改进YOLOv8的输电线路异物检测算法

Improved YOLOv8 algorithm for foreign object detection on transmission lines

摘要:

针对输电线路异物检测任务中存在的目标被遮挡、尺度不规则及复杂背景干扰等问题，本文提出了一种基于YOLOv8改进的输电线路异物检测算法SEPC-YOLOv8。该算法在主干网络中引入基于拆分的卷积(Split Based Convolution，SPConv)，以减少信息冗余并提高特征提取效率；在SPPF(spatial pyramid pooling fast)中嵌入空间增强注意力模块(Spatially Enhanced Attention Module，SEAM)，构建SPPF_SE模块，通过结合全局上下文感知和局部特征细化，改善因目标被遮挡引起的漏检问题；设计C2f_SAL模块替代颈部C2f(csplayer_2conv)，用于捕捉多尺度目标的关键特征，增强模型在复杂背景下的检测精度；同时引入Inner-PIoU(inner powerful intersection over union)损失函数，优化目标框定位精度。实验结果表明，所提算法的均值平均精度达到了96.36%，相比基线算法提升了3.02%，FPS达到51.69，能够满足输电线路中异物检测的实际需求。

Abstract:

To address the challenges of target occlusion, irregular scales, and complex background interference in transmission line foreign object detection tasks, this paper proposes an improved YOLOv8-based detection algorithm named SEPC-YOLOv8. The algorithm introduces Split-Based Convolution (SPConv) into the backbone network to reduce information redundancy and enhance feature extraction efficiency. A Spatially Enhanced Attention Module (SEAM) is embedded into SPPF (Spatial Pyramid Pooling Fast), constructing an SPPF_SE module that combines global contextual awareness with local feature refinement to mitigate missed detections caused by target occlusion. The C2f_SAL module is designed to replace the neck's C2f (csplayer_2conv) module, enabling effective capture of critical features in multi-scale targets and improving detection accuracy in complex backgrounds. Additionally, the Inner-PIoU (Inner Powerful Intersection over Union) loss function is introduced to optimize bounding box localization precision. Experimental results demonstrate that the proposed algorithm achieves a mean average precision of 96.36%, representing a 3.02% improvement over the baseline algorithm, while maintaining a detection speed of 51.69 FPS. These performance metrics confirm the algorithm's capability to meet practical requirements for foreign object detection in transmission lines.