A ship infrared image target detection algorithm based on YOLO11n, named AGT-YOLO, is proposed to address the issues of low model accuracy and recall rate, difficulties in identifying small targets, and multi-scale recognition challenges under complex sea conditions. By introducing an improved GhostHGNetv2 network, the background discrimination capability is enhanced; the designed ASFP2 optimized neck network improves detection capabilities for low-resolution images and very small targets; the proposed Tack Adaptive Alignment Detection Head ( TAADH ) replaces the original detection head, enhancing localization and classification performance;meanwhile, the AFGCAttention mechanism is integrated to improve global information processing capability and the model's generalization ability. Experimental results show that compared to the baseline model YOLO11n, AGT-YOLO achieves a 4.4% increase in recall rate and a 3.1% increase in mean average precision at IoU=0.5 ( mAP@ 50), demonstrating strong multi-scale recognition capability and robustness in complex environments.

To solve the problems of tracking and positioning of airborne early warning aircraft and other long-range targets，a high-precision passive positioning strategy based on the collaborative networking of UAVs and passive radar is proposed. By analyzing the impact of the layout of multi-station passive radar on positioning accuracy，a layout strategy that can improve positioning accuracy and reduce layout costs is designed. Time Difference of Arrival (TDOA) positioning method is applied in the study，and on this basis，the Geometric Dilution of Precision ( GDOP) for multiple stations is analyzed to evaluate the impact of the layout form on positioning accuracy. A collaborative strategy is proposed that，when the initial direction of the target is unknown，first uses a star-shaped layout for initial positioning and then switches to an inverted triangular layout for high-precision secondary positioning. The optimal secondary stations are selected using the “virtual structure method”and the flight trajectory of the UAV is optimized using an improved Particle Swarm Optimization ( PSO) algorithm to achieve high-precision layout. Simulation results show that this strategy can significantly improve positioning accuracy. Compared with traditional passive radar systems，positioning error is significantly reduced，and system response speed is faster. The research results have certain application value in practice.

Target detection in autonomous driving scenarios faces challenges such as complex environmental interference, multi-scale target distribution and target occlusion, and existing algorithms are still deficient in feature fusion capability, detail characterization accuracy and localization regression performance. To this end, an improved YOLOv8 detection algorithm, DMP-YOLO, is proposed. The original neck structure is optimized using Multi-Branch Auxiliary Feature Pyramid Network (MAFPN) to enhance the multi-scale feature fusion capability in complex traffic scenarios; C2f_DEConv is proposed in backbone network module, which replaces the standard convolution with Detail-Enhanced Convolution (DEConv) to significantly improve the detail capturing ability of small-scale vehicles and occluded targets through high-frequency feature preservation and local texture enhancement; the Powerful Intersection over Union version 2 (PIoUv2) loss function is introduced to optimize the improved bounding-box loss, which improves the regression accuracy of the target bounding-box through the optimization of dynamic scale-sensitive factors and geometric constraints. Experiments on the KITTI dataset demonstrate that DMP-YOLO achieves significant improvements across all key performance metrics, with mAP@0.5 reaching 89.0% (2.6% improvement compared with the baseline YOLOv8) as well as 2.9% improvement for mAP@0.5: 0.95, which provides an effective solution for high-precision real-time detection in autonomous driving scenarios.

To address the communication technology requirements for online monitoring and digital operation & maintenance of high-voltage transmission lines, as well as the shortcomings of existing communication methods such as optical fiber, 4G/5G in terms of adaptability to complex environments, coverage integrity, and cost control, the characteristics and requirements of current communication methods for high-voltage transmission lines are analyzed. Relying on a National Key Research and Development Plan project, a solution for a highly reliable broadband ultra-multi-hop wireless ad hoc network communication system is studied and proposed, which overcomes the technical issue of a sharp decline in quality of service after multi-hop wireless transmission, and a secure broadband ultra-multi-hop wireless ad hoc network communication system is constructed, realizing long-distance broadband service transmission with Quality of Service(QoS) assurance. The constructed ultra-multi-hop wireless ad hoc network communication system is simulated and tested though the OMNeT++ simulation platform, 9-node outdoor field tests, and on-site operation in the 220 kV Binxing First Line of State Grid Tianjin. The simulation and test results show that the system can achieve 50-hop broadband wireless data transmission with an end-to-end traffic of no less than 2 Mb/s. Compared with traditional technical route, this system features stronger technical adaptability and lower operation and maintenance costs. It enhances the digital operation and maintenance level of power grids and provides a reliable solution for the construction of communication networks in new-type power systems.

