Abstract:When towing for rescue in certain hazardous environments, it is difficult for rescue personnel to approach. Rescue personnel can use remote control to operate the trailer bar to complete the installation of the trailer hook. This paper proposes a trailer hook detection and positioning method ECSA-YOLOv5 for rescue vehicles. Firstly, the YOLOv5 algorithm is improved by designing an efficient attention module ECSA, which replaces the module on the previous layer of the spatial pyramid pooling module. Additionally, a small object detection layer of 160×160 is added to obtain the pixel coordinates of the trailer hook in the image more accurately; By incorporating guided filtering in the preprocessing stage of the SGBM stereo matching algorithm and introducing weighted least squares (WLS) filtering and outlier handling in the post-processing stage, a more optimized disparity map can be obtained, resulting in more accurate target depth information and improving the accuracy of trailer hook position information calculation. Experimental verification was conducted based on the Jetson Agx Xavier development board, and the results showed that the ECSA-YOLOv5 model improved the AP value by 5.8% compared to the YOLOv5s model, reaching 99.0%. The average realtime detection frame rate was 14 fps, and when the positioning distance was within 3 meters, the error was below 3.5%, which can meet the accuracy and real-time requirements of trailer hook detection and positioning.

Abstract:It is a challenging and meaningful task to achieve more accurate emotion recognition. Because of the complex diversity of emotions, it is difficult to measure emotions comprehensively and objectively with a single mode of EEG signal. Therefore, a multi-modal lightweight hybrid model PCA-MWReliefF-GAPSO-SVM is proposed in this paper. The hybrid model consists of a PCA-MWReliefF feature channel selector and a GAPSO-SVM classifier. Electroencephalogram (EEG), electromyographic signal (EMG) and temperature signal (TEM) were used for emotion recognition. Through many experiments on DEAP public data set, the classification accuracy of 97.500 0%, 95.833 3% and 95.833 3% in titer dimension, wake dimension and four categories were obtained, respectively. The experimental results show that the proposed mixed model can improve the emotion recognition accuracy and is significantly better than the single mode emotion recognition. Compared with the recent similar work, the hybrid model proposed in this paper has the advantages of higher accuracy, less computation and fewer channels, and is easier to be applied in practice.

Abstract:Every mode of orbital angular momentum vortex beam is orthogonal to each other, which can solve the problem of spectrum shortage well. In order to solve the problem of narrow bandwidth of orbital angular momentum antenna, an ultra-wideband orbital angular momentum four-arm spiral antenna in terahertz band is designed. The relationship between the continuous phase difference of the feed ports and the generated modes is studied. A graphene double ring structure is used and the antenna performance is improved by adjusting and optimizing the antenna structure size. The experimental results show that the vortex beams with mode number 0, 1, 2 and 3 can be generated by simply adjusting the phase difference, and the gain under different mode number is above 7.5 dBi. At the same time, the absolute bandwidth of the designed antenna reaches to 8.85 THz, and the relative bandwidth at the center frequency of 6 THz can reach to 147% with the S11 is -50 dB. The designed antenna is greatly improved compared with the traditional antenna, and provides a certain significance for the terahertz band mode multiplexing in practical application.

Abstract:In view of the problems of low driving capability and high output ripple in traditional LDO with charge pump and NMOS as power transistor, a high-current LDO with isolated AC-DC loops was designed based on Huahong 0.35 μm BCD process. The demand for charge pump driving capability in this LDO is reduced by isolating the DC loop and AC loop, thereby ensuring low ripple in the gate driving voltage of the NMOS power transistor and achieve high current output. The PSRR of LDO is enhanced by adding ripple current absorbing circuit. The results show that in the input voltage range of 3.41~5.5 V, the output voltage of LDO is 3.3 V and the output current can reach 3 A. The PSRR of LDO under light load is 111.261 dB@DC, 86.900 5 dB@1 kHz, 78.947 2 dB@1 MHz. The PSRR under heavy load is 111.280 dB@DC, 84.123 1 dB@1 kHz, 39.263 8 dB@1 MHz.

