Abstract:Volumetric tube calibration is an important process of the traceability of flowmeter, but the standards and benchmarks are different from each other in the countries, leading to the inconvenience for the users. In this paper the main technical contents between the domestic JJG regulations and international API standards for volumetric tube calibration are firstly compared. Then we systematically analyzes the volume value of the volumetric tube and repeatability calculation model. Based on the validation for the calibration data of the piston type volumetric tube in one foreign oil field, we demonstrate that the difference between the standard volume value calculated by the two standards (specifications) is very small (1ppm) but the repeatability has large difference (up to 200 ppm). The number of measurements has a direct impact on the repeatability. The repeatability differences of the API standard and the JJG standard are 50 ppm and 179 ppm, respectively. The analysis indicates the differences arise from Bessel formula of the JJG standard and the polarization method of the API standard free of the polarization coefficient and confidence probability. Through theoretical derivation and data validation, the coefficient of polarization and confidence probability can be reduced to the repeatability calculation formula to obtain a universal counteraction method. For the n≥5 measurements between API standards and JJG regulations, repeatability differences are small. This work can provide reference for the future revision of the regulations and method selection.

Abstract:In order to further improve the fault diagnosis accuracy of fiber optic composite submarine cable, a fault diagnosis method based on VMD and SO optimization SVM is proposed. Firstly, VMD was used to decompose the fault data, several IMF components were obtained, and Pearson correlation coefficient was used for further screening. Secondly, feature extraction is carried out on the selected IMF components to extract the kurtosis, approximate entropy and fuzzy entropy of each component respectively. Finally, the eigenvectors composed of the above eigenvalues are input into the SVM optimized by SO for training and classification, and the fault diagnosis results are obtained. The experimental results show that the fault recognition accuracy of fiber-optic composite submarine cable can reach 100% by using the optimized SVM method based on VMD and SO, which is 7.5%, 5%, 5% and 7.5% higher than that of SVM, GA-SVM, GGO-SVM and CNN respectively.

Abstract:In view of the complex background of solar cell image, changeable defect morphology and large scale difference, a method of solar cell defect detection based on SimAM-Ada Pool YOLOv5 algorithm was proposed. First, deformable convolution is incorporated into the CBL module to achieve adaptive learning of feature scales and perceptual field sizes; then, Ada Pool is incorporated into the SPP module to increase the degree of defect information retention; finally, the feature extraction capability of the model is further improved by introducing the SimAM attention mechanism. To further optimize and improve the YOLOv5 algorithm, the Mosaic and MixUp fusion data enhancement, K-means++ clustering anchor box algorithm, CIOU loss function, and Hard-Swish activation function are used to enhance the performance of the improved model. The experimental results show that the improved YOLOv5 algorithm achieves 89.86% detection mAP on the solar cell electroluminescence image dataset, which is 8.07% higher than the mAP of the original algorithm, with a speed of 37.92 fps, and can complete the solar cell defect detection task more accurately while meeting the real-time requirements.

Abstract:When the mobile robot sets the navigation point in the depth of the complex corner, the mobile robot can not plan the path because of the oscillation and rotation of the inaccurate positioning. This paper proposes a local planner based on TEB combined with lidar to collect corner information, and realize the rapid positioning and navigation of the mobile robot by setting auxiliary points. Through testing in the simulation platform Gazebo and the actual environment, the results show that the method can solve the positioning problem of complex corners, and realize the fast and effective path planning and accurate parking of mobile robots.

Abstract:Aiming at the problems of long transient time and large torque ripple in steady state in PI controller and traditional sliding mode method of switched reluctance motor, this paper proposes a control method based on adaptive reaching law and disturbance observer. Firstly, state variables are introduced to construct nonsingular fast terminal sliding surface to improve the convergence time of the system. Secondly, the inherent chattering of the traditional exponential reaching law is analyzed, and an adaptive reaching law algorithm is designed to reduce the system chattering. Then a nonlinear disturbance observer is designed according to the disturbance change, and the observed disturbance value is compensated to the sliding mode controller as a feedforward signal. Finally, compared with the traditional sliding mode control, the proposed control strategy is verified. The results show that the response time is shortened by 0.08 s and 0.4 s respectively in the start-up stage of simulation and experiment, and the speed tracking without overshoot is realized. In the loading stage, the speed decrease is reduced by 19.9%, 9.2%, 7% and 2.9% respectively, and the output torque fluctuation of the motor at steady state is reduced, and the torque ripple is reduced by about 10%.

