Abstract:Based on the problem of ignoring the slight difference between normal data and fault data and insensitive detection of traditional data-driven process fault detection, this paper proposes a fault detection method based on the combination of FRFT and CNN. Starting from amplifying the small differences between normal data and fault data, a residual matrix is constructed by CVDA for data monitoring to enhance sensitivity. The second is to use FRFT to transform the data, convert some faults with low amplitude and easy to be masked by noise from the time domain to the frequency domain, and amplify their characteristics as much as possible to make them easy to detect. Finally, CNN is used to detect the processed data, which solves the problems of ignoring small differences and low detection sensitivity, and experiments are verified by TE process, which improves the fault detection rate and shows the effectiveness of the proposed method.

Abstract:Aiming at the small output torque and the coupling problem between torque and suspension of traditional bearingless switched reluctance motor, a four-phase 16/14/8-pole hybrid-excitation double stator bearingless switched reluctance motor is proposed in this paper. The topological structure, torque principle and suspension principle of the motor are introduced, and the excitation mode of the torque winding is optimized. The electromagnetic characteristics of the motor with the same parameters without permanent magnet are compared and analyzed by finite element method, including the torque characteristics and decoupling characteristics of the motor. The simulation results verify that the self-decoupling structure of the motor is reasonable and can effectively improve the output torque. On the basis of Maxwell stress method, the equivalent magnetic circuit diagram is established according to the magnetic flux distribution of the motor, and the core magnetization curve is fitted by piecewise function. The mathematical model of torque considering magnetic saturation is derived. The torque model agrees well with the simulation results verified by the finite element, with a minimum difference of about 1.4% and a maximum difference of about 26%.

Abstract:Aiming at the problem of low detection accuracy in the field of network intrusion detection, we study the intrusion detection model when there are few samples of anomalous traffic and the performance of classifiers is poor, and propose an intrusion detection model based on improved generative adversarial network and hybrid spatio-temporal neural network. The improved generative adversarial network generates artificial anomalous traffic samples with specific labels by learning the distribution characteristics of the anomalous traffic samples; the fusion convolutional neural network and bidirectional long and short-term memory neural network extracts the spatio-temporal fusion features of the attacking traffic, and utilizes the attention mechanism to weight the spatio-temporal fusion features and constructs a hybrid spatio-temporal neural network to classify and predict the network traffic. Simulation experiments of the proposed model are conducted on the UNSW-NB15 dataset, and the accuracy and F1 score are 92.93% and 94.81%, respectively, indicating that the proposed model can effectively improve the problem of category imbalance in the original dataset, and improve the detection capability of abnormal traffic samples and the detection accuracy of network intrusion.

Abstract:To solve the problems of artificial weighing, which is time-consuming and laborious, easy to cause animal stress, and electronic instruments are easy to destroy in the process of poultry breeding, a set of wireless dynamic adaptive weighing systems is designed in this paper. The system combines empirical mode decomposition and robust Kalman filter to improve the algorithm and proposes an innovative queue-based automatic peeling algorithm to solve the weighing zero offset problem caused by feces and feed accumulation. Through the practical application and monitoring verification in broiler farms, the results show that the wireless dynamic adaptive weighing system designed in this paper can obtain animal weight quickly and accurately, and has good adaptability, stability, and robustness.

Abstract:In view of the fact that the electric field sensor in the market can only detect the vertical component of one-dimensional electric field, and there are some errors in the electric field measurement results, in order to measure the intensity of atmospheric electric field more accurately and improve the accuracy of thunderstorm cloud early warning, this paper proposes a field-grinding three-dimensional atmospheric electric field sensor. The main structure of the sensor is composed of a shield, an induction electrode, a photoelectric switch, a blade and a motor. The circuit design mainly includes an I-V conversion circuit, a differential amplifier circuit, a peak detection circuit and a filter circuit, etc., and the field electric field measurement experiment of the sensor is carried out and the experimental data are sorted The experimental results show that the sensor can measure the horizontal and vertical components of the environmental electric field, which effectively reflects the characteristics of the environmental electric field, and verifies the effectiveness of the structural design and circuit design of the field grinding three-dimensional atmospheric electric field sensor.

