In the task allocation problem for multiUAV(Multiple Unmanned Aerial Vehicles) cooperative reconnaissance and strike missions, insufficient solution accuracy often leads to redundant mission paths and reduced strike effectiveness. This paper constructs an optimized task allocation model tailored for complex battlefield conditions. To enhance computational efficiency, a Genetic-Wolf Pack Algorithm (GWPA-PT) integrated with genetic strategies is proposed. This approach discretizes the wandering, summoning, and sieging behaviors of the wolf pack algorithm while introducing an individual replenishment mechanism to enhance its applicability to discrete task allocation scenarios. Additionally, an elite experience learning mechanism was designed to guide population reconstruction using the superior allocation experience of elite groups, thereby significantly enhancing iteration efficiency. Comparative experiments demonstrate that compared to typical multi-UAV task allocation algorithms in recent years, the GWPA-PT algorithm exhibits outstanding accuracy in solving problems, providing an efficient and reliable solution for collaborative task allocation in complex environments.

In this paper, we propose a robust signal reconstruction and direction of arrival (DOA) estimation method based on a generative adversarial network (GAN) framework under array element failures. The proposed signal reconstruction GAN (SR-GAN) consists of two components: a generator and a discriminator. The generator integrates both upsampling and downsampling operations to capture multi-resolution information from the input data and jointly leverages adversarial loss and content loss to ensure signal generation capability and reconstruction accuracy. The discriminator is designed to perform multi-scale feature extraction, enabling it to distinguish between real and reconstructed covariance matrix effectively. Simulation results demonstrate that the proposed method achieves excellent signal recovery and parameter estimation performance under varying signal-to-noise ratios (SNRs), numbers of snapshots, and configurations of missing array elements.

To address the dynamically changing navigable space caused by incomplete environmental information and conflict oscillations in parallel planning for multiple unmanned aerial vehicles (UAVs), a collaborative real-time path planning method integrating A* and Model Predictive Control (MPC) is proposed. This method maintains goal-directed planning by dynamically generating A*-guided reference paths and setting temporary sub-goals when the goal is temporarily unreachable. It improves tracking stability by downsampling and smoothing discrete paths. Furthermore, MPC-based optimization is performed over a finite prediction horizon, explicitly considering system dynamics constraints and control input bounds. The safety distance between UAVs is incorporated into receding-horizon optimization through constraints and penalty terms, and closedloop updates are achieved by applying the first control action. This approach improves the real-time performance, stability, and safety of online path planning.

In multi-agent reinforcement learning (MARL), the action-state space grows exponentially as the number of agents increases, resulting in the curse of dimensionality. This leads to a significant decline in sample efficiency and algorithm performance. In this paper, we propose a novel network model, GAIN, which integrates graph theory with action interactions. Using a self-attention mechanism, GAIN generates a weighted graph-based action interaction network that captures the complex interaction dynamics between agents, thereby mitigating the complexity of the action-state space. Experimental results demonstrate that, in six scenarios of StarCraft II, GAIN achieves a winning rate close to 100