Abstract:Aiming at the problem that specific emitter identification basically follows the closed set hypothesis and cannot effectively identify unknown classes, a specific emitter identification model based on zero sample learning is proposed to identify known and unknown specific emitter. By building a convolution neural network to extract the semantic features hidden under the emitter signal data, the attention module is introduced to enhance the focus on key features, and a combined loss function is proposed to separate different types of emitter signals in the semantic feature space, and the specific emitter classification and recognition are carried out according to the distribution of emitter signals in the semantic feature space. The experimental results show that compared with traditional closed set recognition, the proposed model can recognize unknown classes and distinguish between unknown classes while maintaining the recognition rate of known classes, with an average recognition rate of more than 90%. In engineering verification, the established unknown specific emitter identification platform can achieve fast and accurate recognition in indoor and outdoor scenes.

Abstract:Aiming at the problems of low estimation accuracy and slow convergence of existing array signal direction of arrival(DOA)estimation algorithms, a DOA estimation algorithm based on hyperbolic composite function approximating the l0 norm is proposed. Firstly, a hyperbolic composite function is given to approximate the l0 parametric number, and the problem of solving the l0 parametric minimum is transformed into the optimization problem of solving the hyperbolic composite function. Then, to improve the global convergence efficiency of the algorithm, the modified Newton method is used to optimally solve the hyperbolic composite function. The approximate l0 norm solution is obtained through the inner and outer loops of the algorithm. The outer loop provides the approximation factor for the inner loop, and the inner loop solves the modified Newton iterative expression according to the decreasing approximation factor. Finally, the optimal solution of the hyperbolic compound function is obtained, from which the DOA estimate is obtained. The effectiveness of the proposed algorithm is verified by simulation experiments, The results show that the root mean square error of DOA estimation is 0.685 6° for the proposed algorithm with a signal-to-noise ratio of 5 dB, and the estimation success rate is higher than 98%.

Abstract:To better evaluate the quality and safety of reinforced concrete structures used in power transmission and transformation, a corrosion detection method was proposed for reinforced concrete structures in power transmission and transformation based on the microwave transmission method. The accelerated corrosion test of steel bars was carried out, and reinforced concrete models with different corrosion degrees were obtained. A microwave detection test platform was built. The variation of the amplitude and phase of the transmission coefficient S21 was analyzed through microwave nondestructive testing of reinforced concrete models with different degrees of corrosion. The results show that the accelerated corrosion test in a wet salt sand environment can obtain steel bars with different corrosion degrees in a short time. When the lift-off distance is 4 mm, the corrosion state of steel bars in concrete can be effectively identified by the amplitude and phase change trend of transmission coefficient S21. The research can provide a reference for corrosion detection of power transmission and transformation-reinforced concrete structures.

Abstract:Multi-sensor data fusion technology is widely used because of its multi-scale and in-depth processing of data. In order to reduce the impact of conflicting data on the fusion accuracy, this paper proposes a multi-sensor data fusion method based on hybrid strategy. Firstly, the conflict factor K in Dempster-Shafer(DS) evidence theory is introduced to group the evidence, and the evidence is retained for low conflict evidence, and the high conflict data is weighted and modified. The weighted correction method uses information entropy and Bray-Curtis distance to calculate the uncertainty and difference of evidence, and synthesizes the two to obtain the corrected weighted evidence. Finally, the weighted evidence is fused according to DS combination rules, and then fused with low-conflict data to obtain the final result. The experimental analysis results show that the method can obtain correct results for various conflict situations, and the accuracy rate in the face of high conflict evidence reaches 98.12%. At the same time, in the application of fault diagnosis, the accuracy of this method reaches 89.98%, which proves the effectiveness and practicability of this method.