Abstract:Solid-state circuit breakers require multiple IGBT in series to cut off the short-circuit fault current, and a hybrid voltage-balancing control circuit topology is proposed to address the problems of uneven voltage distribution, high local voltage, and high loss of multiple IGBTs. Analyze the influence factors of voltage equalization of multiple IGBTs in solid-state circuit breaker, and study the performance of voltage balance topology. Optimize the buffer circuit structure, improve the charging and discharging buffer circuits, and reduce the loss; introduce the dual-threshold clamp control circuit to improve the IGBT over-voltage; and propose a hybrid balancing voltage control strategy that combines the passive voltage equalization and the auxiliary feedback active voltage equalization, which accelerates the response speed and achieves the dynamic adaptive regulation of voltage equalization. The prototype is fabricated, and the simulation and experimental verification of the solid-state circuit breaker design topology and control are carried out, and the results show that the hybrid balancing voltage control circuit can reduce the over-regulation of the IGBT circuit, have stronger over-voltage suppression capability, and improve the response speed.

Abstract:A method based on attenuation tomography is proposed for the high-precision and anisotropic reconstruction of underground shallow seismic energy field in response to the demand for high accuracy and anisotropy in the process. Firstly, the propagation path of underground seismic waves is calculated through ray tracing based on the velocity field model; Secondly, the logarithmic spectral ratio method is used to invert the quality factor of underground media; Then, the absorption and attenuation coefficients of seismic waves at different propagation distances are calculated; Finally, the absorption attenuation and geometric diffusion attenuation of seismic waves are used to obtain the energy field of shallow underground vibrations. The simulation results show that the root mean square error between the reconstructed vibration energy field and the forward result is 7.653, and the maximum relative error between the reconstructed energy value of a single grid and the forward result is 15.3%, with an average relative error of 3.4%. The proposed method can achieve the reconstruction of the shallow underground vibration energy field, and has certain application reference value for the reconstruction of the shallow underground vibration energy field.

Abstract:A Chip-less Radio Frequency Identification (CRFID) metal crack sensor with a single resonant structure is proposed to simultaneously detect crack width and direction. The circular and the rectangle with tangent corners resonators are integrated into the design, and the electromagnetic simulation software HFSS is utilized to optimize the structure and simulate the performance of the sensor. The response characteristics of Radar Cross Section (RCS) of the sensor under different defect cracks are studied systematically. It is found that frequency shift is proportional to the crack width. The results show that the crack direction and crack width on the metal structure can be identified by the change of the resonant frequency of the sensor. The frequency shift direction in RCS amplitude-frequency characteristics in the two polarization directions correspond to the direction of the crack, and the resonant frequency shift is proportional to the crack width. The CRFID sensor can detect cracks in submillimeter width in three directions: 0°(horizontal), 90°(vertical), 45° or 135°(oblique), among which the crack detection sensitivity can reach up to 43.5 MHz/0.1 mm.

Abstract:In order to solve the problems of local minimum and unreachable target in traditional artificial potential field method, an artificial magnetic field method is proposed. Firstly, an artificial magnetic field is established around the obstacle, and the Lorentz force is introduced. The Lorentz force is perpendicular to gravity, so as to avoid the robot falling into the local minimum point; Secondly, the distance factor between the robot and the target point is introduced into the Lorentz force function to solve the problem of target inaccessibility; Thirdly, the direction of Lorentz force is optimized by establishing a virtual line between the target point and the obstacle, avoiding detours and reducing the number of steps of path planning. Finally, the traditional artificial potential field method and artificial magnetic field method are simulated in the MATLAB environment. The experimental results show that the artificial magnetic field method can overcome the local minimum problem and the target unreachable problem, and the planned path effectively avoids obstacles, preventing situations of oscillation and hesitation, thus improving the quality of path planning.

