Abstract:Reconfigurable wideband components play an important role in wireless communication systems, however, designing a straightforward wideband 1-bit unit remains a significant challenge. This design is based on the principle of polarization conversion. By controlling the on-off state of the PIN diodes, the polarization of the incident wave is rotated by ±90°, provide dual-polarization regulation and a 180° phase shift. The unit operates from 25.58 to 31.87 GHz, with a relative bandwidth of 21.9% and the cross-polarization reflection coefficient is greater than -2 dB. Meanwhile, the unit demonstrates excellent angular stability, supporting an oblique incidence angle of up to 30°. Utilize this unit to construct a phase-reconfigurable reflectarray with an aperture size of 70 mm × 70 mm. The gain is 20.6 dB at 30 GHz, which can realize ±60° full-space beam scanning capability. This wideband tunable design can be utilized in wireless communication, radar detection, and other fields.

Abstract:During the wind tunnel testing, the balance will be subjected to aerodynamic dynamic loads for a long time, which may lead to fatigue failure of the balance. In severe cases, cracks or fractures may occur, which can not only cause damage to the balance, but also catastrophic situations such as test pieces falling off and being blown away along the surface of the wind tunnel. These unexpected situations often delay the entire development cycle of the aircraft. To reduce the risk of fatigue failure of the balance and ensure the safety of wind tunnel testing, the FL-9 wind tunnel pressurization test balance was taken as the research object. Combined with the finite element analysis results of the balance, high-risk stress nodes of the balance structure were extracted, and fatigue monitoring was carried out throughout the calibration and wind tunnel testing process of the balance. The test results indicate that the fatigue stress monitoring technology of the balance can ensure the safety of the balance. Based on this, the load range of each unit of the balance has been expanded by 1.5 to 2 times. The wind tunnel test results show that the balance still has sufficient safety margin to obtain a wider range of test data.

Abstract:A method based on the DR-IFMM is proposed for inpainting damaged rock art images. This method determines two optimal repair radii based on the pixel density of the damaged region, and is applied to the IFMM algorithm to generate the repaired image respectively. The IFMM algorithm improves the weight calculation rules based on the FMM algorithm, and then fuses and restructures the two images into the optimal repaired image. The experimental results show that the DR-IFMM method outperforms the MSMM, IK-means, COTR, STDecomposition, SFIIM, AutoFill and ICriminisi methods in inpainting rock art images with various types of damages, and effectively addresses the issues such as color loss and texture clutter. Compared with the LaMa method, the advantage of the proposed approach is that it still can achieve better inpainting results without model training and high-performance computer. Inpainting damaged rock art images can inherit and develop the rock art through the form of digital. and provide cultural relic researchers with a complete record of China′s ′history′ etched on stone walls.

Abstract:Cloud phase is not only an important parameter in meteorological and climatological research but also a key element in cloud parameter inversion. Accurate identification of cloud phase is crucial for weather monitoring and forecasting. Traditional cloud phase recognition methods often rely on threshold setting, which is highly subjective and not very reliable. Therefore, this paper proposes a semi-supervised adaptive possibility C-means algorithm that enhances the processing capability of multi-dimensional data and the robustness of classification through semi-supervised learning combined with an adaptive feature weighting mechanism and regularization techniques. By applying this method to Raman lidar and millimeter-wave cloud radar data, it is possible to accurately classify ice clouds, water-dominated mixed clouds, ice-dominated mixed clouds, and supercooled water clouds. Compared with the algorithm before improvement, the classification accuracy has been significantly increased from 0.699 to 0.967, greatly improving the accuracy of cloud phase classification.