Abstract:To solve the problem of the formation control of multi-UAV based on the second-order consensus theory. Taking fixed-wing UAVs as the research object and presenting a formation control method of UAVs based on an improved secondorder consistency algorithm. To improve the consistency algorithm, by introducing a virtual leader that constructs a relative motion coordinate system capable of describing the formation directly. Simultaneously considering the characteristics of the UAV′s kinematic model with decoupled transverse and longitudinal directions, as well as the performance constraints of the UAV system. To accommodate the UAV formation transformation under varying scenarios by employing the KM algorithm to design a redistribution strategy for the positions of formation members during formation transformation. Additionally, using the improved L1 guidance law as the virtual leader’s trajectory tracking method. Finally, using simulation to verify the feasibility and effectiveness of the formation control method proposed in this paper. The results indicate that the method can form a stable formation while satisfying the performance constraints of the UAV system, and the method can quickly perform formation transformation with switching topologies and eliminate position errors across a variety of flight scenarios.

Abstract:Aiming at the problems of poor prediction accuracy and unstable prediction results of BP neural network, an improved slime mold algorithm (ISMA) is proposed to optimize the prediction model of BP neural network, and Tent chaotic mapping is introduced to overcome the shortcomings of uneven initial population distribution. The leader strategy and Levy flight strategy are introduced to solve the randomness of the position update and the problem of falling into local optimality. The adaptive reverse learning strategy is used to expand the search space and 23 groups of benchmark functions are tested. Then the ISMA algorithm was used to optimize the initial weights and thresholds of the BP network model, and the ISMA-BP Air quality index prediction model was constructed. At last, 779 sets of AQI data were collected and put into the prediction model for testing and analysis. The experimental results showed that, Compared with BP neural network model, GWO-BP model and SMA-BP model, ISMA-BP model has higher accuracy in predicting AQI. The mean square error of ISMA-BP model is 3.840 2, and the mean absolute error is 1.507 8 respectively.

Abstract:To address the shortcomings of the current traffic flow prediction model in extracting data features that easily ignore the differences in traffic flow trends between weekdays and rest days, a feature selection-based spatio-temporal fusion traffic flow prediction model STTF-XGB is proposed. The model improves the extraction of data features by the model from both data and model levels. First, Pearson correlation coefficient is used to calculate the correlation between data, and the data set is reclassified into weekday and rest day data sets according to the magnitude of the correlation. Secondly, the global spatial features of the road network data are extracted by using the adjacency matrix which can reflect the global relationship and the self-attention mechanism to build a graph self-attention mechanism, and the spatio-temporal feature extraction module with a "sandwich" structure is built based on the Transformer model to build a spatio-temporal fusion model STTF. Then, at the end of the STTF model, the XGBoost model is used to filter the features extracted by the multi-head attention mechanism to build the STTF-XGB model. Finally, the model was experimented on the UK freeway traffic flow dataset, and the results show that STTF-XGB can be effectively used for traffic flow prediction with about 5%~10% improvement in prediction accuracy over the GCN-BiLSTM and GAT-BiLSTM model in the short and medium term, and the prediction error fluctuation range is minimal.

Abstract:Aiming to address the insufficient diagnostic accuracy in existing rolling bearing fault diagnosis research, this paper proposes a rolling bearing fault diagnosis algorithm based on optimized intrinsic mode function adaptive noise-assisted ensemble empirical mode decomposition (CEEMDAN) and variational inference. First, the intrinsic mode function components of the original signal are obtained using CEEMDAN. A sensitive intrinsic mode function component screening algorithm is then designed to optimize the CEEMDAN method, generating a feature vector. For the training dataset, a Gaussian mixture model is established. Through variational inference, the Gaussian mixture model approximates the probability distribution of the feature vector to achieve rolling bearing fault diagnosis. The effectiveness of the proposed algorithm is validated through examples. Compared with CEEMDAN combined with variational inference, local feature scale decomposition combined with variational inference, and optimized CEEMDAN combined with particle swarm optimization support vector machine, the diagnostic accuracy is improved by 4.3%, 4.3% and 21.7%, respectively.

