A lightweight RKM-YOLO algorithm for transmission line fault inspection

The inspection of transmission lines using Unmanned Aerial Vehicles (UAVs) is crucial for ensuring reliable power supply systems. However, environmental complexity often leads to difficulties in locating equipment faults and maintenance delays, compromising power supply security. Given these challenges, this research presents a lightweight algorithm utilizing an enhanced version of YOLOv8n. First, the Receptive Field and Channel Attention Convolution (RFCAConv) module is introduced, capturing global and local features through a receptive field optimization mechanism, suppressing noise interference and mitigating image degradation caused by light refraction and reflection. We designed the K3_RFCA structure, which effectively integrates shallow and deep feature information to enhance the network’s feature extraction capability. Second, an improved KBiFPN structure is employed during feature fusion stage to enhance multi-scale adaptability while reducing computational costs. Finally, to tackle small target detection in transmission line inspection, an innovative Multi-scale and Multi-dimensional Cooperative Attention (MMCA) is proposed, adaptively weighting regional contextual information through multi-scale convolution and spatial-channel fusion, improving critical region representation. Experimental results demonstrate that RKM-YOLO achieves an average detection accuracy (mAP) of 90.7%, with an 18.27% reduction in parameters and a 3.1% improvement in recall and mAP50, while its detection speed meets the requirements for UAV inspections.