Abstract:In response to the issues of high cost, inadequate accuracy, and limited flexibility of weak magnetic field measurement devices available on the market, a three-dimensional magnetic field measurement device has been developed using PNI magnetic field measurement technology, centered around the RM3100 sensor. This device is capable of measuring magnetic field strengths at the nT level. Utilizing an ESP32 as the main control chip, it integrates modern communication technologies and cloud services to control servo modules, Bluetooth modules, Wi-Fi modules, and display modules, enabling real-time synchronous display of measurement data on both the device and PC, as well as precise close-range control of a three-dimensional mobile platform via a mobile phone. This enhances the device′s flexibility and potential application scenarios. Experimental results indicate that the device measures the Earth′s magnetic field strength to be 28.93 μT, with a measurement range covering ±800 μT and an error margin within 5%, while achieving a minimum measurement precision of 10 nT. The device is characterized by low cost, high precision, great flexibility, and ease of operation, making it suitable for various fields such as physics classroom demonstrations, experimental teaching, and non-destructive testing, with promising application prospects.

Abstract:To address the frequent severe rainfall disasters along the Beijing-Zhangjiakou Railway, this study employs a combined weighting analysis method integrating the Analytic Hierarchy Process and Random Forest algorithm for risk assessment and zoning to facilitate disaster warning. Utilizing data from rainfall sensor networks along the railway line, a risk model was established by assigning weights to three categories of indicators: hazard factor risk, sensitivity of disaster-pregnant environments, and vulnerability of disaster-bearing bodies. Risk levels were calculated by integrating rainfall and geological data, with visualization achieved through ArcGIS software. Experimental results identified the highest-risk areas in the Badaling-Nankou section and Qinglongqiao segment, consistent with historical disaster records. Compared with single-method AHP approaches, this hybrid method demonstrates enhanced accuracy, providing valuable references for optimizing disaster warning systems and improving emergency response mechanisms.

Abstract:Aiming at the common problems of incomplete algorithms and complex implementation in structured light 3D measurement systems based on embedded platforms, a monocular surface structured light 3D measurement system based on ZYNQ was designed. This system combines FPGA and ARM processors, utilizing Vitis HLS technology to achieve high-speed decoding and reconstruction based on complementary Gray code combined with phase-shifting algorithms. First, an FPGA image acquisition module was designed, and the structured light images were preprocessed through grayscale conversion and median filtering. Subsequently, Vitis HLS was used to implement structured light image binarization, phase calculation, and 3D point cloud solving. Finally, a control program for the CMOS camera and image processing was developed based on the ARM processor, enabling system module control and flexible system parameter calibration, with a reprojection error as low as 0.092 1 pixels. Experimental results show that the system exhibits high robustness for complex object surfaces, with a root mean square error of diameter fitting for standard sphere measurements as low as 0.103 7 mm. The decoding and reconstruction time is only 112.67 ms, approaching 31.5 times the speed of a CPU and comparable to an integrated GPU, while the power consumption of the ZYNQ platform is only 2.696 W, verifying the feasibility and effectiveness of the system.

Abstract:To address the issues of low efficiency and high error rates in green tires logistics management for small and medium-sized tire manufacturing enterprises, this study designed an UWB-based FIFO system for green tires storage and retrieval. The system integrates a UWB positioning network with multi-base station redundancy collaboration, centroid correction algorithms, and FIFO prioritization to dynamically determine the retrieval sequence of green tires. A dual-mode display panel provides real-time visual guidance for operators. Comparative experiments demonstrated that after system implementation, the average retrieval time decreased significantly from 3.017 minutes to 1.009 minutes (a 66.6% reduction), while the error rate dropped from 7.38% to 0.23%, markedly enhancing retrieval accuracy. Operational stability across different time periods was also improved. The results indicate that the system effectively resolves challenges such as green tires expiration, time-consuming retrieval, and operational errors inherent in traditional manual workflows, offering a viable solution for intelligent warehouse management in small and medium-sized tire enterprises. This research provides a practical framework for optimizing logistics efficiency and precision in the manufacturing sector.