Abstract:To address the issue of traditional algorithms being prone to local optima during the maximum power point tracking (MPPT) process due to the multipeak characteristic of photovoltaic array output power curves under partial shading conditions, this paper proposes an MPPT control strategy combining the improved Black-winged kite algorithm (TBKA) and the Perturb and observe method (P&O), referred to as TBKA-P&O. In the global search phase, the population is first initialized using the Tent-Logistic-Cosine chaotic mapping. Then, a tangent flight strategy is introduced to enhance the search efficiency and convergence accuracy of the TBKA. Additionally, a dynamic lens imaging reverse learning strategy based on a greedy approach is designed to improve search diversity and prevent local optima. In the local search phase, the P&O method is incorporated to achieve rapid localization and high-precision tracking of the maximum power point. To verify the effectiveness of the proposed algorithm, a photovoltaic power generation system simulation model was constructed, incorporating the traditional P&O algorithm, the BKA-P&O algorithm, the quantum CS-P&O algorithm, and the TBKA-P&O algorithm. Experimental results demonstrate that the TBKA-P&O algorithm achieved tracking accuracies of 100%, 99.97%, 99.96% and 99.96% under four operating conditions, with corresponding tracking times of 0.093, 0.090, 0.077 and 0.047 s. Compared to other algorithms, the TBKA-P&O algorithm exhibited significant advantages in terms of dynamic tracking speed, steady-state tracking accuracy, and power oscillation control.

Abstract:In order to solve the problem of difficulty in target detection caused by noise interference, illumination fluctuation and complex background in UAV aerial infrared images, an infrared target detection model for UAV based on YOLOv8 was proposed. Firstly, the SCDown module in YOLOv10 was introduced to maximize the preservation of contextual semantic information for each scale. Secondly, the dynamic upsampler DySample was introduced to improve the sensitivity of the model to image details. At the same time, the triplet attention mechanism is introduced to improve C2f to strengthen the model′s understanding of the relationship between spatial and channel dimensions and the processing ability of complex data. Finally, a lightweight decoupling head Efficient_Head module is designed to ensure the detection accuracy and greatly reduce the model parameters. Experimental results show that the improved algorithm mAP50 reaches 83.7%, which is 4.2% higher than YOLOv8n, the accuracy is increased by 1.2%, the recall rate is increased by 3.8%, the number of floating point operations isreduced by 2.5%, and the FPS reaches the detection speed of 323.17 fps, which fully shows that the overall performance of the improved algorithm is better than that of other mainstream algorithms, and it can better complete the task of UAV infrared target detection.

Abstract:To address the issues of low convergence accuracy and susceptibility to local optima in the PKO algorithm, this paper proposes a multi-strategy improved IPKO algorithm. First, Latin hypercube sampling is used to avoid overconcentration or neglect of potentially beneficial areas in high-dimensional problems, thus reducing the risk of local optima. Secondly, the positioning fishing mechanism from the OOA algorithm is introduced to enhance exploration of the optimal region and improve the ability to escape from local optima. Finally, a new falling mechanism is integrated to improve search stability and prevent premature convergence. An adaptive mutation rate termination condition is also applied to dynamically balance global exploration and local exploitation, optimizing solution quality and search efficiency. The training-testing accuracy and runtime under different feature dimensions are compared, and the impact of population size and iteration count on the algorithm′s performance is analyzed. Experimental results on 12 benchmark test functions show that IPKO outperforms other comparison algorithms in terms of convergence speed, solution accuracy, stability, and the Friedman test. When applied to the microgrid scheduling problem, IPKO demonstrates lower costs compared to other algorithms, with a reduction of 1.92% over the original PKO, confirming its effectiveness and reliability in practical applications.