Abstract:This paper presents a yaw angle random error correction method based on a line detection model and multisensor data fusion, aimed at enhancing the accuracy of a lowcost sensor-equipped positioning platform in agricultural and forestry environments. The method achieves dynamic state adjustment by tuning the line detection threshold to improve the robustness and precision of the navigation system. Subsequently, it fuses multisensor data using Kalman filtering to correct yaw angle random errors. Experimental results demonstrate the method′s effectiveness under various paths and velocities. In straight-line progress experiments, the positioning accuracy of this method remains within 5 cm, with a yaw angle error within 5°. In rectangular progress experiments, the trajectories closely resemble those of the differential RTK method, with an average error of only 2.7 cm and a standard deviation of 3.9 cm. This yaw angle correction method provides robust support for autonomous operations in agricultural machinery and vehicle environments. It is adaptable to different environmental conditions, thereby enhancing the performance and measurement accuracy of navigation systems.

Abstract:It is one of the key technologies to extract the center of laser stripe quickly and accurately when using linear Structured light for 3D measurement of workpiece.In this paper, a method that a laser center extraction method based on spatial gray gravity center advancing is proposed.which method bases on advancing the gray center of gravity in the laser stripe space, combining the eight neighborhood decision method to select the light stripe space area, then extracting the new spatial gray Barycentric coordinate system coordinates of the selected area;The pauta criterion eliminates abnormal center points and ultimately obtains the laser center coordinate information behind completing the entire image extraction.The experimental shows that the Root-mean-square deviation of the extraction center of the proposed algorithm is 0.492 pixel, which is 9.8% higher than the optimal extraction precision Steger improved algorithm, and the extraction speed is 5 times higher. Compared to the internal advancement algorithm with the best processing speed, while maintaining the extraction speed, the accuracy has been improved by 24.1%.At the same time, the proposed algorithm greatly enhances the processing ability of under exposed light stripe images, effectively reduces the restriction of the environment on the measurement of line Structured light.

Abstract:In complex factory environments with multiple obstacles, in order to solve the problem that the path generated by rapidlyexploring random tree star algorithm(RRT*) has redundant points, is close to the obstacle and has jagged turns, the path planning algorithm of SSRRT*［Safe-Smooth RRT*］ is improved. Firstly, a target biasing strategy is introduced. Secondly, the algorithm utilizes a new node expansion approach that combines the concept of target point attraction and an improved nearest neighbor point metric to reduce the blind expansion of the tree and accelerate growth towards the target point; Node security constraints are then imposed to add the security nodes to the tree; Improved path simplification eliminates redundant points while taking into account security; Finally, the local smoothing of the B-spline is used to improve the smoothness of the path. By comparing with the standard RRT* algorithm, the adaptive target bias RRT algorithm, and improved RRT algorithm，the maximum decrease in average path length is 7.1%, the maximum decrease in average effective node number is 64.1%, and always keep a safe distance from obstacles. The results indicate that the improved algorithm effectively improves the smoothness and safety of the path.

Abstract:Permanent magnet synchronous motors require sampling of the motor phase current during vector control. During the sampling process, many factors such as reference level fluctuations can cause the phase current to contain DC components. In addition, online parameter estimation methods such as DC voltage injection also introduce DC components into the phase current. This article provides a detailed analysis of the generation mechanism of DC components in different situations, and investigates the impact of DC components on control effectiveness, torque ripple, efficiency, and other performance. Propose a data processing method that can filter out the DC component in three-phase AC signals. This method can effectively suppress the DC component introduced by different factors, thereby increasing control stability and improving the efficiency of the motor system. Simulation and application practice have verified the effectiveness of the algorithm.

Abstract:In to achieve efficient search capability for path planning of mobile robots in complex environments, a hybrid multi-strategy improved dung beetle optimizer has been proposed. Firstly, the ISPM chaos strategy is introduced to initialize the initial population of fireflies. This ensures a more uniform distribution of the initial population and reduces the likelihood of the algorithm getting stuck in local optima. Then, the greedy selection strategy is combined with the improved lens imaging reverse learning strategy to update the positions of the fireflies during their foraging behavior. This balances the algorithm′s local exploration and global search capa-bilities, thereby enhancing its convergence ability. Finally, the Levy flight strategy and an improved dynamic weight update mechanism are employed to update the positions of the fireflies during their stealing behavior. This helps to change the optimal global solution and prevent the algorithm from getting trapped in local optima. To evaluate the performance of the improved algorithm, comparative experiments are conducted with four other swarm intelligence algorithms using benchmark test functions and simulation of path optimization. The experimental results demonstrate that the improved dung beetle optimizer significantly improves convergence speed and optimization accuracy, while maintaining good robustness.