Abstract:In response to the problem of narrow gain range of traditional LLC resonant converter, this paper proposes a novel LLC resonant converter topology with wide gain range. In this circuit, the output of a four-switch Buck-Boost converter is cascaded with the front bridge arm of the LLC resonant converter. The full-bridge LLC resonant module switching tubes operate near the resonant frequency point, and the input voltage AC amplitude of the LLC resonant module is varied by changing the duty cycle of the four-switch Buck-Boost module switching tubes. The four-switch Buck-Boost module only needs to carry part of the load power to realize the adjustment of the wide voltage gain range of the whole machine. Through the modal analysis and gain derivation of this new converter, the conditions for all switching tubes to realize soft switching and the control method of wide gain are derived. Finally, a 30 V/300 W experimental prototype is constructed, and the peak efficiency of the converter is 95.27%, which is 2.13% higher than that of the traditional two-stage topology. The experimental results prove the feasibility of the proposed topology circuit and the correctness of the theoretical analysis.

Abstract:With the rapid advancement of modern military technology, radar systems are widely used on battlefields. However, due to constraints like airspace and terrain, testing their performance and reliability in real signal environments is challenging. To address this, a UAV-mounted multi-mode broadband radar waveform simulation electronic target based on the ZYNQ multicore processor and ADRV9009 RF transceiver chip is proposed. The system employs a modular design for radar algorithms, generating waveforms from inter-pulse characteristics such as agility and PRI, and intra-pulse characteristics like FMCW, supporting flexible radar scalability. Tests show peak-to-peak and bandwidth errors of 0.94% and 0.121% in agility testing, RMS errors of 0.51, 0.42, and 0.47 for jitter, stagger, and sliding PRI measurements, 0.54% FMCW range error, and RMS symbol width errors of 0.005 3, 0.004 8, and 0.003 8 for 2, 4, and 8-phase coding. Compared to traditional simulators, this design offers high integration, low power, compact size, rich waveform concurrency, and wide frequency adjustment, enhancing radar testing and countermeasure training.

Abstract:To solve the problems of traditional path planning algorithms, such as low efficiency in path planning, unsmooth paths, and poor dynamic obstacle avoidance, this paper proposes a path planning method that combines the A* algorithm with the APF algorithm. For the A* algorithm, a dynamic weighting method is used to adjust the weight of the heuristic function according to the position of the traveling robotic fish and the distance between the robotic fish and the obstacles as an index; then the Gaussian filtering method is used to curve-smooth the obtained optimal path. Then the path generated by the improved A* algorithm is used as the search path of the APF algorithm, and dynamic obstacle avoidance is carried out on the basis of realizing the shortest path planning. The results of simulation experiments show that the improved A* algorithm obstacle different maps, the average reduction of the search time is 52.32%, the average reduction of the number of search nodes is 56.60%, the average reduction of the path length is 6.33%; under different sizes of maps, the average reduction of the number of search nodes is 49.60%, the average reduction of the search time is 40.89%, the average reduction of the path length is 5.55%. The fusion algorithm can perform successful dynamic obstacle avoidance and path planning under maps with dynamic obstacles.

Abstract:Acquiring surface orthophoto image by UAV technology can quickly and effectively realize the comprehensive monitoring and analysis of mining surface morphology. However, shadows are prevalent in the open-pit mine UAV orthophoto image, which not only interferes with the acquisition of some ground object information, but also reduces the interpretation and recognition accuracy of UAV images. There are few researches on shadow extraction in mining area at present. Existing methods can not meet the needs of shadow identification in open pit mines. And the problems of not establishing shadow data set in opencast mining area, a UAV ortho image shadow dataset is constructed by using manual annotation for the first time. Based on the UNet3+model, a shadow extraction method combining mixed attention mechanism (CBAM) and depth separable convolution layer (DSC) is proposed. By introducing the ResNet feature extractor, feature extraction on five scales is carried out on the original image, and performs full-scale jump connection according to the extracted features to carry out feature fusion. And introduce CBAM attention mechanisms to enhance useful features. The category of each pixel is predicted by the feature map recovered by the deep monitoring mechanism and the decoder. Finally, proposed method is compared with four typical target extraction networks FCN, UNet, UNet++and UNet3. The experimental results show that, compared with Unet3+network, the mPrecision, mRecall, mF1 and mIoU improved by 4.9%, 0.44%, 2.24% and 4.51%, respectively. Proposed method was compared with a variety of existing shadow extraction methods on AISD public data. The experimental results showed that compared with residual supervision network, F1 and IoU improved by 0.27% and 2.62%. It is proved that this method is accurate in shadow extraction, and is suitable for shadow extraction in open pit mining area.