Abstract:In the context of local shading affecting photovoltaic arrays, the traditional maximum power point tracking algorithms exhibit slow convergence, poor accuracy, significant power fluctuations, and a susceptibility to getting trapped in local optima. For this reason, a composite algorithm based on the combination of a novel adaptive cuckoo algorithm and particle swarm algorithm was proposed. The method introduced adaptive discovery probability and adaptive L-vy flight step control factor into the cuckoo algorithm, and also incorporated the opposing population strategy in order to improve the algorithm′s convergence speed and global optimization seeking ability. In the early stage of the algorithm, the global search with particle swarm algorithm was used to quickly find the vicinity of global maximum power point (GMPP), and in the later stage, the new adaptive cuckoo algorithm was used to accurately search for the optimization in the local range in order to quickly, accurately, and stably track to the global maximum power point. The simulation results show that the convergence time and tracking error of the algorithm proposed in this paper are 0.106 s and 0.012%, 0.108 s and 0.034%, 0.110 s and 0.059%, and 0.106 s and 0.031%, respectively, for the four lighting modes, which are better than the other algorithms, and it validates that the algorithm in this paper has the fastest convergence speed, highest tracking accuracy, minimal power fluctuations, and the least likelihood of getting trapped in local optima among the six compared algorithms.

Abstract:Aiming at the problem of fixed Gaussian window function in the S transform, a Gaussian window adjustment factor is introduced to improve the algorithm, and a harmonic and interharmonic analysis method for photovoltaic power station based on generalized S transform is proposed. Firstly, the simulation model of the photovoltaic power station is built by MATLAB/Simulink, and the three-phase output voltage of the inverter is collected under natural sampling bipolar SPWM modulation mode. Then, taking the A phase voltage as an example, the generalized S transform is used to process the voltage signal to obtain a modular time-frequency matrix. Finally, the harmonic and interharmonic parameters of the photovoltaic power station are accurately calculated, which is achieved by analyzing the matrix. Simulation results show that the maximum amplitude calculation error of the method is only 3.30×10-4%, the average calculation error of frequency is 0%, which is much smaller than the S transform amplitude error of 35.19% and frequency error of 2.39%. The need for detection accuracy of harmonics and interharmonics can be met in photovoltaic power station.

Abstract:As an important part of automobile transmission device, the machining quality of synchronizer tooth hub has a direct impact on the performance and reliability of the transmission. Aiming at the problem of low efficiency in judging the range of tooth hub error source by manual experience, this paper proposed an error tracing method based on bat algorithm to optimize BP neural network. The error sources in the tooth hub machining process were analyzed, and the bat algorithm was used to optimize the weights and thresholds. The BA-BP error tracing model was constructed after obtaining the optimal value, data samples were collected to verify the model and compared with the error traceability method of BP neural network before optimization. Compared with the accuracy of the BP neural network traceability model before the optimization was 83.56%, the optimized accuracy was 96.34%, which significantly improved the traceability accuracy，this method allows the production personnel to trace the error causes of the subsequent out-of-tolerance workpieces, which is convenient to directly deal with and eliminate the problems in the production process, so as to improve the production efficiency.

Abstract:Resampling in standard particle filters can lead to particle impoverishment, affecting the accuracy of tracking systems. To overcome this deficiency, an improved cuckoo search algorithm-based optimization method for particle filtering in multi-object tracking is proposed. In this method, particles are treated as host nests for cuckoo birds, simulating the behavior of cuckoo birds in locating nest positions. The algorithm consists of two stages: global search and local search, which collectively guide particles towards high likelihood regions. Furthermore, enhancements are made to the cuckoo search algorithm, introducing dynamic search step sizes and reinforcing the local search mechanism, thereby improving the convergence speed of the algorithm in global search. Additionally, the improved algorithm incorporates joint probability data association for addressing multi-maneuver object tracking problems. Two sets of experiments are conducted in one-dimensional and two-dimensional environments to compare the tracking performance of the optimized particle filtering algorithm with the standard particle filtering algorithm. The experimental results demonstrate that the algorithm proposed in this paper exhibits not only faster global convergence but also an enhanced precision in multi-object tracking. In comparison to the standard Cuckoo Search Optimized Particle Filter algorithm, it showcases a 28.5% increase in global convergence iteration speed. Furthermore, when juxtaposed against the particle filter joint probability data association and particle swarm optimization particle filter joint probability data association algorithms, it shows respective accuracy enhancements of 24.7% and 11.81% in estimation precision.

