Abstract:The resistivity of magnetic samples at the micro-and nanoscale is affected by various factors such as temperature, size, and applied magnetic field. The traditional two-wire method to measure the sample resistance is easily affected by external interference, and the presence of contact resistance will lead to the measurement results not being precise enough, which affects the experimental results. In this paper, the four-point probes method is used to study the resistance characteristics of magnetic microwire samples. Firstly, two different sizes of nickel microwire are prepared based on photolithographic process, and at the same time, a measurement system that can accurately control the temperature and the magnetic field is used to study the effects of the ambient temperature and the external magnetic field on the resistance of the samples of different sizes. The experimental results show that the resistance value of the nickel microwire almost does not change with the temperature in the low-temperature region below -250℃, and then increases approximately linearly with the increase of the temperature; at room temperature, the resistance value of the nickel microwire increases with the increase of the strength of the external magnetic field, and the resistance value of the samples is basically stable after the magnetic field strength reaches 2 000 Gauss. The article explains the temperature effect, size effect, and magnetoresistance effect of magnetic microwire resistance from the microscopic electron transport perspective.

Abstract:For the problem that the speed measurement error of high-speed laser Doppler velocity measurement system cannot be effectively calibrated, this paper establishes a calibration scheme for the indicated value error based on the echo simulation method, in which the optical frequency modulation and optical wavelength modulation simulate the optical echo that carries the target information in the speed measurement process of the system, and then indirectly simulate the reference speed, so as to realize the calibration of the indicated value error of laser Doppler velocimetry system. Based on the self-developed calibration device, a calibration experiment is built, and a laser Doppler velocimeter with an upper speed limit of up to 10 km/s is used as the calibration object to carry out the calibration experiment of the display value error. By analyzing the experimental results, it can be seen that the echo simulation calibration scheme based on optical frequency modulation and wavelength adjustment can realize the calibration of speed measurement error of the speedometer in a wide speed domain, and the measurement accuracy analysis shows that the calibration uncertainty of the optical frequency modulation method is U=0.26% (k=2), and the calibration uncertainty of the wavelength adjustment method is U=1.5% (k=2). This index meets the calibration demand of laser Doppler velocimeter indicating value error, and provides a method reference for the calibration of laser Doppler velocimeter indicating value error.

Abstract:Harmonic governance is very important to power quality control. The traditional ip-iq method has problems such as low detection accuracy and slow speed. Therefore, it is proposed to increase the negative feedback loop based on quasi-proportional-resonance(QPR) filter based on the traditional second-order generalized integrator (SOGI), and replace the traditional PI link with the P link, and change the low-pass filtering link in the traditional ip-iq method to obtain an improved ip-iq harmonic current detection method. The simulation and experimental results show that the improved phase lock loop shows excellent frequency locking ability in the condition of abnormal power grid. In addition, through the improved ip-iq harmonic detection method, the current distortion rate of the fundamental wave current calculated by the algorithm can be significantly reduced, and the current distortion rate is reduced from 0.29% to 0.20%. And reduced from 8.63% to 0.51% in the presence of a DC component. At the same time, the harmonic can still be accurately detected in the complex power grid environment.

Abstract:A fault diagnosis method based on an improved black winged kite optimization algorithm (GBKA) optimized variational mode decomposition (VMD) and wide convolutional neural network (WDCNN) is proposed to address the issue of poor diagnostic performance caused by noise interference and other factors in gearbox fault diagnosis. Firstly, in response to the shortcomings of the black winged kite algorithm (BKA), which is prone to falling into local optima and premature convergence, genetic algorithm′s gene crossover recombination and mutation operations are introduced to improve BKA; secondly, using the improved GBKA to optimize VMD parameters, modal components are screened through correlation coefficients and the signal is reconstructed; finally, the reconstructed signal is input into the WDCNN model to achieve fault classification. The results indicate that GBKA has better optimization performance compared to BKA in the test function; under two operating conditions, the average fault classification accuracy of this method reached 99.645% and 99.978%,which is superior to other comparative methods. In addition, it was less affected by noise in the noise experiment, verifying the effectiveness and stability of the proposed model and providing a reliable solution for gearbox fault diagnosis.

