【摘要】 针对光伏电站入侵检测中存在的高漏检率、高误检率及实时性差的问题，提出了一种融合YOLOv10目标检测与改进射线-四叉树算法的智能模型。该模型通过设计轻量化网络、采用一致双重标签分配策略(无需NMS)并结合多尺度注意力模块，将参数量压缩至原版的78%,显著提升了小目标检测能力。改进的射线-四叉树算法通过动态区域分割与方向加权扫描，将计算复杂度从O(n)降至O(log n)。在自建数据集Solar-Security-1K上的实验表明，模型综合检测精度达96.2%,误检率降低41%,并在Jetson Xavier边缘设备上实现了45 FPS的实时检测，有效解决了复杂背景干扰与夜间场景下的安防需求。更多还原

【Abstract】 Aiming at the problems of high missed detection rate, high false alarm rate, and poor real-time performance in intrusion detection for photovoltaic power stations, this study proposes an intelligent model that integrates YOLOv10 object detection with an improved Ray-Quadtree algorithm. By designing a lightweight network, adopting a consistent dual-label assignment strategy(eliminating the need for NMS), and incorporating a multi-scale attention module, the model compresses the parameter count to 78% of the original version, significantly enhancing small object detection capability. The improved Ray-Quadtree algorithm reduces the computational complexity from O(n) to O(log n) through dynamic region partitioning and direction-weighted scanning. Experiments on the self-built dataset Solar-Security-1K show that the model achieves a comprehensive detection accuracy of 96.2%, reduces the false alarm rate by 41%, and achieves real-time detection at 45 FPS on the Jetson Xavier edge device, effectively addressing security requirements under complex background interference and nighttime scenarios.更多还原