Abstract:As the traditional ORB feature point extraction and matching method is not rich in image texture information or when the lighting changes drastically, it is very easy to produce feature point loss, uneven distribution and other problems, which is not conducive to the location and construction of the SLAM system. In this paper, a set of more robust and higher accuracy extraction matching algorithm is proposed. Firstly, the extraction algorithm is improved based on the ORB feature points, the adaptive threshold is calculated and the feature points are extracted based on the grid model, which can improve the robustness of feature point extraction and make its distribution uniform. In addition, the G-R image matching algorithm is also proposed, which calculates the neighborhood support estimator based on grid features to distinguish between positive and incorrect matches, and then combines with the RANSAC algorithm that introduces the evaluation function to further eliminate incorrect matches, which improves the matching accuracy by 9.36% compared with the original matching algorithm of ORB-SLAM2, and reduces the time consumption by about 13.6%. Finally, the feature point extraction matching algorithm proposed in this paper is added to the ORB-SLAM2 algorithm framework, which is verified by the dataset and the actual scene that the method in this paper can effectively improve the positioning accuracy of the ORB-SLAM2 system by more than 36.6% and make the system more robust.

Abstract:Rich cardiovascular information is encompassed within the arterial blood pressure (ABP) waveform, offering valuable insights for the prevention and diagnosis of cardiovascular diseases. Despite the availability of several photoplethysmography (PPG)-based blood pressure prediction methods, they primarily focus on predicting systolic blood pressure (SBP) and diastolic blood pressure (DBP). This paper proposes a novel method for blood pressure measurement that predicts the entire ABP waveform from PPG signals. The proposed approach involves linearly mapping the PPG signal to a high-dimensional space and feature extraction using a Transformer encoder structure. A linear layer is then utilized to output the predicted ABP waveform, enabling the calculation of SBP and DBP. Experimental results demonstrate that the Transformer network provides an accurate fit to the actual ABP waveform in the MIMIC dataset, with predicted SBP and DBP errors averaging (3.76±5.66) mmHg and (2.20±3.77) mmHg, respectively. Additionally, the proposed method complies with the standards of the Association for the Advancement of Medical Instrumentation (AAMI) and achieves Grade A according to the British Hypertension Society (BHS) criteria.

Abstract:The intelligent defect detection algorithm based on electromagnetic ultrasonic technology can be used to monitor the quality status of important parts and ensure the safe and reliable operation of equipment. In the actual detection process, on the one hand, the collected signals are often polluted by noise, which interferes with the detection results. On the other hand, the defect signals of important parts often have less data and cannot meet the needs of neural network training. Therefore, this paper proposes a noise reduction algorithm based on variational mode decomposition to pre-process detected signals to improve signal quality, proposes an improved virtual sample generation technology to expand the sample set, and uses transfer learning technology to reduce the number of parameters in neural network training to solve the problem of insufficient sample number. The average prediction accuracy of this method is 97.2% in the depth detection example of aluminum plate surface defects. Therefore, this method has certain reference significance for the surface defect detection of non-ferromagnetic materials.

Abstract:In response to low efficiency and high leakage rate caused by different sizes of transmission tower components and more defects in the current defect detection of power line towers, this paper proposes a Dynamic Position Query-Guided Multi-Scale Instance Segmentation method and a Graph Feature Memory-based Defect Detection method. The proposed instance segmentation method extracts multi-scale aerial image features, selects low-resolution pixels with the highest attention scores from the features, maps them to the corresponding positions in high-resolution features, and incorporates a bounding box detector to enhance the segmentation accuracy of power transmission towers. In the defect detection algorithm, a learnable graph feature descriptor is introduced, a memory bank is constructed to extract key elements for more accurate sample feature extraction, thereby improving defect detection efficiency. The power transmission tower defect detection method presented in this paper is compared with other state-of-the-art algorithms on two self-constructed defect detection datasets, the box_APand mask_AP of instance segmentation saw significant improvements of 7.6% and 0.5%, respectively, compared to Mask2Former. The AUROC indicator of defect algorithm was 7.3% and 1.6% higher than the second-best algorithm for the two datasets, and the F1-Score was improved by 6.7% and 6.9%, respectively. These results strongly demonstrate the outstanding performance of our algorithm in the transmission tower defect detection.