Abstract:Stem and leaf angle detection is an important part of tobacco phenotype detection, which is of great significance in increasing yield and efficiency and disease prevention in tobacco farming. Aiming at the problems of low efficiency, long cycle time and inconvenience of manual stem and leaf angle detection in different environments, a lightweight tobacco stem and leaf angle detection model, FAL-YOLO, was designed and constructed.The algorithm builds the FAI backbone network structure to sufficiently reduce the amount of computation and feature redundancy, and increase the efficiency of using semantic information. The SAC detection head module, which integrates the spatial attention and channel attention SA attention modules, is designed to further reduce the number of parameters and improve the perception of stem and leaf angle features. GSConv lightweight convolution is introduced to reduce model complexity and the number of model parameters. The MPD-IoU loss function is introduced to improve the overall performance of the model. A self-constructed tobacco stem and leaf angle detection dataset is used to carry out the comparison and ablation experiments of the FAL-YOLO model. The experimental results show that the mAP of the FAL-YOLO model on the self-constructed dataset reaches 99.2%, compared with the YOLOV8-POSE model in the GFLOPs, the Params are reduced by 56.7% and 52%, respectively, and the improved model is capable of identifying the stem and leaf angles of tobacco plants faster and more accurately, which can support the wisdom of tobacco agricultural seed selection and breeding.

Abstract:To address the issues of low optimization accuracy and the tendency to fall into local optima in the northern goshawk algorithm, an improved version is proposed that integrates the subtraction optimizer and t-distribution wavelet mutation. In the initial phase of the algorithm, the Tent map combined with the dynamic reverse learning strategy is utilized to improve the quality and diversity of the initial population, thereby accelerating the iteration speed of the algorithm.Secondly,in the exploration stage, the subtractive average optimizer and the best value guidance strategy are introduced to update the population position. Finally, an adaptive t-distribution wavelet mutation strategy is employed to perturb the population, preventing it from falling into local optima.Through simulation experiments using test functions and integrating the improved algorithm with the extreme learning machine, the approach was applied to predict photovoltaic power generation. Additionally, it was implemented in two engineering design applications. The experimental results demonstrate that the improved algorithm significantly outperforms other modified algorithms in terms of convergence accuracy and robustness, and effectively enhances the performance in solving complex problems.

Abstract:In view of the problem that the current PCB defect detection algorithm cannot simultaneously take into account the number of model parameters and detection accuracy, this paper proposes an improved lightweight PCB detection algorithm ST-YOLO based on YOLOv8n. First, the backbone network of YOLOv8n was replaced by the lightweight backbone network StarNet to adjust the network structure. Delete the large target detection layer and add the small target detection layer. Secondly, C2f module is combined with Star Block and CA attention mechanism to design C2f-Star-CA module, which can better integrate local and global context information. Finally, lightweight detection headers are designed to reduce the number of parameters in the model by using shared convolution. The experimental results show that compared with YOLOv8n, the number of model parameters is reduced by 45.5%, the calculation amount is reduced by 56.8%, and the mAP%0.5 is increased by 0.2%. It provides new possibilities to meet the needs of mobile deployment.

Abstract:In order to solve the problems of multi-entity and multi-matching categories and complex context information in the field of discourse level relation analysis and recognition, this paper proposes an English discourse structure analysis and recognition method considering multifeatures by fusing entity feature information and context feature information. Firstly, a structural analysis and recognition framework considering multiple features is proposed. Then, the mechanisms of entity recognition unit and context relation recognition unit are introduced in detail. Finally, comparative experiments and ablation experiments are carried out through public datasets and self-selected datasets to verify and analyze the superiority of the proposed model, and the recognition efficiency of the model and the influence of related parameters are experimented and analyzed. On the two datasets, the proposed method can maintain the optimal performance in both indicators F1 andIgnF1. Compared with other suboptimal recognition models, the proposed method improves 1.65% and 1.13% respectively on F1, 2.78% and 1.58% respectively onIgnF1. Experiments show that the proposed model can extract the key feature information representing the discourse relation, help to understand the context and the relationship between each part of the text, and grasp the overall structure of the text.