The BeiDou-3 Global Navigation Satellite System ( BDS-3) can provide data of six frequencies at present, which provides more choices for Multi-frequency Carrier Ambiguity Resolution (MCAR). Focusing on BDS-3, the basic method of Geometry and Ionosphere Free (GIF) model in MCAR is comprehensively studied, including the application of three-frequency GIF model in ambiguity resolution. Based on the theory of three-frequency linear combinations, the basic mathematical model under three-frequency is given. The optimal frequency combination for ambiguity resolution using GIF model is discussed under the possibility of 20 combinations of any three of the six frequencies. Meanwhile, the optimal linear combination of each frequency combination is also systematically discussed. In addition, the high-quality linear combinations for single-epoch ambiguity resolution using GIF model are also analyzed. The experiment is carried out by using the real BDS-3 six-frequency data. Through theoretical analysis and practical demonstration, the results show that when using the GIF model for ambiguity resolution, the optimal frequency combination is (B1C, B3I, B2a). In this method, if the sum of the coefficients of two fixed Extra Wide Lane /Wide Lane (EWL/WL) combinations equals zero, the standard deviation of the ambiguity for the third Narrow Lane (NL) combination is theoretically dependent solely on the frequency characteristics. However, due to the influence of unmodeled errors, the actual results may deviate from theoretical expectations. The GIF model effectively eliminates ionospheric delay effects and avoids Geometry Base ( GB) errors, demonstrating significant advantages. The ambiguity resolution based on GIF model exhibits strong potential particularly in ionosphere-active environments and medium-to-long baseline scenarios.

An algorithm for weed recognition in beet fields based on improved YOLOv11 model is proposed to address the problems of low efficiency, low accuracy, and missed detection of small targets in complex real-world scenarios. The PoolFormer module and AKConv module are introduced into the backbone network to enhance the model's ability to capture global semantic information to improve detection accuracy, enhancing the detection performance in low resolution images and small objects. The AKConv module improves the feature extraction ability of the model for beets and weeds with irregular growth patterns by dynamically adjusting the convolution kernel parameters and shapes, while the PoolFormer module can effectively segment the edge features of beets and weeds that cover each other. Secondly, the High-level Screening Feature Pyramid Network (HS-FPN) module is added to the head network to enhance the efficiency of multi-scale fusion and improve the feature extraction efficiency and speed of beets and weeds during the seedling stage. Through experiments, it is found that the improved YOLOv11 model achieves increases of 6.9%, 7.8%, 7.9%, and 7.8% in precision, recall, mAP@0.5 and mAP@0.5: 0.95, respectively, compared to the original model. The results show that this algorithm has achieved significant improvement in weed recognition in beet fields, providing a more feasible solution for detecting weeds in beet fields in complex scenarios.