Abstract:Accurate segmentation of diabetes retinopathy is the prerequisite and key step to achieve automatic diagnosis of retinopathy. However, most of the existing segmentation models have limitations such as large parameters, unsatisfactory model training effect, and even inability to process data sets normally. To this end, an improved Ghost convolution module and multi-scale feature fusion module are added to the original U-Net network, and an improved U-Net algorithm for fundus lesion segmentation images is proposed. This model can achieve good segmentation results with a small number of parameters and low computational complexity. Using the Ghost Model to replace the original convolution, design Ghost convolution and Ghost down sampling convolution modules to ensure accuracy while reducing the number of parameters; Design a lightweight Half U-Net multi-scale feature fusion module to obtain multi-scale information, and introduce CBAM attention mechanism to improve its adaptability for different scale lesion targets, thereby better extracting small lesion information. The improved model is implemented in the mIoU on the two publicly available datasets, e_optha and IDRiD, were 61.42% and 61.84% respectively, while the F1 Score was 70.59% and 69.41%, respectively. The model parameters and FLOPs are only 5.48 M and 35.46 GMac, respectively, which are more streamlined and have higher segmentation accuracy compared to U-Net, Att-UNet and other models.

Abstract:Aiming at the problem of low manual measurement of the size of continuous casting rolls in the sector section in the continuous casting workshop, a continuous casting roll size measurement method based on binocular vision system is proposed. Firstly, the images collected by the binocular camera are preprocessed and the Otsu method was used to segment the background before and after the workpiece; And then, in view of the problem of low edge detection accuracy, the gradient templates of the traditional Canny edge detection algorithm are increased to 8 to extract the workpiece contour, and the sub-pixel level feature points are extracted by combining the polynomial interpolation formula. Then, based on the stereo matching of SAD and Census transformation fusion, the RANSAC algorithm was introduced to eliminate false matching. Finally, the triangulation principle is used to calculate the dimensions of the part. The experimental results show that the average relative error of the system measurement is 0.14%, and the measurement method has high accuracy, and its stability and accuracy meet the task of automatic dimensional detection of continuous casting rolls.

Abstract:In order to solve the problems of low detection accuracy and weak generalization ability of steel cord surface defects, a steel cord defect detection method based on DCGAN and improved YOLOv5s was proposed. Firstly, by adjusting DCGAN network parameters and optimizing hyperparameters, the generator can generate steel cord defect images with rich features and clear texture, thus expanding the data set. Secondly, the K-Means++ algorithm is used to re-cluster the anchor frame to obtain better anchor frame parameters and achieve accurate matching between anchor frame and actual defects. Then, coordinate attention mechanism was added to C3 module of YOLOv5s backbone network to enhance the feature extraction capability and accurate localization capability of the model. Finally, MPDIoU loss function is introduced to replace YOLOv5s original loss function to further improve the detection accuracy. The experimental results show that on the measured steel cord defect data set, the average accuracy of defect detection is increased by 6.6%, reaching 89.4% by using the YOLOv5s detection model enhanced and improved by DCGAN data, and the detection accuracy and recall rate are also improved. Compared with other mainstream detection models, this model not only improves the detection speed by about 30%, but also maintains high detection accuracy. On the publicly available NEU-DET dataset, the mAP value of this model reaches 82.6%, which is 3.8% higher than that of the original YOLOv5s model.

Abstract:At present, there is a rapid growth in communication services in mobile communication systems. To alleviate the power consumption caused by the increasing base station load, renewable energy production equipment has been equipped for mobile communication system base stations. By matching the information flow and energy flow in the communication system, communication services and renewable energy storage in the communication system can be accurately paired. This can further improve the utilization rate of renewable energy within the communication system, which is the key to optimizing the network performance of the communication system and reducing system operating costs in the next step of research. Therefore, this paper constructs a multi-dimensional utility function. This function comprehensively considers three factors: user signal interference noise ratio, renewable energy utilization, and base station load. This paper solves the initial problem of minimizing the operating cost of a multi base station system by transforming it into a problem of maximizing the utility value of a multi-dimensional utility function. The transformed problem is a non-convex problem of mixed integer nonlinear optimization. To solve this problem, this paper proposes the Multidimensional Utility Function Iterative Optimization Algorithm. This algorithm divides the problem into three subproblems: user scheduling, power allocation, and load balancing. Then, this problem can be iteratively solved by using alternating optimization and continuous convex approximation techniques. The simulation results show that compared to the Maximum SINR Association Optimization Algorithm and the "Maximum SINR and Renewable Energy Utilization" Optimization Algorithm, the algorithm in this paper has improved the utilization efficiency of renewable energy by 58.68% and 29.74%, respectively. At the same time, the total cost of applying the algorithm proposed in this paper has been consistently lower than other algorithms during the simulation period. This indicates the advantages of the algorithm proposed in this paper.