Abstract:Timely detection of Self-exploding insulators during the inspection process can effectively prevent power grid accidents. In response to the problems of large memory and slow detection speed required for convolutional neural network training, which do not have advantages in real-time detection on mobile devices, self-explosion fault detection algorithm for Insulator based on improved YOLOv7-tiny is proposed. Firstly, deformable convolution and dynamic snake convolution is introduced into the YOLOv7 tinyIncorporating dynamic serpentine convolution into the YOLOv7-tiny algorithm and designing a more efficient layer network to enhance perception; then, the Gold-YOLO network is introduced to enhance the information fusion of the intermediate layer; subsequently, the MPDIoU loss function is used to reduce the redundancy of the prediction boundary; finally, designing a Self-exploding insulator detection system to enable staff to quickly identify self-exploding insulators in a massive collection of images. The research results show that the mean average precision of the improved algorithm is 96.3%, which is 1.1% higher than the original YOLOv7-tiny algorithm. The average precision of the improved algorithm is 99.5% for identifying Self-exploding insulators, which is 0.2% higher than the YOLOv7-tiny algorithm and 0.1% higher than the YOLOv7 algorithm. Moreover, the scale of the improved algorithm is only 28% of that of the YOLOv7 algorithm, and the FPS has increased by 11.3, reaching 60.6. The improved algorithm can meet the requirements of real-time detection while ensuring recognition accuracy.

Abstract:Accurately localizing and recognizing human moions in both spatial and temporal dimensions is of significant importance for applications such as intelligent sports analysis. However, existing step-by-step human motion recognition methods are often limited by the fixed receptive field of RoI features, making it difficult to achieve effective modeling and semantic representation in complex scenarios. To address this issue, this paper proposes a fine-grained motion & situation fusion (FMSF) network that integrates human representation features and global spatiotemporal situation features through parallel semantic modeling and motion proposal units. The semantic modeling unit employs a human localization model to generate finegrained human candidate features from key frames and leverages a 3D video backbone network to extract global spatiotemporal features. The motion proposal unit then uses a shared Transformer framework to jointly model these multi-modal features, capturing complex interactions between humans and their surroundings, resulting in highly discriminative motion predictions. Furthermore, a weighted score aggregation strategy is introduced to integrate the motion classification results of multiple key frames and short video segments for long-video motion recognition. On the AVA-60 v2.2 dataset, the FMSF model achieved a frame-level mAP of 30.01%, while the long-video strategy-based FMSF-Prolonged reached 30.74%. On the Charades dataset, the mAP of FMSF increased to 30.68%, and that of FMSF-Prolonged increased to 31.29%.

Abstract:Accurately detecting insulator defects is one of the main tasks of power grid maintenance. In response to the problems of low recognition accuracy of current insulator defect detection algorithms and large models that are difficult to deploy to mobile devices such as drones, a method based on YOLOv8 is proposed to improve the detection accuracy and lightweight the model. This method uses the feature fusion mode in a bi directional feature pyramid network BiFPN to fully fuse multi-scale features, and then integrates a deformable attention mechanism DAttention into the original algorithm to extract features with lower complexity. In addition, it introduces a fusion of average pooling and maximum pooling coordinate attention DAF-CA to enhance key information, and finally uses the minimum point distance based Intersection over Union MPDIoU as the loss function to improve the training effect of bounding box regression, thereby improving the accuracy of the algorithm. Multiple comparative experiments were conducted on the dataset, and the results showed that the proposed method achieved an average accuracy of about 91.0%. The model had a floating point count of 7.2 G and a parameter count of 2.07 M, respectively, and all performance indicators were superior to commonly used detection algorithms. This method can provide reference for intelligent inspection of power grids.

Abstract:In order to ensure the safe travel of the blind, an obstacle avoidance method combining object detection and binocular stereo vision was proposed to solve the problems of blind path occupation, damage and absence. Firstly, the sidewalk information is collected by binocular camera, and the obstacles on the sidewalk are detected by the improved YOLOv8s model. Then, an improved stereo matching algorithm is used to match the obstacles, which uses FAST algorithm with adaptive threshold to find the feature points on the scale space, and uses least square method to obtain the sub-pixel coordinates of the feature points and reduce the dimension of the feature descriptors. Finally, the two-dimensional pixel coordinates are converted into three-dimensional spatial coordinates by the ideal binocular model, and the depth value of the obstacle is obtained. Experiments show that the obstacle avoidance system can accurately identify the types of obstacles in the range of 10 meters, and the FPS can reach 149.1; when measuring the depth of obstacles, the maximum error is controlled within 5.6%, and the FPS can reach 3.8, which meets the requirements of real-time and distance accuracy required by the blind to avoid obstacles.