Vision-Language-Action (VLA) models are a core technology for achieving general embodied artificial intelligence, aiming to integrate visual perception, language understanding, and action decision-making within a unified end-to-end framework. The current research status and development trajectory of VLA models are comprehensively and systematically reviewed. The theoretical origins of VLA models are traced, and the paradigm shift from modular designs to unified architectures is clarified. Along the evolutionary path of VLA, representative works such as SpatialVLA, TLA, and GR00T N1 are presented with a focus on multimodal fusion and cognitive hierarchies. A detailed taxonomy of VLA models is constructed from two key dimensions-macro architecture and system hierarchy. Key technologies and design principles are deeply analyzed, ranging from pioneering works such as RT-1, to models introducing large-scale knowledge transfer such as RT-2, OpenVLA, and ECOT, and further to cutting-edge dual-system architectures such as Helix, OpenHelix, DexVLA, and DexGraspVLA. Mainstream simulation environments, core datasets, and benchmarks supporting VLA research are systematically integrated and reviewed. The application status and prospects of VLA models in robotic manipulation, autonomous navigation, and industrial automation are explored. Core challenges in current VLA research are analyzed, including generalization and data efficiency, long-horizon task planning, and real-time responsiveness. Future research directions are discussed, including integration with world models and enhancement of data efficiency.

With the development of technologies such as artificial intelligence, multi-agents ( e. g. , unmanned aerial vehicle swarms) have been increasingly applied in practical combat operations. The Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, designed to solve the coordination problems of multi-agents in cooperative environments, has become one of the mainstream applied algorithms in the multi-agent field owing to its unique Actor-Critic framework. To address the problems in multi-agent collaborative tasks during command and decision-making—including ambiguous role division and slow convergence of the algorithm's policy caused by information overload—an improved MADDPG algorithm incorporating a Dynamic Role Attention(DRA) mechanism, namely DRA-MADDPG, is proposed. This algorithm embeds a DRA module into the Actor-Critic framework, and achieves accurate optimization of division of labor and collaboration by dynamically adjusting the attention weights of each agent towards peers with different roles. Specifically, the role set ( reconnaissance, assault, command) and phase division ( exploration→execution→encirclement) for command tasks are defined, and on this basis, a role coordination matrix and phase adjustment coefficients are constructed. A DRA module is designed in the Critic network to calculate weights and filter key information by leveraging role relevance and task phases. Additionally, the Actor network is improved to generate targeted actions by integrating role responsibilities. Simulation experiments show that compared with MADDPG, the Area Under the Curve (AUC) of the cumulative training reward of DRA-MADDPG increases by 2.4%, and the task completion time decreases by 19.3%. Furthermore, comparative analysis of training reward curves reveals that DRA-MADDPG exhibits better learning efficiency in short-term training. It is demonstrated that this method is suitable for complex command and decision-making scenarios and provides a relatively efficient solution for multi-agent coordination.

A wideband radial power combiner based on Ridge Gap Waveguide ( RGW) is designed for high-power combining applications in the wide millimeter-wave frequency band. The combiner consists of two metallic plates, with the center of the lower plate fed by a coaxial line. The energy is reflected by multiple metallic conical structures positioned above the plate, and directed into the radial transmission lines, achieving equal power division. The radial transmission lines employ ridge gap waveguides, which is built up by placing pin-type Electromagnetic Band-Gap (EBG) structures on the metal bottom plates adjacent to the ridge, effectively reducing coupling among adjacent channels without the need for welding the upper and lower plates. Simulation results show that the combiner operates within the frequency range from 14.7 to 37.5 GHz, with a reflection coefficient of less than -15 dB and a transmission coefficient of approximately -6.1 dB. The measurement results show good agreement with the simulation results.

The rising incidence of skin lesions has made early screening for skin cancer increasingly critical. However, existing methods for skin lesion image segmentation often suffer from limitations in channel-wise information modeling, structural adaptability, and feature fusion, which can lead to inaccurate boundary delineation and insufficient utilization of crucial contextual information. To address these issues, a skin lesions image segmentation method based on attention mechanism and wavelet transform, termed AW-SkinNet, is proposed. The proposed approach employs a dual-branch collaborative attention module to extract spatial and channel-dependent features, integrates wavelet transform to enhance frequency-domain representations, and incorporates lightweight attention-guided sub-pixel upsampling to improve detail restoration and contextual understanding. Experimental results on the ISIC-2017 and ISIC-2018 skin lesion segmentation datasets demonstrate that the proposed method achieves higher segmentation accuracy compared with existing approaches for skin lesion image segmentation.