Abstract:In order to solve the problems of low accuracy, slow detection speed and high model complexity of obstacle detection of campus intelligent sweepers, a modified YOLOv8s obstacle detection and ranging algorithm for campus intelligent sweepers YOLOv8s-FDR was proposed. On the basis of the YOLOv8s algorithm framework, the backbone network is replaced by the FasterNet network with smaller parameters and memory access, so as to reduce the complexity of the model and improve the detection speed. Then, the SPPF-DAM module was designed to introduce the deformable attention mechanism (DAM) in the form of residuals to improve the model′s perception of multi-scale target features. Secondly, Partial-RFEM was used for downsampling in the feature fusion network to capture the non-local context features and local target features to improve the detection accuracy. Finally, the ranging function is added to reduce hardware costs. Experimental results show that compared with the original algorithm, the mAP of the improved algorithm is increased by 3.6%, and the amount of model computation and parameters is reduced by 19.72% and 15.27%, respectively. The actual environment test shows that the detection speed of the YOLOv8s-FDR algorithm reaches 38.44 fps, which is much higher than the 17.12 fps of the original algorithm, which can meet the performance requirements of the normal operation of the campus intelligent sweeper.

Abstract:The bleed air system of aircraft engines is a critical system to ensure flight safety, and fault detection is essential for maintaining it. This study addressed the fault detection problem of the bleed air system by first using the modified density-adaptive synthetic oversampling algorithm (MDADASYN) to handle the imbalanced fault data. Then, a multi-strategy improved energy valley optimization algorithm (IEVO), enhanced with good point set population initialization, Gaussian-Cauchy mutation strategy, and dynamic parameter adjustment, was applied to optimize the general regression neural network (GRNN) for fault diagnosis. Results from CEC2014 test functions demonstrated that the proposed strategy effectively improved population diversity as well as global and local search capabilities. Simulation experiments based on real fault data from the bleed air system verified that the MDADASYN-IEVO-GRNN fault diagnosis model significantly enhanced diagnostic accuracy for bleed air system faults in aircraft engines, contributing to improved operational safety and maintenance efficiency.

Abstract:Aiming at the ranging error problem of ultra-wideband in actual environment, a UWB ranging error mitigation algorithm based on improved sparrow search algorithm and convolutional neural network bi-directional long short-term memory attention model is proposed. Tent mapping is adopted, adaptive adjustment method is used, combined with northern goshawk algorithm, and spiral flight strategy is adopted in the following stage to improve the SSA algorithm, improve the global search performance of the algorithm and avoid falling into the local optimum. The BiLSTM model and attention mechanism are used to improve the CNN-LSTM model, and the CNN-BiLSTM-Attention model is constructed to improve the model′s ability to process long sequence data, so that the model has more accurate weight distribution for data. TANSSSA is used to optimize the hyperparameters of the CNN-BiLSTM-Attention model, and the TANSSSA-CNN-BiLSTM-Attention model is constructed. In the model performance verification experiment, the average absolute error of SSA-CNN-BiLSTM-Attention, CNN-BiLSTM-Attention, CNN-BiLSTM, CNN-LSTM-Attention, CNN-LSTM, GRU and TCN models was reduced by 12.05%~62.31%. In the actual environment, the average absolute error of TANSSSA-CNN-BiLSTM-Attention was reduced by 45.70%~83.82% compared with the other seven models, and the ranging error was effectively alleviated.

Abstract:In order to solve the problem that the component defect detection in the transmission line inspection image is easily disturbed by the background environment and the defect target scale is different, resulting in low detection accuracy, an improved YOLOv10n transmission line component defect detection algorithm was proposed. Firstly, the RepViTBlock and ELA attention mechanism were used to redesign the C2f, and the ERC2f module was constructed to suppress the background environment interference, enhance the feature extraction ability of the model, and reduce the parameter redundancy. Secondly, the DASF neck structure was designed by combining dynamic upsampling DySample and attention scale sequence fusion module ASF to improve the multi-scale feature fusion ability of the model. Thirdly, based on the DBB of diversified branch blocks, a reparametric shared convolutional detection header RSCD is proposed, which reduces the redundancy of the header parameters and strengthens the interaction ability of feature information by sharing parameters. Finally, the MPDIoU loss function is optimized to Inner-Wise-MPDIoU by drawing on the ideas of Inner-IoU and WIoUv3 to accelerate the model convergence process and improve the defect positioning accuracy. The experimental results show that the accuracy of the improved algorithm for the detection of transmission line component defects reaches 92.1%, which is 3.4% higher than that of the original algorithm, and the number of parameters and GFLOPs are reduced by 19.4% and 0.4, respectively, which proves the effectiveness of the improved algorithm.