Abstract:With the advancement of artificial intelligence technology, baby monitoring systems have become increasingly prevalent in daily life. This paper presents an AI-based infant behavior monitoring system that utilizes computer vision techniques and deep learning algorithms, integrated with hardware components such as the Raspberry Pi 4B and Camera V2, to achieve real-time monitoring and intelligent analysis of infant behavior. The system employs the Google MediaPipe pose recognition algorithm to extract infant joint features within predefined safety zones and uses an optimized Moondream 2 model for multimodal data inference, significantly enhancing the system′s real-time responsiveness and accuracy. Additionally, the system incorporates a lightweight time-series analysis module to improve sensitivity to behavioral changes and integrates dynamic alert functions to ensure efficient and reliable monitoring. By leveraging the Home Assistant platform, MQTT protocol, and network tunneling technology, the system supports remote access and real-time notification capabilities. Experimental results demonstrate excellent performance in terms of accuracy and stability, making the system widely applicable in home monitoring and intelligent caregiving scenarios, and providing a novel solution for the safety management of infants and young children.

Abstract:Existing security research on microgrid frequency control systems lacks a comprehensive analysis of severe attack scenarios, particularly high-concealment attacks executed by adversaries using internal information. The system vulnerabilities and the extent of their potential impact remain insufficiently assessed. This paper develops a load frequency control model of microgrid that incorporates wind, solar, and storage, and performs a vulnerability analysis of its communication layer to identify potential attack vectors. To address concealment constraints, an optimized attack model is formulated by introducing slack variables, which transforms the nonlinear optimization problem into a linear programming problem, enabling faster solutions and the generation of specific attack sequences. Finally, multiple attack tests are conducted on microgrids in islanded operation mode. Compared to traditional random attack methods, the proposed optimized attack sequence achieves approximately 40% improvement in attack effectiveness while maintaining over 95% stealth. The effects of key microgrid system parameters, different operation modes, and various renewable energy penetration rates on optimal attacks are analyzed. Results show that the proposed optimizationbased attack can significantly improve attack success rate and effectiveness while maintaining stealthy, indicating that microgrid systems remain potentially vulnerable to well-designed attacks.

Abstract:To address the challenge of short-term fluctuation of large-scale photovoltaic power generations pose a challenge to accurate energy measurement, this paper proposes a new method for run clustering for short-term fluctuation of photovoltaic based on improved Gaussian mixture model. Firstly, the characteristics of short-term fluctuation signals of photovoltaic output are analyzed based on the run theory. Secondly, to address the issue of excessive run and difficulty in extracting typical features in the power generation of photovoltaic, the clustering method based on the improved Gaussian mixture model is adopted to cluster the massive run. Furthermore, a subjective-objective fusion evaluation method for clustering results is proposed. Finally, the simulation results of on-site recorded waveforms from photovoltaic power stations show that, compared with other methods, the proposed method has an improvement in clustering result scores ranging from 1.1% to 61.4% in different aspects. The proposed method can maintain good clustering effects under different noise and outlier levels, with a decrease in the composite index score that is less than that of other algorithms by 0.92% to 18.24%. The proposed method achieves adaptive clustering of the Gaussian mixture model through deep learning technology and the Bayesian information criterion, enhancing its adaptability and stability to noisy and outlier data, and enabling reasonable clustering of run-lengths of photovoltaic power station short-term fluctuation signals.