Abstract:Method for classification of point cloud is presented based on LiDAR function in order to solve the problem of fast and accurate classification of point cloud application of power-line inspection. Detect probabilistic of different kind of targets such as power-line, tower and ground target are built based on LiDAR function. Based on the presented model of detecting probabilistic model, point cloud is meshed and parameters of each mesh are calculated based on the detecting probabilistic model. Fast classification method is designed based the parameters of each mesh, in which point cloud from power-line, tower and ground target is extracted. In order to verify the effectiveness of the presented method of classification, serval groups of point cloud are applied in experiment of point cloud classification. According to the results of experiments, recall ration of the presented method can reach as high as 98%, and time-consuming of classification for one segment of power line can reach 14 s. Results of the presented experiments show that the presented method of classification of point cloud can reach higher accuracy and efficiency.

Abstract:When the livestock is weighed dynamically, the EMD algorithm is selected for processing according to the non-stationary characteristics of the dynamic weighing signal. In order to solve the endpoint effect in the process of EMD decomposition signal, an improved EMD algorithm based on segmented envelope fitting endpoint effect is proposed. The algorithm first uses the least squares method to extend the endpoints, and then divides the envelope into the internal segments, extension points and extreme point segments, constructs the inner end of the extreme point by using cubic spline interpolation between the extreme points, and constructs the extension point and the extreme point segment by connecting the extreme point and the extension point according to the continuous condition of the second derivative. This effectively simulates the endpoint effect while reducing the error introduced by the extension endpoint. The experimental results show that the average relative error of the proposed algorithm is 0.21% and the maximum value is 0.33% after comparing the dynamic weighing signal of livestock processed by the proposed algorithm with the real weight data, indicating that the proposed algorithm has good accuracy and stability when applied to dynamic weighing.

Abstract:Unequal capacitance grouped capacitors have high switching flexibility and excellent reactive power regulation ability, but there are problems with complex control strategies and frequent switching of equipment. Therefore, this article proposes a reactive power optimization configuration method that coordinates the contradiction between the two. Firstly, the probability distribution curve of reactive power demand in the substation is obtained according to the substation load, and then the optimal coverage method is used to calculate the reactive power mismatch area formed by the intersection of the distribution curve and the trapezoidal curve of capacitor compensation capacity. The smaller the area, the better the flexibility of capacitor regulation.Then considering the regulation complexity index, an optimization model of unequal capacitance grouping of capacitor is established with the minimum mismatch area and regulation complexity as the objective function. The multi-objective optimization model is transformed into a single-objective optimization model by the fuzzy weighting method. The improved genetic algorithm is used to solve the problem, and the reactive power configuration scheme of the substation is obtained, which can provide decision-making basis for reactive power planners. The final case simulation verifies that the proposed method in this paper can balance the advantages and disadvantages of unequal capacity grouping by comparing it with existing literature on equal capacity grouping and unequal capacity grouping.

Abstract:In order to solve the problem of low accuracy of segmentation model caused by the complex shape of liver tumor and blurred boundary with surrounding normal tissues in the liver tumor image, this paper proposes a novel liver tumor image segmentation model HFU-Net based on hybrid dilated convolutions and high-level feature fusion. In this model, a high-level feature fusion recalibration module is added to enrich the skip connection part of U-Net, so that it can calibrate the feature information by using feature fusion and squeeze and attention module to enhance the ability of network encoder to obtain feature information. And, in order to further improve the feature extraction effect of each layer of the network, the conventional convolution module in the original model’s encoding network is replaced by the hybrid dilated convolution to obtain dense tumor feature information and expand the network’s receptive field. The experimental results show that Dice coefficient, volumetric overlap error (VOE), sensitivity and precision are improved by 3.3%, 4.59%, 4.39% and 2.04% respectively compared with the U-Net algorithm. The proposed model significantly improves the segmentation precision of liver tumor images, and provides a reliable basis for the diagnosis and treatment of liver cancer.