Abstract:Accurate prediction of the state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries is of great significance for their safe and stable operation. In this paper, a novel Algorithm combining the improved hippopotamus optimization algorithm (IHOA) and deep extreme learning machine (DELM) is proposed. Six health indicators (HIs) were extracted from the charge and discharge process of lithium batteries, and the five HIs with high capacity correlation were retained by Pearson correlation analysis. Then, the outliers in the feature data were removed and normalized by Hampel filtering. Finally, the DELM model of SOH and RUL is established. In addition, In order to improve the prediction efficiency of the model, an IHOA is proposed to optimize the hyperparameters of DELM. Compared with the traditional hippopotamus optimization algorithm (HOA), it solves the limitations of the traditional hippopotamus algorithm in search efficiency, convergence speed and global search. The experimental simulation results based on CALCE lithium battery data set show that the prediction accuracy of IHOA-DELM algorithm is high, the root mean square error (RMSE) of SOH prediction of the proposed method ranges from 1.21%~1.31%, and the mean absolute error (MAE) value ranges from 0.89%~0.95%. The mean absolute percentage error (MAPE) was between 1.59%~1.93%. The maximum absolute error (AE) value predicted by RUL does not exceed 3 cycles, and the minimum absolute error value is only 1 cycle.

Abstract:To improve the convergence speed and optimization accuracy of the red-billed blue magpie optimizer, a multi-strategy improved red-billed blue magpie optimizer is proposed for the first time. Firstly, in order to improve the diversity and coverage of the initial population, circle chaotic mapping is used to initialize the population; secondly, combining the spiral search strategy with hunting behavior to expand the optimization range, while balancing the algorithm′s global exploration ability and local development ability; finally, the Cauchy mutation perturbation strategy is introduced during the iteration process to avoid the algorithm falling into local optima in the later stages, further improving the overall efficiency of the algorithm. Using 15 test functions for simulation experiments, the results show that the improved algorithm has improved optimization accuracy, convergence speed and stability. Compared to the original algorithm, the average error was reduced by 90.68% and the average standard deviation was reduced by 98.60%. The application of the improved algorithm to engineering optimization problems has verified its feasibility.

Abstract:When using traditional Mel-frequency cepstral coefficients (MFCC) as features for target recognition in wild filed environments, their sensitivity to environmental interference often leads to a decline in recognition accuracy. To address this issue, this study proposes replacing the discrete cosine transform used in the MFCC extraction process with principal component analysis (PCA) and applying a nonlinear function to map the logarithmic energy obtained after Mel filtering. The improved MFCC are more aligned with actual data, can enhance the weighting of frequency bands containing the target′s signal, and have better separability and robustness. Experimental results show that using the improved MFCC based on PCA and nonlinear mapping as classification features achieves an accuracy of 93.36%, surpassing the performance of traditional MFCC.

Abstract:To address the limitations of existing optical dust detection methods, which are susceptible to environmental factors and lack precision, a novel method and system for detecting dust accumulation on photovoltaic modules has been developed. This system integrates temperature and humidity data into its detection algorithm to mitigate environmental influences on measurement accuracy. The optical path structure is optimized, and incident light intensity data are collected to compensate for variations in the light signals received by photosensitive elements. A synchronous demodulation circuit is designed to eliminate background light interference, thereby enhancing detection accuracy. Furthermore, a new calibration method reduces both calibration complexity and cost while further improving precision. Test results demonstrate that this system achieves high accuracy and robust anti-interference performance, with a maximum deviation of 1.67% under outdoor high-temperature conditions, outperforming traditional equipment by 0.55%. Additionally, the stability in low-temperature and high-humidity environments has been enhanced by approximately 2.9%, enabling longterm monitoring capabilities, meeting the operational requirements of photovoltaic power stations.

Abstract:Most of the existing gait recognition methods are based on silhouettes or skeletons, however, the 2D information lacks a complete description of the spatial geometry of the human body, and the performance of the recognition effect is limited under complex conditions such as view angle change and occlusion, for this reason, this paper proposes a point cloud gait recognition method that combines global multiscale and local fine-grained features. The method projects the point cloud as a depth gait map, introduces a cross-view data transformation module to improve the viewpoint invariance of the model, uses an improved residual network to extract rich global multi-scale gait features, and finally uses a KAN network to enhance the representativeness of local fine-grained gait features. The experimental results show that the gait recognition method based on point cloud is far better than the method based on 2D information, which achieves an average Rank1 accuracy of 92.65% on the SUSTech1K public dataset, which is a 6.02% improvement compared to the advanced method LidarGait, which fully verifies the effectiveness of the method.