Abstract:Underwater object detection often faces challenges such as complex environmental interference, unstable system performance, and low detection accuracy. To address these issues, this paper proposes WAD-YOLOv8, a lightweight object detection algorithm based on adaptive feature extraction and cross-scale feature fusion strategies. First, a context-aware residual feature extraction module (CCRF) is introduced in the backbone network, enabling the model to effectively integrate global and local information. Second, an adaptive down-sampling module (ADFE) guided by variable large kernel convolution attention mechanisms is employed to adjust sampling features dynamically, enhancing the network′s adaptability. Finally, the neck network is restructured by incorporating new cross-scale feature fusion connections, significantly improving the model′s robustness against environmental interference. Experimental results demonstrate that, compared to the baseline model, WAD-YOLOv8 achieves a 3.0% improvement in detection accuracy and a 2.6% increase in mAP50, while reducing model parameters and computation by substantial margins. The detection speed reaches 64 FPS, outperforming classical algorithms in both effectiveness and stability. These improvements highlight the model′s capability to address the challenges of underwater object detection, offering a highly efficient and reliable solution for complex underwater environments.

Abstract:To improve wind power prediction accuracy, a combination model based on SSA-optimized Transformer-BiGRU is proposed. First, CEEMDAN decomposes the original sequence into multiple modal components and a residual component, reducing data complexity and instability. Then, a high-efficiency combined model is constructed by integrating the self-attention mechanism of the Transformer with the bidirectional time-series modeling capability of BiGRU. To address the challenge of hyperparameter optimization for the Transformer-BiGRU model, the SSA algorithm is introduced to optimize the hyperparameters, further enhancing prediction accuracy. Finally, using the Longyuan Electric Power wind power prediction dataset, comparative and ablation experiments are conducted to show that the proposed model outperforms other traditional models and demonstrates the effectiveness of each component. The experimental results indicate that the method achieves an R2 of 0.981 0.

Abstract:To address the issues of high parameter count and computational complexity in existing traffic surveillance detection models, which limit their deployment on edge devices due to hardware resource constraints, this study proposes a lightweight network architecture-based road surveillance detection model by specifically modifying the YOLOv8 model. In the backbone network, the minimalist architecture VanillaNET is introduced to replace the intermediate part of the original network for feature extraction, significantly reducing the model′s parameter count and overall computational complexity. The advantages of FasterNet are combined with the EMA attention mechanism and applied to the C2f module in the backbone network, effectively reducing memory access and enhancing the model′s detection capability. Additionally, the G-SPPCSPC module is proposed by integrating SPPCSPC with grouped convolutions, improving the model′s ability to extract multi-scale information under varying surveillance perspectives. Finally, in the neck network, the lightweight attention mechanism MLCA is incorporated into the C2f module to reduce interference from irrelevant background information in road surveillance detection. Experimental results show that the improved model reduces the parameter count by 53.3%, model size by 51.3%, and computational complexity by 48.1%, while achieving a mAP50/% of 93.7% and an FPS of 280.5. The model maintains high detection accuracy and speed while significantly reducing parameter count and computational complexity, making it suitable for deployment on edge devices and demonstrating high practical value.

Abstract:The control system, as a safety-critical system in nuclear energy equipment, often needs to operate without human intervention for a long time in actual engineering. It demands extremely high levels of automation and operational reliability. The highly robust signal voting algorithm can ensure that the control system can automatically respond and return to normal operation when it encounters a fault or abnormal situation. At present, the signal voting algorithm commonly used in nuclear energy control system is threshold detection method. This kind of method is simple in structure and easy to understand, but its control accuracy, reliability and automation are not good. Therefore, based on the design of the nuclear energy control system under the condition of long period unmanned operation, this paper proposes a multi-level voting algorithm for the redundant signal of the nuclear energy control system based on voting mechanism to meet the high reliability and automation requirements of the system. The first-level monitoring algorithm determines the signal fault point, and on this basis, the second-level monitoring voting algorithm outputs the final voting value. In this paper, the signal sequence in the simulation process is intercepted, and the functions of the proposed algorithm such as output voting value, fault count, fault point elimination and recovery and output safety value are verified. At the same time, a variety of test signals are simulated for comparative verification. The results show that the proposed algorithm can effectively reduce the error probability of voting on step signals, ramp signals and sine signals. Its voting result is generally superior to that of traditional voting algorithms such as average and median, and the threshold operation result is better than that of traditional threshold detection methods. Finally, by discussing the time complexity of the algorithm and the average running time of the algorithm, the method is verified to meet the realtime requirements of the nuclear energy control system.

