The digging force of hydraulic excavator is an important parameter reflecting the working conditions of the excavator. In this study, by analysis of the working mechanism of the excavator, a mathematical model reflecting the relationship between the cross-section stress of the linkage mechanism and the digging force of the bucket was established, and a new method for measuring the digging force was proposed. Furthermore, the cross-section stress values of the linkage mechanism under different digging forces were obtained by finite element simulation, and the error between simulation results and theoretical results was about 8%, which verifies the feasibility of the proposed measurement method for the digging force of hydraulic excavator.

In object tracking, the traditional correlation filtering algorithm is unable to perceive the change of scale aspect ratio for moving targets, and it is easily affected by a complex environment, resulting in tracking failure. Therefore, a spatial-temporal regularized correlation filtering algorithm with adaptive aspect ratio(AAR-SRCF) was proposed. Firstly, the average peak-to-correlation energy(APCE) and peak score were used as references to weigh and fuse each feature response map to achieve accurate results. Additionally, a set of novel one-dimensional boundary filters were presented, integrating near-orthogonality and spatial regularization. These filters can adaptively detect changes in the target scale and aspect ratio by precisely locating the boundaries of the target's bounding box. Moreover, spatial regularization effectively mitigated the negative impact of the boundary effect for boundary filters. Finally, the learning rate of each boundary filter was adjusted separately according to the peak-to-sidelobe ratio (PSR) to prevent the model from degradation. Through extensive experiments on OTB datasets, the proposed algorithm shows excellent tracking performance, achieving better results than other excellent algorithms in each challenge attribute.

To address the challenge of automatic recognition of electronic components on an assembly line, an improved YOLOv5 was used to implement instance segmentation of four categories of electronic components. Firstly, multi-channel histogram equalization was used for image preprocessing. Then, the YOLOv5 was improved: Segmentation head was added; Sequeeze-and-excitation net(SE-Net) channel attention module was embedded to enhance the feature extraction capability and to compress the useless information without increasing the model complexity; GhostNet was used to make the model lightweight; and BiFPN was used to enhance model feature fusion capability. Finally, experimental results showed that the mAP of the proposed method could reach 96.7% and the detection time of a single image was 45.5 ms. The results prove that proposed method has superior performance than that based on mask region-based conventional neural network(Mask RCNN) and initial YOLOv5, and has practical significance for automatic detection of electronic components.

Aiming at the problems of edge blur and distortion in the current damaged face image inpainting, a two-stage hierarchical gated convolutional network(HGCN) was proposed and then combined with edge adversarial network for face image inpainting. Firstly, the edge adversarial network was adopted to generate edge images. Secondly, the edge images, the masks and the occluded images were combined to train the generative adversarial network (GAN) model of the HGCN to generate the inpainted face images. In the HGCN, traditional convolution was replaced by gated convolution and the dilated convolution was introduced. The main structure of the HGCN is composed of coarse inpainting module and fine inpainting module. In the coarse inpainting module, the encoder and decoder network structure was used for coarse inpainting. In the fine inpainting module, the attention mechanism was introduced to enhance the feature extraction ability so as to further refine the inpainting results. In the experiment, the Celeba-HQ dataset and NVIDIA irregular mask dataset were used as the training datasets, the gated convolution network and attention module were adopted as comparing networks, and PSNR, SSIM and MAE were used as evaluation indicators.The experimental results demonstrated that for the face images with missing areas less than 20%, the proposed network works better than the two other networks on the above three indicators, and for the face images with missing areas greater than 20%, the proposed network is close to the comparison networks on three indicators. In terms of visual effects, the proposed method also surpasses the two contrasting networks in details. The proposed network can evidently improve the inpainting effect, especially image details.

It is difficult to fully mine the information from that one-dimensional vibration signal that expresses the state characteristics and then early recognize the wear of the valve plate of a piston pump. In view of the excellent image processing capabilities of the convolutional neural networks(CNN), we proposed an optimized VMD-CWT-CNN model to solve the above-mentioned problem. Firstly, continuous wavelet transform(CWT) was used to preprocess the signal to obtain a two-dimensional time-frequency diagram of the signal, which iwasused as one input of the CNN model to convert the state recognition problem into a CNN image recognition problem. Secondly, after optimizing variational mode decomposition(VMD) parameters based on correlation coefficient, the vibration signal was preprocessed by using the optimized VMD, and then based on the principle of maximizing the correlation coefficient and the kurtosis value, three groups of Intrinsic mode function(IMF) with fault characteristics were selected and reorganized into a three-channel one-dimensional signal as another input of the CNN model. Finally, in the CNN model, two paths were converged, and the results of the recognition and classification of the valve plate wear states of the piston pump were obtained. In the experiment, the proposed method we first use the optimized VMD and the CWT to preprocess the vibration signal, respectively, and then combined with the CNN to classify the wear states of valve plates. Experimental results show the recognition effect of the proposed method on the three states of valve plate wear is significantly better than that of the single-input CNN model, the typical deep learning method and the machine learning classifier. The optimized VMD-CWT-CNN method can more accurately recognize the valve plate wear states of the piston pump.