Abstract:In response to the problem of poor detection accuracy in current transmission line inspections, a transmission line inspection algorithm based on YOLOv8 (GCAF-YOLOv8) is proposed. Firstly, designed a global channel spatial attention module GCSA to enhance the expressive power of input feature maps. This module combines channel attention, channel shuffling, and spatial attention mechanisms to capture global dependencies in feature maps. Secondly, introduce the StarBlocks structure from StarNet and fuse it with the existing C2f modules in Neck and Backbone to construct a new C2f_Star module, which improves the accuracy of object detection and reduces the overall parameter count of the model. Then, it is proposed to replace the traditional convolution in the baseline model with the ADown convolution module to improve the detection accuracy of subtle features. Finally, combining Focal Loss with the original CIoU in YOLOv8, a Focal CIoU Loss function is designed to solve the problem of class imbalance and improve the accuracy of detecting box position prediction. The experimental results show that the proposed GCAF-YOLOv8 model has improved detection accuracy P by 3.3% and average accuracy detection mean mAP by 3% compared to the original model. It can effectively detect various defects in power components and foreign objects on transmission lines.

Abstract:Breast cancer is one of the diseases that pose a serious threat to women′s health. Early diagnosis is crucial for the cure of breast cancer, and computer-aided breast cancer classification and diagnosis has been widely used. Although the mask autoencoders breast cancer classification method can improve the model performance under the premise of the lack of labeled data in breast cancer pathology images, the existing mask autoencoders breast cancer pathology image classification method does not adequately extract and fuse the feature information between breast cancer pathology images with different magnifications. To solve this problem, a multi-magnification mask autoencoders breast cancer pathology image classification method is proposed, which combines the advantages of magnification independence and magnification specificity on the basis of mask autoencoders. First, a uniform noise masked module is designed to avoid the loss of important features in breast cancer pathology images. Then, blocks of breast cancer pathology images with different magnifications are combined together and fed into an encoder incorporating cross convolution mapping to extract and fuse features from images with different magnifications. Finally, a residual cross attention mechanism is incorporated into the decoder to enhance the fusion of cell density and alignment order under low magnification images and cell texture features under high magnification images. Experiments on the BreakHis public dataset show that the proposed method improves at least about 2% in Top-1 Accuracy, Precision, Recall, and F1-Score compared to existing classification methods. The results demonstrate that the proposed method exhibits good performance in accurately classifying benign and malignant breast cancer pathology images.

Abstract:In recent years, convolutional neural networks have achieved great success in the study of building extraction from remote sensing images, but they still face problems such as low overall extraction accuracy, misclassification, omission, and fuzzy boundaries. Aiming at the above problems, a building extraction method based on VDSEC-UNet for remote sensing images is proposed. Firstly, VGG-16 is used as the encoder to extract the building feature information. Secondly, dynamic up-sampling is used instead of traditional up-sampling to enhance the model′s ability to perceive the details so as to improve the extraction accuracy of the building boundaries. Next, a multi-scale context information extraction module is embedded in the middle of the coder and decoder in order to take the influence of other objects around the building into account and introduce sufficient context information and global information under different sensing fields to reduce the loss of spatial information and enhance the extraction effect of buildings at different scales. Then, the ECA attention mechanism is embedded in each jump connection part to improve the model′s attention to the building features in the image. At the same time, the joint loss function is used to alleviate the category imbalance problem. Finally, the CA-DPGHead module is constructed and added at the end of the decoder to enhance the distinction between buildings and background so that the model can locate and identify the building information in the image more accurately, which in turn improves the extraction accuracy of small buildings and refines the extraction effect of building boundaries. The experimental results show that the mIoU of VDSEC-UNet on Massachusetts and Inria datasets reaches 82.07% and 84.35%, respectively, and the F1 index reaches 83.34% and 86.66%, respectively, which is better than other classical methods.

Abstract:Intelligent recognition of rice leaf diseases is of great significance in modern agricultural production. Focused on the issue that the traditional Vision Transformer network lacks inductive bias and is difficult to effectively capture the local detail features of the image, an improved Vision Transformer model was proposed. This model′s the ability to model multi-scale context as well as local and global dependencies was enhanced by introducing intrinsic inductive bias, while reduced the need for large-scale datasets. In addition, the multi-layer perceptron module in the Vision Transformer was replaced by the Kolmogorov-Arnold networks structure, thereby improving the model′s ability to extract complex features and interpretability. Experimental results show that the proposed model achieved excellent performance in the rice leaf disease recognition task, with an accuracy of 98.62%, which was 6.2% higher than the original ViT model, effectively improving the recognition performance of rice leaf diseases.

