RD-YOLO: Towards Power Components Identification and Rust defect Detection for Future Unmanned Transmission Lines Maintenance

Abstract As a common defect type, rust defect of power components is one of the important hidden dangers that harm the safe operation of transmission lines. How to quickly and accurately discover and repair the rusted power components is an urgent problem to be solved in power inspection. Aiming at the above problem, in this study, a Rust-Defect YOLO (RD-YOLO) is proposed to perform power components identification and rust defect detection in aerial images. Firstly, Coordinate Channel Attention and Residual Module (CCARM) is proposed to enhance multi-scale detection precision, which is applied to the backbone CSPDarknet-53 to suppress the influence of background noise and enhance features extraction capability of power components. Secondly, Receptive Field Block (RFB) and Efficient Convolutional Block Attention Module (ECBAM) are introduced to PANet to strengthen deep and shallow features fusion, which alleviate the information reduction or loss of small targets, and reduce the inconsistency of features between different levels. Finally, Contrast sample strategy and Focal loss function are adopted to train and optimize RD-YOLO, and experiments are executed on a self-bulit dataset. Experimental results show that the mAP of power components identification reaches 92.33%, which is improved 7.82% compared with the original YOLOX. More importantly, the AP of rust defect detection attains 95%, which is increased by 18%, 20%, 11%, 9% and 5% compared with those of Faster R-CNN, SSD, YOLOv5, YOLOX and YOLOv7. Comparative experimental results demonstrate that the proposed model achieves good performance in power components identification and rust defect detection, and is promising for the future automatic visual inspection of transmission lines.