Iris verification has gained extensive attention because of its uniqueness, stability, and non-invasiveness. Deep learning techniques have made significant progress in the field of iris verification. By using convolutional neural networks (CNNs), features of iris images can be automatically extracted and learned, enabling high-precision identity verification. However, challenges such as intra-class variability and limited dataset size can compromise verification accuracy. To address these issues, we proposed an iris verification method based on dynamic data augmentation and contrastive learning. Four data augmentation strategies were carefully designed for online iris enhancement and dataset expansion, and the accuracy of iris verification was further improved by using data augmentation probability scheduler(DAPS). The MobileNetV3 was employed as the backbone network, which was optimized with contrastive learning for 3-channel iris pairs. The proposed method was evaluated on two benchmark iris databases, CASIA-V4-Interval and CASIA-V4-Thousand, achieving high accuracies of 99.85% and 98.82%, respectively. Experimental results demonstrate that the proposed method can achieve competitive performance with a small number of training samples.

In view of the problem of multiple peak values of P-U characteristic curve under a local shading environment,the traditional gray wolf optimization (GWO) is slow in convergence speed and low in accuracy of steady state at the late stage when tracking the maximum power point. Combining the advantages of GWO and perturbation & observation (P&O) method,an improved hybrid maximum power point tracking(MPPT) algorithm based on GWO-P&O was proposed. Firstly, optimized by the GWO, the algorithm was gradually close to the global MPPT. Then, P&O was introduced into the GWO at the late convergence stage, so that the local maximum power point of of photovoltaic power can be found at a faster speed while maintaining a high steady-state accuracy of the GWO,which overcomes the shortcomings of the traditional GWO algorithm. Finally, the proposed method was compared with the GWO under different environments. The results show that the proposed GWO-P&O method can improve the convergence speed in the late stage of the GWO when tracking the maximum power while ensuring high steady-state accuracy.

The actual working environment of unmanned excavation robot is harsh. In order to improve the trajectory tracking accuracy of bucket under load disturbance, a nonlinear mathematical model of electro-hydraulic system of digging robot was established, and a sliding mode controller(SMC)based on linear extended state observer(LESO), called SMC-LESO, was designed. Based on the displacement signal of the piston rod of the bucket cylinder, the velocity, the acceleration, the load disturbance and uncertain factors of the system were estimated by the LESO. On this basis, SMC-LESO was completed, and the Lyapunov stability of the controller was proved. The co-simulation model of electro-hydraulic proportional control system of the excavator was established. Compared with proportional-integral-derivative (PID) controller and SMC, the simulation results show that the designed controller can effectively suppress the disturbance, and has high displacement tracking accuracy and robustness.

Motivated by the goal of enhancing the accuracy and robustness of visual inertial navigation systems(VINSs) across a wide spectrum of dynamic scenarios, protracted missions and expansive navigation ranges, we designed a monocular visual inertial odometry (VIO) augmented by planar environmental constraints. To attain efficient feature extraction and precise feature tracking, we employed a methodology that involved the extraction and tracking of uniformly distributed using features from accelerated segment test(FAST) feature points from video images, with the subsequent removal of outliers through symmetric optical flow. Additionally, we outlined the process of identifying coplanar feature points from the sparse feature set, enabling efficient plane detection and fitting. This approach constructed spatial geometric constraints on the three-dimensional coordinates of visual feature points without resorting to computationally expensive dense depth mapping. The heart of this method lied in the formulation of a comprehensive cost function, which integrated the reprojection error of visual feature points, the coordinate constraints derived from coplanar feature points, and the inertial measurement unit(IMU) pre-integration error. These integrated measurements were then utilized to estimate the system states through a nonlinear optimization methodology. To validate the accuracy and effectiveness of the proposed approach, extensive experiments were conducted using publicly available datasets and large-scale outdoor scenes. The experimental results conclusively demonstrate that compared to VINS-Mono and ORB-SLAM3, the proposed method achieves higher positioning accuracy. It can deliver precise and stable navigation results even in challenging conditions, thereby imparting significant practical value to the fields of robotics and unmanned driving.