Abstract:An active suspension body attitude control strategy based on attitude compensation is proposed to address the body attitude imbalance problem generated during vehicle driving. Based on the establishment of a seven-degree-of-freedom dynamics model for the vehicle and the verification of the accuracy of the vehicle model based on random road tests, a model predictive controller is constructed to solve the vertical control force for each suspension based on the signals estimated by the state observer to attenuate the vehicle droop vibration; and then the vehicle attitude compensation control strategy is designed based on a fuzzy algorithm to cause the electromagnetic linear actuator to generate a reaction force to suppress the body attitude deterioration. A certain type of linear motor is selected as the active suspension force source, and the electromagnetic force required for suspension control is obtained by combining the vertical control force with the attitude compensation force, which is used to calculate the target current required by the motor, and the active suspension system is simulated by the MATLAB/Simulink platform. The simulation results show that the proposed active suspension control strategy based on attitude compensation can significantly reduce the root mean square values of the vehicle centre of mass and pitch angle without affecting the vehicle droop control effect, and the body attitude is effectively controlled.

Abstract:The existing methods for measuring the sag of power lines are cumbersome and lack a high level of automation. This paper proposes a power line sag measurement method that integrates monocular depth and RTK positioning. Firstly, unmanned aerial vehicles capture key images of power line routes, and these images are input into the constructed monocular depth estimation model, EleDep-Net, to generate corresponding depth maps. The model incorporates a strip-like pyramid module and a boundary fusion attention module to accurately capture semantic information in the context of the conductors. Secondly, a depth correction algorithm is introduced to further refine the depth values in the depth map, obtaining depth information for the key points. Finally, by combining UAV RTK positioning and key point depth information, spatial coordinates for the key points are generated in the reference coordinate system, and a parabolic formula is fitted to derive the sag of the power line. Tested in a real-world environment of distribution network lines, the results indicate that this method significantly improves operational efficiency while ensuring a relative measurement error of less than 5%, demonstrating its high engineering application value.

Abstract:For the automated access process of intelligent warehouse materials, for the access of materials stored in cartons, it is necessary to carry out tasks such as carton size measurement, carton disassembly, material dumping, material picking, etc. Therefore, the carton should firstly be unmanned size measurement, and the SGBM improvement algorithm of Adaptive GrabCut is proposed to realize the automatic measurement of carton size. The method firstly completes the calibration of binocular camera, uses NLM algorithm to denoise the collected pictures, completes the stereo correction of carton pictures, uses NCC, SGBM, AD-Census three stereo matching algorithms to get the parallax map of carton, on the basis of analyzing the effect, proposes the SGBM improvement algorithm based on the template matching Adaptive GrabCut to be used for the stereo matching of carton and get its size information. The experimental results show that the improved algorithm can realize the accurate measurement of carton size information, and the size error is less than 10 mm, which meets the actual production requirements.

Abstract:A new underwater image enhancement algorithm based on improved U-Net is proposed to address the problems of color distortion, fuzzy fogging, and low contrast in underwater degraded images. A new residual attention structure and edge detection module are designed and introduced into the U-Net network to construct the improved underwater image enhancement algorithm. The experimental results show that the algorithm proposed in this paper obtains good results in both correcting the underwater color bias and enhancing the contrast, with an average improvement of 14.2% in the IE value compared with the original image and an average improvement of 24% in the UCIQE value compared with the original image. The results of the ablation experiments show that the residual attention structure, edge detection module, and loss function proposed in this paper have positive effects on underwater image enhancement.

Abstract:To address the issues of small inter-class differences and difficulty in distinguishing fine-grained images, this paper proposes a method that improves the network’s ability to express image detail features, aiming to alleviate this problem. To achieve this, an improved Transformer-based algorithm for finegrained recognition is designed in this study. Firstly, deformable convolutional token embedding adjusts the sampling points adaptively to modify the convolution operation range and the shape of its kernel, enhancing the network’s perception of spatial information for more accurate spatial details. Secondly, an efficient correlation channel attention mechanism automatically selects channels to transform the computation from neighboring channels to semantically similar channels, capturing semantic-related channel information. The precise spatial information and semantically related channel information effectively enhance the network’s perception of local features. Experimental results demonstrate that compared to the baseline algorithms, the proposed method improves recognition results by 1.5%, 2.4%, and 1.5% respectively on the CUB-200-2011, Stanford Cars, and Stanford Dogs datasets. These results indicate that the proposed approach effectively enhances the effectiveness of fine-grained image recognition by improving the expression capability of image detail features.