Abstract:In response to the issue of excessive reliance on professional equipment for detecting air PM2.5 concentration, an air quality detection algorithm based on multi-source information and image feature generalization is proposed. Firstly, the EfficientNet-B0 was used as the backbone network for feature encoding of the input atmospheric visible image, the multi-source meteorological information, such as temperature, humidity, wind speed, pressure and light intensity, was mapped into feature vectors corresponding to the atmospheric image, and fused with the atmospheric image features. Then, the global features were output as scalars using a fully connected layer, and the PM2.5 concentration in the air was detected using a loss function. Finally, the features of atmospheric images at different scales were randomly generalized and enhanced in the training phase of the network model to enrich the sample distribution space, making the network to learn more features from limited data samples, thereby effectively improving the performance of the detection network. The experimental results show that the designed air quality detection method has higher detection accuracy and stability compared to several mainstream methods, the RMSE and R-squared obtained on the test set are 21.55 μg/m3 and 0.923, respectively. The average error obtained by detecting 8 scenarios is only 5.2%, and the maximum error is only 7.6%, which can adapt to air quality testing tasks in various extreme atmospheric pollution environments.

Abstract:A deep feature interaction network is proposed to address the problems of insufficient utilization of complementary information between multimodalities and the tendency of feature interaction to introduce noise in existing methods. First, a deep feature multilayer interaction module is proposed in the coding stage, which uses depth features as cues for feature interaction to fully utilize the texture information of visible light and the position information of thermal imaging. Second, a texture-position feature interaction module is designed to interact texture information with position information to achieve feature complementarity between the same layers. Then, the inflated convolutional feature fusion module is proposed in the decoding stage, which improves the model sensory field by inflating the convolutional block so that the model focuses on the multi-scale information in the network. Finally, extensive experiments are conducted on the public RGB-T datasets VT5000, VT1000 and VT821, which show that the average absolute errors of the proposed networks reach 2.2%, 1.5% and 2.5%, respectively, and achieve excellent performance compared with the stateof-the-art methods in the field.

Abstract:In the field of computer vision, monocular depth estimation has garnered significant attention due to its importance in applications such as autonomous driving and scene reconstruction. However, existing self-supervised monocular depth estimation methods fail to fully exploit low-level features, resulting in poor depth estimation performance for object contours. To address this issue, this paper proposed a multi-scale feature fusion decoding method. The original RGB image is progressively downsampled using a Gaussian approach to obtain feature maps at various levels, which are then upsampled using Gaussian processes. During upsampling and downsampling, Laplacian pyramids are constructed using feature maps of the same dimensions. During decoding, the lost contour cues from downsampling are fused with the features extracted by the encoder at each scale, guiding the decoder to generate more accurate depth maps and maximizing the utilization of low-level features from the encoder. Compared with the experimental results of the baseline method Monodepth2 on the KITTI dataset, this method reduced the absolute relative error Abs Rel by 1.69%, the squared relative error Sq Rel by 6.80%, and the root mean square error RMSE by 1.00%, indicating that this method has improved the accuracy of global depth estimation, and the visual analysis also verified that the method has significantly improved in the depth estimation effect of object contours.

Abstract:In the process of rail surface defect detection, due to the influence of external factors such as uneven illumination and lens shake, the collected images have problems such as low contrast, uneven background and blurred defect details. Therefore, an image enhancement algorithm for rail surface defects based on improved Retinex and dual CNNs was proposed. Firstly, after converting the RGB image of rail surface defects into HSV space, the Retinex algorithm that introduces the mean and mean square deviation and controls the dynamic parameters of the image is added to adjust the contrast of the V component, and then the image exposure is corrected by adaptive gamma transform. Secondly, the S component is enhanced according to the brightness to solve the problem of uneven background caused by lighting changes. Thirdly, in order to further solve the problem of blurring the details of defective images caused by lens shake, a dual CNNs network composed of a deblurring sub-network and a super-resolution detail recovery sub-network based on U-Net structure was designed to learn the semantic features of the original image and the enhanced image, and extract their texture features to obtain the texture and detail information of high-quality images. Finally, the RSDDs dataset and the self-made rail surface defect fuzzy image dataset were used to train and test the model. Experimental results show that compared with the existing mainstream algorithms, peak signal-to-noise ratio and structural similarity are increased by 2.61 dB and 0.026, respectively, and visually have higher contrast, clear defect details and rich texture information than the rail surface defect images obtained by the other 10 methods.