The rapid advancement of unmanned aerial vehicle (UAV) technology has underscored the significant potential of multi-UAV systems in managing complex tasks. Efficient task-allocation strategies are crucial for enhancing the overall performance of these systems. Although conventional methods perform adequately in simple environments, they often struggle in more complex scenarios where environmental disturbances and resource constraints hinder their effectiveness, resulting in suboptimal task allocation outcomes. By contrast, reinforcement learning (RL), as a powerful optimization technique, is particularly suitable for addressing the challenges inherent in multi-UAV task allocation. Unlike conventional approaches, RL does not rely on predefined models or external knowledge, enabling the system to learn optimal strategies via continuous interactions with the environment. This flexibility enables the system to adapt to dynamic conditions and improve its decision making over time. This study proposes an innovative approach based on deep reinforcement learning to address the challenges encountered in multi-UAV task allocation, with specific consideration given to the uncertainties typically prevalent in real-world battlefield scenarios. These uncertainties include variable wind conditions, precipitation, and other environmental factors that can potentially affect UAV performance. The primary objective of this study is to ensure that multi-UAV systems can respond rapidly to multiple simultaneous tasks while optimizing resource utilization. Traditional task allocation methods, which are often heuristic or rule-based, lack the flexibility required to handle environmental complexity or dynamic changes. They are typically rigid and struggle to adapt to unanticipated situations, which results in inefficiencies and delays in task allocation. To address these challenges, this study modeled the task allocation problem as a Markov Decision Process. In this framework, the system can select the most appropriate task allocation strategy based on the current state of the environment, ensuring flexibility and timeliness in decision making. To enhance the stability and robustness of the model, an evaluation network and a target network were designed in tandem to ensure reliable learning. By separating the state and advantage values, the model effectively reduces the noise introduced by action selection, resulting in more accurate predictions and enhanced decision making. In addition, this study introduces a prioritized experience replay module that ranks the importance of each experience sample based on its temporal difference error, thereby prioritizing the most useful experiences for learning. This approach enables the model to focus on more informative samples, thereby accelerating the learning process and improving algorithm efficiency. By addressing the inefficiencies of traditional experience replay methods, which often reuse low-value samples, this technique ensures a more efficient use of the available training time. Moreover, this study employed neural network approximation techniques to reduce the computational demands of online learning, which is particularly important in real-time applications with limited processing power. Experimental results demonstrate that the proposed method substantially reduces resource waste in UAV task scheduling. On average, each UAV assignment is completed in just 0.24 s, indicating substantial improvement in task allocation efficiency. The proposed algorithm outperforms traditional methods in efficiency as well as in convergence speed and stability, owing to the prioritized experience replay module. Furthermore, the scalability of the algorithm was validated via simulations involving larger UAV fleets, where performance remained robust without degradation. Additional simulation tests confirmed that the proposed method can optimize resource allocation, reduce system interference, and accelerate convergence. In conclusion, the proposed method offers significant improvements in multi-UAV system task allocation, particularly in terms of task allocation efficiency and system adaptability.

High-impedance fault (HIF) in distribution networks poses detection challenges due to inherently weak signal characteristics, which are easily confused with normal operating conditions. Traditional detection methods employ single-domain feature extraction, resulting in inadequate characterization and poor performance under adverse interference conditions. To address this, we propose a novel deep learning framework for HIF detection based on multi-domain feature fusion. The method comprises three components. A xLSTM-KAN temporal feature extractor is developed, which utilizes Extended Long Short-Term Memory (xLSTM)'s matrix memory and exponential gating to identify subtle temporal features in HIF signals. Kolmogorov-Arnold Networks (KAN) serve as feature transformers with learnable activation functions to optimize HIF feature representation. Subsequently, Markov Transition Fields (MTF) convert zero-sequence voltage signals into two-dimensional image representations. The spatial-frequency collaborative module (SFC) extracts spatial domain and Fast Fourier Transform (FFT)-based frequency domain features from the MTF images. Furthermore, a feature crossing enhancement module enables bidirectional interaction and fusion between temporal features and MTF image features. MATLAB/SIMULINK validation confirms the effectiveness of the proposed method. The approach accurately and rapidly identifies HIF under ideal conditions and maintains 98% detection accuracy at 20 dB signal-to-noise ratio, demonstrating significant potential for practical engineering deployment.