Abstract:In response to challenges in practical road target detection, such as low accuracy in detecting small targets and the occurrence of missed and false detections for occluded targets, an improved YOLOv5s road target detection algorithm, termed YOLOv5s-OEAG, is proposed in this study. The label assignment strategy of YOLOv5s is replaced with a more efficient OTA label assignment strategy to enhance the model′s detection accuracy and generalization ability. Additionally, a lightweight decoupled prediction head is introduced to decouple classification and regression tasks for different-sized feature layers, thereby improving the model′s capability to detect small targets on roads. The original nearest-neighbor interpolation upsampling module is replaced with the lightweight and versatile CARAFE module to better preserve fine details in the image, thereby enhancing the model′s accuracy. Furthermore, a novel C3 module, GMC3, is proposed to reduce model computational complexity while improving the model′s feature capturing capability. To enhance the model′s generalization ability, the KITTI dataset is augmented, increasing the number of small targets. Experimental results demonstrate that the improved model achieves a mAP of 90.4% on the augmented KITTI dataset, representing a 2.8% improvement over the original model′s accuracy. With a frame per second (FPS) rate of 75, meeting real-time requirements, the model exhibits enhanced adaptability to complex traffic scenarios.

Abstract:In order to solve the problem of insufficient extracted features in high-resolution human pose estimation using lightweight convolutional neural network, a high-resolution human pose estimation network based on feature enhancement is proposed in this paper. Firstly, the dilated convolution completion operation was used to extract image features to avoid the loss of feature information and basically keep the model parameters unchanged. Then, the pooling enhancement module was used to select the features of convolution extraction, which retained important features and reduced the damage caused by traditional pooling module on extracted features. Finally, the depthwise separable convolution module that strengthens the channel information interaction was used for feature extraction, so as to keep the number of parameters of the module small and improve its feature extraction ability. The performance of the proposed algorithm and DiteHRNet-30 algorithm were tested on the COCO2017 dataset. The AR values of the proposed algorithm and DiteHRNet-30 algorithm are 77.9% and 77.2%, respectively. The performance of the proposed algorithm and DiteHRNet-30 algorithm are tested on the MPII dataset. The PCKh values of the proposed algorithm and DiteHRNet-30 algorithm are 32.6% and 31.7%, respectively. Experimental results show that the proposed algorithm can achieve a good balance between the accuracy of human pose estimation and the complexity of the algorithm.

Abstract:Aiming at the problems of complex operation and high cost of traditional contact measurement of bearing size in industrial environment, an improved edge detection algorithm for HED network is proposed by adding attention mechanism and Canny algorithm. The method is based on the HED network, the convolutional layer of the fourth and fifth stages of the backbone network is replaced with a continuous hollow convolutional layer, and the pooling step of the third and fourth layers of the network is set to 1, which increases the model receptive field and improves the output edge image accuracy. The efficient channel attention mechanism ECA module is added to effectively suppress the influence of irrelevant texture features and non-edge pixels. Using the non-maximum value suppression and double threshold processing algorithms in Canny algorithm, the coarse edges detected are refined to obtain more accurate bearing edges. The inner and outer diameter parameters of bearing are obtained by using the least square circle fitting method. The experimental results show that the improved HED model achieves 0.811 and 0.833 respectively in ODS and OIS, which can effectively realize the bearing edge detection and ensure the bearing size measurement accuracy.

Abstract:To address the issue of existing object detection algorithms struggling to meet real-time detection requirements in UAV remote sensing, we propose a model compression method based on ShuffleNetv2 and structured pruning. Using YOLOv5m as the foundation, we incorporate the ShuffleNetv2 model as the backbone network of YOLOv5m, reducing the model′s parameter count and computational complexity while improving inference speed. Furthermore, we employ the ECA attention mechanism to replace the SE module in ShuffleNetv2, enhancing the feature extraction capability of the backbone network. Additionally, we adopt FocalEIoU as the loss function for the YOLOv5 algorithm, improving the model′s regression ability. Finally, we use channel pruning to eliminate redundant parameters in the Neck structure, further compressing the model′s parameters and computational complexity, and enhancing the pruned model′s accuracy through fine-tuning.Experimental results show that, under the same testing environment, compared to YOLOv5m, the proposed model reduces the parameter count and floating-point operations by 86.3% and 80.0%, respectively. The model achieves an mAP@0.5 of 92% and an mAP@0.5:0.95 of 50.4%, outperforming other mainstream detection algorithms. Moreover, the proposed model achieves a detection speed of 35 frames/s on the AGX edge computing platform, satisfying the requirements for real-time detection.