Abstract:A laser inertial SLAM system based on the fusion of lidar and inertial unit (IMU) is proposed to address the issue of inaccurate pose estimation using a single sensor laser odometer when unmanned vehicles are mapping in outdoor large scene environments, and the accuracy may decrease with accumulated drift. The front-end of the system is assisted by IMU information to remove distortion from the point cloud, and through point cloud registration, it forms a LiDAR odometer. The back-end optimization is realized by factor graph, which is jointly optimized by the front-end odometer factor, IMU pre Integrating factor and loopback detection factor. At the same time, this article proposes an improved fast loop detection method based on the global descriptor (Scan Context), which can effectively improve the accuracy and accuracy of loop detection while ensuring real-time performance. The results of publicly available datasets and unmanned vehicle experiments show that compared to the classic laser algorithms A-LOAM and LeGO-LOAM, the trajectory accuracy of the proposed method in this paper has been improved by about 40%, and the efficiency of loop detection has been improved by about 25%, effectively improving the performance of the SLAM system.

Abstract:Aiming at the problem that it is difficult to extract key features clearly due to the cluttered change information of multiple feature types in different time-series remote sensing images and complex backgrounds, this paper proposes a new method of fusing Swin Transformer with twin networks to achieve building change detection. The method obtains features at different levels through the structure of four Swin Transformer blocks, and performs difference calculations for feature maps at different scales to obtain change feature maps. In addition, a difference feature fusion module and an edge-aware attention module are introduced based on the algorithm in this paper. The difference feature fusion module can better express the features under different perceptual fields and improve the fusion effect on detailed features and global features; the edge-aware attention module refines the edge features of buildings in the feature map during feature extraction, expands the local perceptual field of the model, enhances the detection ability of the model for detailed information, and thus improves the extraction ability of the network structure for building edge features. The experimental results show that the F1 values of this paper′s method are improved by 7.36% and 19.67% on two public datasets, respectively, compared with the existing classical change detection network FC-EF.

Abstract:The Back Projection (BP) algorithm does not use approximation in the imaging calculation process, and the imaging quality is high. Any array configuration is suitable for imaging. In recent years, it has been widely used in the field of radar imaging technology. However, in millimeter wave three-dimension holographic imaging, the computational efficiency is low, which affects the implementation of real-time imaging. Under the conditions of three-dimensional polar coordinates, the Fast Factorization Backward Projection (FFBP) algorithm uses sub-aperture division for imaging, which solves the real-time imaging problem to a certain extent. This article implements the FFBP algorithm based on the four threaded CPU and GPU accelerated CUDA platform, and compares and analyzes the imaging of multi-point targets. The results are basically consistent, thereby verifying the effectiveness of the accelerated algorithm. Furthermore, through electromagnetic simulation software, the resolution board is modeled and simulated to simulate real targets, and GPU accelerated imaging is performed. The calculation time is 33.97 times faster than the four threaded CPU, making it suitable for 3D near-field real-time imaging systems and better applied in the field of human security inspection.

Abstract:Acoustic detection technology has gradually become a key means of underwater target detection. Due to factors such as underwater hydrological environmental noise and equipment accuracy, sonar images inevitably have problems such as low resolution, low contrast, and blurred target edges, which are not conducive to subsequent target detection and recognition. In this paper, we propose an improved denoising algorithm that combines adaptive guided filtering (AGF) with three-dimensional block matching (BM3D). This algorithm uses the BM3D algorithm to suppress gaussian and speckle noise in the image for initial denoising; then, the AGF algorithm is used for secondary filtering of the image. At the same time, by introducing the improved edge detection Canny operator, we optimize the guided filter by adaptively adjusting the size of the regularization parameter to retain more image details and edge features. The combination of the two algorithms not only optimizes the shortcomings of the BM3D denoising performance but also effectively retains the edge features of the image. Experimental results show that the proposed algorithm not only has good suppression effects on speckle noise and Gaussian noise in sonar images but also improves the peak signal-to-noise ratio (PSNR), mean square error (MSE), and structural similarity(SSIM) index by 10%, 15%, and 15%, respectively, compared to other traditional algorithms.