The search for Multi-UAV cooperative targets in complex environments remains challenging because the joint action space increases rapidly with swarm size, while sensor observations are constrained by limited range, field of view, and obstacle-induced sight occlusion. To address the common limitations of search redundancy and local-optimal traps in existing methods, this study proposes an integrated multi-UAV active cooperative search framework driven by EIG-TS. The framework is underpinned by the symbiotic synergy between macro-space decoupling and micro-heuristic exploration. Specifically, a dynamic Voronoi partitioning mechanism utilizes real-time swarm positions to mathematically decouple the high dimensional joint action space into subspaces, eliminating cooperative search redundancy at the macro level. Within each dynamically assigned region, the proposed EIG-TS strategy combines expected entropy reduction, Beta Thompson sampling, and distance-cost scoring to balance exploitation, uncertainty driven exploration, and flight cost control, while suppressing micro level local-optimal behavior. Simulation results demonstrate that the proposed integrated strategy achieves a target detection success rate of 96.67% and reduces invalid flight distance by 89.1% compared with conventional baselines. Ablation experiments verify that the active exploration engine and the partitioning mechanism improve the final uncertainty accuracy by 33.7%, demonstrating superior environmental adaptability and improving the robustness of multi-UAV search.

To address the difficulty of fault line selection for single-phase grounding faults in small current grounding systems, a fault line selection method for distribution networks based on CWT-YOLOv11 is proposed. First, the current signal is transformed into a feature map using Continuous Wavelet Transform (CWT), achieving the conversion from a one-dimensional time series to a two-dimensional spatial domain image. Next, the converted feature map is stitched together, providing valuable material for deep learning-based fault diagnosis in power distribution networks. Finally, an improved lightweight YOLOv11 (You Only Look Once version 11) network is used to train the feature map, extract fault characteristics, and achieve precise fault line selection. Experimental results show that this method achieves a fault line selection accuracy of 99.99%, significantly reducing the model’s parameter count and computational complexity while maintaining high precision, thus realizing an efficient and precise fault line selection process.

To overcome the limitations of software demodulation in laser vibration measurement, a hardware demodulation scheme based on a field-programmable gate array (FPGA) is designed. The reference signal is synthesized using direct digital frequency synthesis and mixed with the Doppler signal. An orthogonal baseband signal is then generated through Butterworth low-pass filtering. Subsequently, a nonlinear error compensation algorithm is introduced to rectify errors. The CORDIC algorithm is applied for inverse tangent demodulation of the laser signal, while a phase detuning algorithm addresses signal hopping issues. Following data processing, the fast Fourier algorithm is used for time-frequency analysis of the vibration data signal. The proposed algorithm undergoes theoretical analysis and experimental verification, ensuring real-time vibration detection and meeting high precision requirements. This offers a more concise and efficient solution for laser vibration measurement systems.

Unmanned aerial vehicle (UAV)-assisted systems have attracted a lot of attention due to its high probability of line-of-sight (LoS) connections and flexible deployment. In this paper, we aim to minimize the upload time required for the UAV to collect information from the sensor nodes in disaster scenario, while optimizing the deployment position of UAV. In order to get the deployment solution quickly, a data-driven approach is proposed in which an optimization strategy acts as the expert. Considering that images could capture the spatial configurations well, we use a convolutional neural network (CNN) to learn how to place the UAV. In the end, the simulation results demonstrate the effectiveness and generalization of the proposed method. After training, our CNN can generate UAV configuration faster than the general optimization-based algorithm.

Feature point detection algorithms have been widely used in the fields of object recognition, panorama stitching, and robot navigation. SIFT algorithm is a robust feature detection method widely used in image processing and computer vision. This design proposes a SIFT algorithm based on FPGA development platform implementation, which reduces the computational complexity and improves the processing speed by adopting a pipelined architecture. The matching results show that the algorithm has good invariance to image rotation, illumination, affine, scale and can meet the needs of feature matching, and there is a certain practical application value.

Most of the railroad fastener images have the problems of low contrast, complex image noise and difficult to recognize edge details.For these difficult problems, this paper improves the traditional Canny algorithm by using a hybrid filter composed of adaptive median filter and Gaussian filter to preprocess the image; improves the Sobel operator by using 5*5 eight-direction templates instead of the traditional 3*3 four-direction templates, and expands the computational direction of the non-maximum value suppression module to eight directions.The experimental results show that the improved algorithm has better image edge detection performance on FPGA, lower resource utilization and faster image processing.