Abstract:The Fabry-Perot demodulation method suffers from low sampling frequency, leading to errors when measuring physical quantities with high frequency variations. To address this issue, this paper proposes a high-performance multi-channel fiber Bragg grating wavelength demodulation system using a tunable laser, designed to meet the application requirements of fiber Bragg grating sensors in high-precision and high-frequency measurements. A fast synchronous refresh program was developed to enhance the demodulation frequency and real-time performance, while a dual-core data processing program was implemented to optimize data processing efficiency. Functional and performance tests were conducted using a motor and an isometric beam. Results showed that the proposed system achieved a scanning frequency of 100 Hz and an average fitting error of 6.23 pm, significantly outperforming the comparison system with an average fitting error of 24.10 pm. The linearity reached R2=0.999 9, higher than the comparison system′s R2=0.999 5, validating its feasibility in high-performance fiber Bragg grating demodulation applications.

Abstract:Early smoke detection is an effective means to eliminate fire hazards in a timely manner, but the small size and complex diffusion form of smoke in the early stage of a fire make its detection extremely difficult. To address the above problems, this paper proposes a multi-path enhanced feature-based YOLO （MEF-YOLO）early smoke detection algorithm, which adopts QA-ELAN to improve the backbone network and optimise the model complexity and accuracy, and develops FGCA to autonomously enhance the feature differences between the sampling regions to effectively capture the spatial information of the smoke. And the feature fusion path is optimised by the MEFAN, which realises the direct interaction between cross-level features and effectively mitigates the loss of detail information; and a Wise-IOU loss function is introduced, which comprehensively takes into account the position and scaling information through the weight adjustment mechanism to improve the robustness of the model in the complex scene. The experimental results show that the algorithm proposed in this paper has an accuracy of up to 92.5% for early smoke detection in experimental scenarios with different lighting and small-scale smoke and smoke diffusion, and has a lightweight advantage, with the number of parameters and GFLOPs reduced by 27.5% and 30.6%, respectively.

Abstract:Helmet detection often faces challenges in complex road scenarios such as heavy traffic, pedestrian interference, and severe occlusion of targets. These conditions can easily lead to low detection accuracy, false detections, and missed detections. This paper proposes a high-performance helmet recognition model based on the CPM-YOLO algorithm. First, a novel cross-scale feature fusion method, CS-FPN, is proposed to better integrate high-level semantic and low-level geometric feature information. Next, the PCT module is introduced to optimize feature extraction capabilities of the model. Additionally, a bounding box regression loss function based on the minimum point distance is adopted to enhance the model′s convergence speed and accuracy. Furthermore, the 20×20 downsampling layer and 20×20 detection head in the backbone network are removed, and a new 160×160 small-object detection head is introduced. Finally, ablation studies validate the effectiveness of each improved module in enhancing the model′s performance, and comparative experiments demonstrate the superiority and generalizability of the CPM-YOLO model.Experimental results show that compared to the baseline model, the proposed method achieves improvements of 5.5% in mAP@0.5. Additionally, the number of parameters and model size are reduced by 69.9% and 67.2%, respectively. The new model significantly reduces complexity while enhancing helmet detection capabilities in road environments.

Abstract:Detecting the wearing of safety helmets by construction workers is an important method to ensure personnel safety. However, existing safety helmet detection methods are mostly manual, which are not only time-consuming and labor-intensive but also inefficient. Moreover, the existing algorithm has low detection accuracy in the face of complex environment or weather. In response to this phenomenon, an improved safety helmet wearing detection algorithm is proposed based on the YOLOv5s algorithm. Firstly, the SLSKA-POOL module is proposed based on the residual idea and large separable module design, and used in the pooling layer. This module can make the network pay more attention to the target features and further improve the network capability; secondly, the CAKConv convolutional module is proposed, which efficiently extracts features through irregular convolution operation to improve the network performance; finally, EMA modules are added to the backbone to aggregate multi-scale spatial structure information and establish short and short dependencies to achieve better performance. The experimental results show that: the improved YOLOv5 compared with the original algorithm, The detection accuracy increased by 2.2%, mAP@0.5 increased by 3.6%, and mAP@ 0.5:0.95 increased by 6.4%, realizing more accurate and efficient helmet wearing detection.