Abstract:In response to the current surge in network traffic leading to a sudden increase in network security incidents and an added burden on network management, a network architecture based on deep learning techniques has been proposed. This architecture involves the parallel use of ResNet and one-dimensional Vision Transformer for the identification and classification of network traffic. ResNet is capable of extracting deep spatial features from flow data, ensuring high accuracy in traffic recognition. Meanwhile, the one-dimensional Vision Transformer excels at capturing more representative temporal features. By employing an attention mechanism to adaptively merge these two types of features, a more comprehensive feature representation is obtained to enhance the network′s capability in traffic identification. Experiments conducted on the ISCX VPN-nonVPN dataset demonstrate that the proposed method achieves an accuracy of 99.5% in application-based traffic classification experiments. Compared to standalone ResNet and one-dimensional Vision Transformer, as well as classical one-dimensional Convolutional Neural Networks (1D-CNN) and CNN combined with Long Short-Term Memory (CNN+LSTM), the proposed method shows improvements of 0.9%, 3.6%, 6.6%, and 3.3%, respectively. On the USTC-TFC 2016 dataset, the proposed method not only easily identifies malicious traffic but also accomplishes the classification of 13 different applications, with an average classification accuracy of 98.92%. This proves its ability to recognize malicious traffic and perform fine-grained classification tasks.

Abstract:In order to accurately describe the wireless channel between the inspection UAV and GBS in the power transmission environment, we propose a three-dimensional channel model based on GBSM. The influence of tower poles and transmission lines in the transmission line environment is considered. The horizontal cylinders are used to describe the scatterer distribution of transmission equipment and the surrounding environment. In order to prevent the electromagnetic field generated by the transmission line from affecting the safety of UAV inspection operations, a safe flight area is set up in the model to ensure the safety of inspections. For the proposed channel model, channel statistical characteristics such as space-time correlation function，Doppler power spectral density，envelope level crossing rate and average fade duration are derived and analyzed. The influence of scatterer distribution and UAV motion status on channel statistical characteristics was studied. The simulation results show that the UAV′s speed, flight direction, and scatterer distribution have a significant impact on the channel. The theoretical results and simulation results are in good agreement, verifying the correctness and effectiveness of the proposed model, and can provide a theoretical reference for the design of wireless communication systems between inspection UAV and GBS in transmission line scenarios.

Abstract:UAV flight data is an important state parameter reflecting its own flight safety, and it is a key initiative to improve the overall flight safety of UAVs through abnormal detection of flight data. Although data-driven methods do not require expert a priori knowledge and accurate physical models, the lack of parameter selection and a single model for the detection network structure make the detection model overfitting due to too many parameters and failing to effectively capture data anomaly patterns. In this paper, a VAE-LSTM based UAV flight data anomaly detection modeling method is proposed by combining the advantages of Variational Auto-Encoders and Long Short-Term Memory networks. First, the Kendall correlation analysis method is introduced for selecting relevant dependent flight data parameter sets; Second, the parameter sets with correlation are trained on the designed VAE-LSTM deep hybrid model to learn the relational mapping between different data features; And lastly, the validation is performed with unsupervised anomaly detection in real multi-dimensional Unmanned Aerial Vehicle flight data. The experimental results show that the various average performance metrics of precision, detection rate, accuracy, F1 score and false detection rate of VAE-LSTM reach 95.24%, 98.71%, 98.8%, 96.82%, and 1.31%, respectively, and show overall better anomaly detection performance compared to KNN, OC-SVM, VAE, and LSTM models.