Raw point clouds are usually sparse and incomplete, inevitably containing outliers or noise from 3D sensors. In this paper, an improved SA-Net based on an encoder-decoder structure is proposed to make it more robust in predicting complete point clouds. The encoder of the original SA-Net network is very sensitive to noise in the feature extraction process. Therefore, we use PointASNL as the encoder, which weights around the initial sampling points through the AS module (Adaptive Sampling Module) and adaptively adjusts the weight of the sampling points to effectively alleviate the bias effect of outliers. In order to fully mine the feature information of point clouds, it captures the neighborhood and long-distance dependencies of sampling points through the LNL module (Local-NonLocal Module), providing more accurate information for point cloud processing. Then, we use the encoder to extract local geometric features of the incomplete point cloud at different resolutions.Then, an attention mechanism is introduced to transfer the extracted features to a decoder. The decoder gradually refines the local features to achieve a more realistic effect. Experiments on the ShapeNet data set show that the improved point cloud completion network achieves the goal and reduces the average chamfer distance by 3.50% compared to SA-Net.

Unmanned Aerial Vehicles (UAVs) have critical applications in various real-world scenarios, including mapping unknown environments, military reconnaissance, and post-disaster search and rescue. In these scenarios where communication infrastructure is missing, UAVs will form an ad hoc network and perform tasks in a distributed manner. To efficiently carry out tasks, each UAV must acquire and share global status information and data from neighbors. Meanwhile, UAVs frequently operate in extreme conditions, including storms, lightning, and mountainous areas, which significantly degrade the quality of wireless communication. Additionally, the mobility of UAVs leads to dynamic changes in network topology. Therefore, we propose a method that utilizes graph neural networks (GNN) to learn cooperative data dissemination. This method leverages the network topology relationship and enables UAVs to learn a decision policy based on local data structure, ensuring that all UAVs can recover global information. We train the policy using reinforcement learning that enhances the effectiveness of each transmission. After repeated simulations, the results validate the effectiveness and generalization of the proposed method.

In recent years, deep learning-based network intrusion detection systems (IDS) have shown impressive results in detecting attacks. However, most existing IDS can only recognize known attacks that were included in their training data. When faced with unknown attacks, these systems are often unable to take appropriate actions and incorrectly classify them into known categories, leading to reduced detection performance. Furthermore, as the number and types of network attacks continue to increase, it becomes challenging for these IDS to update their model parameters promptly and adapt to new attack scenarios. To address these issues, this paper introduces a dynamic intrusion detection system, Dynamic Unknown Attack Intrusion Detection System (DUA-IDS). This system aims to learn and detect unknown attacks effectively. DUA-IDS comprises three components: Feature Extractor: This component employs CNN and Transformer models to extract data features from various perspectives. Threshold-Based Classifier: The second part utilizes the nearest mean rule of samples to classify known and unknown attacks, enabling the distinction between them. Dynamic Learning Module: The third part incorporates data playback and knowledge distillation techniques to retain existing category knowledge while continuously learning new attack categories. To assess the effectiveness of DUA-IDS, this paper conducted experiments using the UNSW-NB15 public dataset. The experimental results show that DUA-IDS improves the classification accuracy of flow network data with unknown traffic attacks. Can accurately distinguish unknown traffic and correctly classify known traffic. When dynamically learning unknown traffic, the classification accuracy of previously learned known traffic is less affected. This indicates the advantages of DUA-IDS in detecting unknown attacks and learning new attack categories.

Ant colony optimization (ACO) is a popular bionic technology optimization method commonly used for shortest path algorithms. However, this method is associated with various issues, including slow convergence rates and deep local optimal solutions. To address these problems, an improved ant colony optimization algorithm is proposed in this paper, utilizing a new heuristic function and an attractive update strategy to enhance the convergence speed and searchability of the ant colony algorithm. The integration of the artificial potential field function formula correctly guides the overall planning path of the ant colony, allowing it to bypass obstacles using the potential field force and disconnect redundant paths. Simulation experiments show that the algorithm proposed in this paper has obtained a better path and fewer iterations compared to the IAACO improved algorithm. This confirms that the algorithm is effective and reliable in path planning, particularly in complex scenarios and special environments.

