robust, and real-time, and is able to achieve effective proactive warning for the common threat states of bowed heads.

Abstract:In response to the maintenance requirements of airborne radar in field operations, this paper proposes a process-oriented automatic measurement method for the parameters of airborne radar signals based on the spectrum analysis module. The principles and steps of automatic parameter measurement for multi-system radar signals are analyzed in detail in the paper. Corresponding testing software is also developed to execute the relevant algorithms and enable streamlined measurements. Additionally, four signal simulation experiments are designed to validate the effectiveness of the proposed method. The experimental results demonstrate that this method can achieve a comprehensive measurement of multiple parameters of typical radar signals in a streamlined manner, without relying on traditional general instruments. It solely utilizes the sampling data provided by the spectrum analysis module in different working modes. The measurement results are accurate and effective, meeting the requirements of field maintenance support. Therefore, it possesses strong engineering application value.

Abstract:For the waveform distortion of narrow pulse signal after amplification, frequency conversion and other analog devices, a waveform distortion correction method for signals based on modified frequency domain filter is proposed. Traditional distortion correction methods only use signals within the effective bandwidth. It results in poor time domain performance after waveform distortion correction. To improve the distortion correction accuracy and ensure the time domain waveform characteristics of the signal, the proposed method uses the frequency responses of the effective bandwidth and the partial high-frequency region outside the bandwidth to solve the waveform distortion correction filter. But the spectrum of the distortion correction filter fluctuates greatly and has many spikes and burrs in the high-frequency region outside the bandwidth. It can′t be directly used to solve the coefficients of the waveform distortion correction filter. The proposed method applies median filtering to the amplitude-frequency response correction curves and phase-frequency response correction curves, followed by polynomial fitting. Better waveform correction performance has been achieved by optimizing the order and coefficients of the waveform distortion correction filter. Finally, the real data processing results verified the effectiveness of the proposed method.