Abstract:In response to the problems of imbalanced samples and low prediction accuracy, an enhanced predictive recurrent neural network EN_PredRNN is proposed. Firstly, the radar data is preprocessed and samples are selected to construct a highquality radar echo dataset; then, deep fusion of spatiotemporal long short-term memory units and dynamic convolution is used to design a dynamic convolution combined with spatio temporal long short term memory module DC_STLSTM, which adjusts convolution parameters in real-time to accurately capture the instantaneous changes in radar echoes. Then, stack 5 layers of DC_STLSTM to extract deeper features of radar echoes, and use gradient highways to alleviate gradient vanishing, improving the model′s generalization ability and prediction accuracy. The experimental results showed that EN_PredRNN performed the best, significantly improving the critical success index and reducing false alarm rates. Compared with PredRNN, it increased the critical success index by 19.3%, 17.3%, 16.5% and 14.0% at 25, 35, 45 and 65 dBZ, respectively, while reducing false alarm rates by 28.3%, 27.5%, 26.7% and 24.9%, effectively. This model effectively learned the spatiotemporal variation characteristics of radar data and accurately predicted the radar echo intensity and location.

Abstract:Currently, the results of iris recognition cannot be applied to judicial trials. The forensic science community has begun to focus on quantitative identification method based on the statistical rules of interpretable iris features. For this purpose, it is necessary to achieve automatic segmentation and extraction of iris texture features. A segmentation network for block-shaped iris features in near-infrared iris images is proposed, which combines residual networks, attention mechanisms, and atrous spatial pyramid pooling. First, a block-shaped iris feature annotation dataset was constructed for model training, validation, and testing. Secondly, improvements were made to the UNet framework as follows: the convolutional modules were replaced with residual modules to promote gradient propagation and enhance feature retention and transmission capabilities; attention gate modules were added to the skip connections to improve the model′s perception of block-shaped features; and atrous spatial pyramid pooling modules were employed in the middle and end parts of the model to expand the receptive field and perform multi-scale feature extraction and fusion. Finally, a hybrid loss function combining cross-entropy and Dice coefficient was proposed, and preprocessing techniques such as normalization and histogram equalization were used to highlight block-shaped iris features. Experimental results show that the RAA-UNet outperforms other comparison models on the test set, with F1 score, mIoU, and Dice coefficient reaching 82.64%, 84.21%, and 81.66%, respectively, representing improvements of 4.42%, 3.37%, and 3.87% over UNet. The loss function experiments determined the optimal weight factor, and histogram equalization significantly improved segmentation performance. Ablation experiments verified the effectiveness of the model improvements. The proposed RAA-UNet semantic segmentation model can accurately segment block-shaped iris features, providing technical support for iris identification research.

Abstract:Forest resources are key natural resources, and forestry economy is also an important component of the national economy. However, natural disasters occur frequently in China's forests, and traditional post-disaster forest resource detection methods face challenges such as low efficiency and insufficient accuracy. This article designs and implements a post-disaster all-terrain forest plant resource detection and positioning system based on drone technology. The aim is to enhance detection efficiency and accuracy through artificial intelligence image recognition,Beidou positioning, and remote sensing technology. The system utilizes the DJI Phantom 4 Pro drone as its platform, equipped with high-resolution cameras, WIFI modules, and Beidou positioning modules, enabling rapid identification and precise positioning of post-disaster forest plant resources. Experimental results demonstrate that the system exhibits high reliability in flight performance, data transmission stability, image recognition accuracy, and positioning precision, achieving a recognition rate of nearly 90% and positioning accuracy down to the centimeter level. This system offers efficient and low-cost technical support for post-disaster forest resource management, highlighting significant application value.