This paper proposes a approach to the problem of low coverage rate in the area coverage path planning of unmanned aerial vehicle (UAV) under the traditional Double Deep Q Network (DDQN). Our approach involves introducing noise parameters into the DDQN and using exploratory noise for deep exploration, which improves the efficiency of exploration. Compared to the traditional DDQN algorithm, our improved method enhances the exploration ability and convergence speed of the trained network, ultimately achieving a higher level of periodic reward. This enables the UAV to better avoid obstacles or no-fly zones in coverage path planning, avoiding it from being stuck or repeating paths in local environments, thereby enabling the UAV to cover more target areas in a shorter number of steps.

Due to the limitation of the resolution of 3D scanning equipment such as lidar, as well as the limitation of the scanning angle and mutual occlusion between objects, processing methods and other reasons, the collected 3D point cloud data is often sparse and incomplete. Existing point cloud completion networks do not sufficiently consider the importance of generating a rough shape with fine-grained details. Therefore, we adopt a coarse to fine strategy and introduce a shape completion network with dual branches. The first branch is used to generate a rough shape with some structural details, which is used to generate global features. In the second branch, we employ fine-grained shape completion from the encoder-decoder. These two branches share a global feature to complete the shape completion. The multi-scale point cloud features extracted by the encoder are fused with the global feature to provide accurate global information for the decoder to reconstruct the geometric details of the missing regions. In the design of the total loss function, we calculate the loss difference between the generated rough point cloud, refined point cloud and the real point cloud, which accumulate them. Quantitative and qualitative results of our method on the ShapeNet dataset demonstrate that our algorithm produces more realistic outputs than mainstream point cloud completion algorithms.

Thanks to the flexible deployment, mobility and low cost, unmanned aerial vehicle (UAV) will be widely used in civil and military scenarios. In this paper, a flying network in the sky is considered, where multiple aerial source nodes are equipped with sensing devices to explore the environment and transmit the collected information to the ground station for data fusion and processing. However, there exists distance restriction due to the signal attenuation. Therefore, a UAV is deployed as a flying relay to provide wireless connection. Firstly, we design a time-division multiple access scheme to satisfy the rate requirements. Secondly, to maximize the system throughput under the no-fly zones, flying speed and available energy constraints, we optimize the time, power and flying trajectory. Then we decompose it into three subproblems. To reduce computational complexity, we optimize the power allocation and flying trajectory by applying the Lagrange dual decomposition and the alternating direction method of multipliers, respectively. In conclusion, an overall algorithm is proposed to iteratively optimize these three variables. In the end, we obtain the solution and the simulation results show the effectiveness of the proposed algorithm.

考察熔体与界面之间的润湿性,对表征界面相互作用及分析材料工艺特性等具有重要意义.针对现有润湿角计算方法自动化程度不足、计算较大润湿角不够精确等局限,基于Sobel算子、矩形定位、Harris角点检测等图像处理算法和手段,提出一种熔体润湿角的全自动计算方法.Sobel算子实现了基座自动定位;应用矩形定位,采用外接矩形法、内接矩形法相结合自动定位样品,消除了主观操作所引入的误差;应用Harris角点检测方法,对非样品轮廓点进行了处理,得到了完整的样品轮廓.为了进一步提高准确性,将轮廓拆分为左右半边,并基于迭代重加权最小二乘法分别用椭圆拟合计算接触角.基于座滴法测量系统对铜和玻璃样品的润湿角进行测量,对轮廓拟合及润湿角计算误差进行统计分析,得出干扰对轮廓拟合的影响,开展精度计算,表明该方法能够在较大的温度区间内准确计算润湿角,并且在多次重复实验中标准偏差不高于0.7°,提高了轮廓拟合的精度;计算方法具有良好的可复现性.