Abstract:Aiming at the issue of small target defect detection and multi-size defect detection in X-ray images, a weld defect detection algorithm based on improved Faster RCNN is proposed. Firstly, the algorithm utilizes ResNet50 and feature pyramid network as the backbone network of Faster RCNN for detecting defects of different sizes on multiple feature maps. Then, the background subtraction layer is added before the backbone network to reduce the interference of the image background on the small target defects. Then, the three-branch region proposal network layer refine the predictions of candidate boxes in the original region proposal network layer, thereby reducing the number of candidate boxes and optimizing the detection speed. Finally, the number of convolutional layers in the network is fine-tuned to enhance the network’s feature extraction ability. The experimental results show that the improved network has a mean average precision of 83.09% and a single image detection speed of 20.8 ms. Compared to the network before improvement, the preset anchor boxes are increased by 10 779, and the mean average precision is increased by 19.43%, while the detection speed is only decreased by 3.1 ms. The improved network effectively improves the detection effect of small target defects and multi-size defects while maintaining detection speed.

Abstract:The study and application of high-throughput plant height data acquisition technology for lettuce crops are limited. A lettuce plant height detection method based on deep learning and drone oblique photography is proposed to address this. Firstly, oblique photography by drone is used to obtain high-throughput plant height data, and a 3D model of plants within the region is generated to extract elevation information. Then, an improved YOLOv5 algorithm with a CBAM attention mechanism embedded in the C3 module of the backbone network is proposed. This algorithm is designed to reduce shallow noise information, enhance the detection capability of small and dense targets, and achieve target detection of plants in the region. This will result in estimated plant heights for each plant. The experimental results show that the CBAM-YOLOv5 model significantly improves the recognition effect, increasing the AP value for lettuce crop recognition to 96.19%. Compared with the original YOLOv5 model, the AP value of our model has increased by 1.5%. The plant target detection has a high correlation between the estimated values calculated from the 3D model and the measured values, with a linear slope of 0.991 1 and R2-value of 0.931 1, achieving the detection of highthroughput plant height data for lettuce crops.

Abstract:Road obstacle detection is a crucial component of automatic driving environment perception. To enhance the precision of existing road obstacle detection algorithms, we propose an improved YOLOv5s road obstacle detection algorithm. The improved coordinate attention module filters invalid information from multi-scale feature maps and strengthens the focus on areas of interest. Additionally, the enhanced downsampling module alleviates the loss of essential information during sampling in the fusion network, thereby increasing feature robustness. The optimized algorithm′s regression loss and wise gradient gain allocation strategy improve the contribution of common mass anchor frame loss. Experimental results demonstrate that the improved model′s average accuracy on the dataset has increased by 4.2% to 78.6%, outperforming Fast R-CNN, YOLOX, YOLOv7, and other algorithms. With a detection speed of 42 frames per second, the algorithm meets real-time detection requirements. Therefore, the proposed improved algorithm in this study can effectively improve the accuracy of road obstacle detection and has practical application potential.

Abstract:Aiming at the problem that the tracking algorithm based on twin network is prone to degradation under the condition of fast moving target, large deformation and complex background, a target tracking algorithm based on the integration of attention mechanism and adaptive template updating is proposed. Based on SiamRPN, the tracking algorithm combines the channel attention mechanism and the spatial attention mechanism in the feature extraction network to suppress the interference information in the image, supplement the target feature information in the channel space, and better locate the target. The template of the object at different times, including the initial template, the accumulated template and the predicted template, is taken as the input of the residual module. The residual learning strategy is adopted to make full use of the semantic information of the initial template and adaptively update the template needed for the current frame, which reduces the phenomenon of tracking drift. Experimental results on the OTB100 dataset show that the proposed tracking algorithm achieves higher tracking success rate and accuracy compared with other tracking algorithms.