Abstract:The partial discharge signal of power equipment is prone to interference from narrow band noise and white noise in the environment. In order to better preserve the characteristics of local discharge signal for fault diagnosis and prediction, a method of denoising transformer partial discharge signal based on compressed sensing reconstruction and variational mode decomposition is proposed. This method firstly uses the window function to suppress the frequency leakage of narrowband interference, and then separates and reconstructs narrowband signals to suppress narrowband noise by taking advantage of the difference in sparsity between narrowband interference and local emission signal and white noise in the frequency domain. Secondly, by improving variational mode decomposition method, different modes are classified and denoised according to the amount of local emission signal information contained in each mode. Finally restore the outgoing release signal. The denoising effect of this method is tested by simulation and actual signal, and the denoising effect is compared with that of singular value decomposition and variational mode decomposition. The results show that this method can effectively suppress the interference of partial discharge signal, and the waveform similarity coefficient is improved by about 2% compared with the traditional algorithm, and the waveform characteristics of partial discharge signal can be better preserved.

Abstract:EEG signal recognition methods rarely integrate spatial, temporal and frequency information, in order to fully explore the rich information contained in EEG signals, this paper proposes a multi-information fusion EEG emotion recognition method. The method utilizes a parallel convolutional neural network model (Parallel Convolutional Neural Network, PCNN) that combines a two-dimensional convolutional neural network(2D-CNN) and a one-dimensional convolutional neural network(1D-CNN) to learn the spatial, temporal, and frequency features of the EEG signals to categorize the human emotional states. Among them, 2D-CNN is used to mine spatial and frequency information between neighboring EEG channels, and 1D-CNN is used to mine temporal and frequency information of EEG. Finally, the information extracted from the two parallel CNN modules is fused for emotion recognition. The experimental results of emotion triple classification on the dataset SEED show that the overall classification accuracy of the PCNN fusing spatial, temporal, and frequency features reaches 98.04%, which is an improvement of 1.97% and 0.60%, respectively, compared to the 2D-CNN extracting only null-frequency information and the 1D-CNN extracting temporal-frequency information. And compared with recent similar work, the method proposed in this paper is superior for EEG emotion classification.

Abstract:Due to the short-term large value characteristics of impulsive noise, the signals parameter estimation method based on Gaussian hypothesis cannot effectively estimate parameters in impulsive noise environment. To solve this problem, this paper proposes an LFM signals parameter estimation method based on Sigmoid-CFRFT by using alpha stable distribution to simulate random impulsive noise. Firstly, an improved Sigmoid function is established to prove that after this nonlinear transformation, the 2nd moment of the signal changes from unbounded to bounded, and the phase information of the signal remains unchanged. Secondly, the transformed signal is discrete-time CFRFT, a mathematical optimization model for LFM signals parameter estimation is established, and the water cycle algorithm is used to search for the optimal value point. Finally, a correction method for non-standard SαS distribution noise is used, and the performance of parameter estimation under standard and non-standard distribution is analyzed. The simulation results show that the proposed method can not only effectively suppress the influence of impulsive noise on the fractional spectral characteristics of LFM signals, but also achieve high-precision estimation of signals parameters with low signal-to-noise ratio. Compared with the existing parameter estimation methods based on nonlinear transformation, the proposed method has better accuracy, stability and noise robustness.

Abstract:The fine identification of ground-based clouds is of great significance for climate prediction and meteorological research. Aiming at the current problems of low accuracy of ground-based cloud recognition, poor generalization, and detrimental deployment of marginalized deployment, a ground-based cloud map recognition model based on residual network is proposed, named GBcNet. The designed model consists of one convolutional layer, two pooling layers, five residual blocks and one fully connected layer, using the first convolutional layer and the first pooling layer to initially extract feature information and reduce the feature map size, and extract more feature information through the residual block, while suppressing the overfitting and gradient disappearance of the network, and finally using another pooling layer to reduce the size of the feature map, and finally output the recognition results through the fully connected layer. The experimental results show that the comprehensive average accuracy of GBcNet model on the dataset reaches 96.02%, and the recognition accuracy of 11 categories of ground-based clouds is between 93% and 99%, and has better generalization, and the recognition performance of single category and overall is better than that of other models. Furthermore, the SWIMCAT dataset was used to experiment with the model, and the comprehensive recognition accuracy reached 99.7%, which proved that the model was universally applicable to the recognition of ground-based cloud maps. The model has a simple structure, which is more conducive to marginalized deployment than other models.