When three-dimensional electric field sensor(3D EFS) with orthogonally arranged capacitive-type sensing units is used to measure space electric fields, measurement accuracy is liable to be affected by the coupling effect between axes. In this study, an electric field shielding electrode was proposed to reduce the interaxial coupling effect in 3D EFS and improve the measurement accuracy. Firstly, the multiphysics field simulation software was used to construct an electric field model. Then, the capacitive-type sensing units of the 3D EFS with shielding electrode was developed by simulation results. Finally, an arbitrary angle test platform was set up to experimentally test the 3D EFS with shielding electrodes and the 3D EFS without shielding electrodes. The experimental results showed the measurement deviation of the 3D EFS with shielding electrodes was within 3.2%, which was 12% lower than that of the 3D EFS without shielding electrodes. It can be concluded that the 3D EFS based on the electric field shielding structure can make the decoupling matrix more reliable and reduce the measurement deviation of space electric field.

Linear Hall sensors are widely used to measure magnetic field strength, but there are few studies on the dynamic response characteristics of Hall sensors. To tackle this issue, a testing platform for the frequency response characteristics of linear Hall sensors was built, which was composed of a controllable constant current source, coils, linear Hall sensors, a low noise amplifier, and a data acquisition device. The system transfer function was constructed and a method of dynamically updating the transfer function was proposed, which realizes the accurate measurement of the dynamic response characteristics of Hall sensors. The dynamic characteristics of NHE520F and P3A were tested on this platform. The results showed that the performance differences in amplitude-frequency and phase-frequency characteristics of these Hall sensors in the range of 2.5 kHz-2 MHz were fully reflected on this testing platform. The dynamic characteristic parameters of Hall sensors were not necessarily consistent with the static characteristic parameters of Hall sensors, and the distributions of dynamic characteristics were also different. Additionally, according to the amplitude-frequency and phase-frequency characteristics of Hall sensors measured under various temperature and humidity conditions, the average dynamic characteristic curves and three standard deviation envelope curves of Hall sensors were plotted. The data obtained by this testing platform are of great significance for the research of dynamic response characteristics of Hall sensors.

The on-board equipment as core equipment of train control system plays an important role in the process of high-speed train operation. At present, its fault diagnosis only depends on the experience of on-site operators and diagnosis efficiency is relatively low. To realize automatic fault diagnosis and improve diagnosis efficiency of the on-board equipment of train control system, a fault diagnosis model called BERT+CNN_BiLSTM was proposed, which combined bidirectional encoder representations from transformers(BERT) model, convolutional neural network(CNN) and bidirectional long short-term memory(BiLSTM). Firstly, the BERT model was used to transform the application event log(AElog) into a text vector representation that can mine semantic information recognized by computer. Secondly, CNN and BiLSTM were used to extract fault features and combine them to enhance spatial and temporal capability of the model. Finally, fault classification and diagnosis of on-board equipment of train control system was realized by using Softmax. In the experiment, taking an actual on-board equipment as the research object, the AElog generated during the train operation was selected as texperimental data to verify the performance of BERT+CNN_BiLSTM model. The results showed that compared with traditional machine learning algorithm, BERT+BiLSTM model and BERT+CNN model, the pecision, recall and F1 of BERT+CNN_BiLSTM model were 92.27%, 91.03% and 91.64%, respectively, which indicates that the proposed BERT+CNN_BiLSTM model has a better overall performance in the fault diagnosis of on-board equipment of high-speed train control system.

Aiming at the problems such as low accuracy and poor robustness of target detection caused by missed detection of small road targets and occlusion between targets in UAV images, an improved road target detection algorithm based on YOLOv5 combining convolutional block attention module(CBAM), called YOLOv5s-FCC, was proposed. Firstly, a small target sensing layer was introduced to improve the multi-scale model, and a small target YOLO detection head was added to improve the feature extraction ability of the network for small road targets. Secondly, the CBAM fused space and channel information to enhance important information in the network after it was introduced into different locations of the Backbone network to obtain the best fusion location of CBAM. Finally, CIoU loss function was used to improve the speed and accuracy of the calculation required for predicting the bounding box of image. The experimental results showed that compared with YOLOv5 algorithm, YOLOV5-FCC algorithm can improve mAP50 and mAP50-95 by 2.0% and 4.2%, respectively. The effectiveness of YOLOv5-FCC algorithm was also verified on VisDrone dataset, and the results showed that the established system can realize automatic detection of road targets.

In the process of crack detection in subway tunnels, it is difficult to detect tunnel cracks due to the complexity of tunnel environments and the limitation of light conditions. To this effect, a tunnel crack detection method based on multi-feature analysis was proposed. Firstly, the quality of the tunnel crack image was improved by the preprocessing algorithm combining Retinex smoothing and piecewise linear stretching, and then the image was preliminarily segmented by Otsu threshold algorithm for block processing. Secondly, the area and rectangularity of connected domain in the image were analyzed, the linear structural features in the image were extracted by probabilistic Hough transform, and the pseudo crack interference was filtered out by image skeleton feature extraction algorithm. Finally, real crack detection was realized, and the detection rate of traditional crack image and tunnel crack image reached 92% and 86%, respectively. It is experimentally verified that the proposed method is practical and effective.