Abstract:Pedestrian re-identification is used to retrieve and recognize the same pedestrian in the non-overlapping fields of cross-camera. Aiming at the feature differences between different fields of cross-camera and the pseudo-label noise generated in the clustering stage, this paper proposes an attention-based unsupervised pedestrian re-identification method. In the feature extraction stage, an adaptive graph channel-spatial attention module (AGCBAM) is proposed, which considers both channel and spatial dimensions, adapts to cross-camera feature distribution by adaptively adjusting channel weights, and pays attention to specific spatial location features to capture details. In training stage, an improved intra-class neighbor spatial attention module is proposed, which combines label smoothing and spatial-level connections of positive instances to better remove pseudo-label noise and enable the model to better learn the real distribution of data. Through experiments on two mainstream datasets, Market-1501 and MSMT17, some existing common algorithms are compared, and the accuracy of mAP and Rank-1 is improved, which verifies the effectiveness of the proposed method.

Abstract:High-resolution imaging with near-field millimeter-wave radar typically relies on extensive data acquisition. Existing time-domain and frequency-domain imaging algorithms process signals under the condition of satisfying the Nyquist sampling rate, which imposes significant burdens on data collection and hardware costs. This paper proposes a millimeter-wave radar sparse imaging algorithm based on compressive sensing theory, leveraging the sparsity of the measured target echo signals to effectively reduce data requirements. The algorithm constructs a sparse model based on the sparsity exhibited by undersampled data in the wavenumber domain, and optimizes it to reconstruct the signal. A matched filtering algorithm is applied in the azimuth direction to achieve two-dimensional imaging of the target. Experimental results demonstrate that under conditions of data undersampling, the proposed algorithm can achieve high-resolution imaging of the target, significantly reducing data requirements. Moreover, the image quality outperforms other compressed sensing optimization algorithms in all metrics. Even under conditions where the target object is partially occluded, the algorithm can effectively restore the occluded portions of the image, demonstrating strong interference resistance and robustness.

Abstract:In this paper, we address the issues of low training efficiency and insufficient pose generalization ability in personalized 3D human avatar creation using neural radiation fields based implicit modeling techniques. We propose a novel method that combines 3D Gaussian splatting with parametric human models to provide an explicit representation. Additionally, we introduce a Point Transformer architecture based on attention mechanisms. This architecture can deeply learn and extract pose information from each frame and effectively integrate it into the Gaussian attribute parameters, thereby enhancing the rendering capabilities of the model. In experiments conducted on the People-Snapshot dataset, our method is compared with current state-of-the-art methods. Quantitative results show that our approach achieves an average PSNR of 29.53, which is a 13.7% improvement over the baseline method, demonstrating a significant advantage. Qualitative evaluations indicate that even with large avatar movements, our algorithm can effectively maintain the integrity and detail of the rendering results.

Abstract:Odometry is a crucial component of Simultaneous Localization and Mapping (SLAM) technology. However, most existing odometry algorithms rely on single data sources such as point cloud data or image data, failing to fully leverage multi-data fusion to improve trajectory estimation accuracy. Additionally, these algorithms often exhibit insufficient accuracy in complex environments and feature-deficient scenarios. To address these issues, this paper proposes a deep network called 3DPointLIO, which fuses LiDAR data and Inertial Measurement Unit (IMU) data. Firstly, a feature pyramid network combined with a weight attention mechanism is introduced to reduce the impact of dynamic information in the environment and enhance the robustness of point cloud features. Secondly, in the IMU data processing network, a convolutional network is integrated with Gated Recurrent Units (GRUs) to mitigate noise in raw IMU data, and a Bidirectional Long Short-Term Memory (BiLSTM) network is employed to extract temporal features from the denoised IMU data. Finally, a pose estimation network composed of multiple linear layers is used to estimate translation and rotation. The proposed algorithm is validated on the open-source KITTI dataset. Experimental results demonstrate that, compared to the baseline model, the proposed odometry algorithm improves rotation estimation by 0.76° and translation estimation by 2.17%. Furthermore, it outperforms other common odometry algorithms in both rotation and translation estimation, particularly achieving higher accuracy in rotation estimation.