Abstract: Aiming at the issues of large parameter count and slow inference speed, which is difficult to meet the demand for real-time online electric power inspection, and considering the complex environmental interferences such as the fog or low illumination during actual unmanned aerial vehicle inspection processes, an insulator defect detection method based on improved lightweight YOLOv5s is proposed. This method achieves a balance between model lightweighting and detection accuracy through a triple optimization strategy. Firstly, the YOLOv5s network is deeply optimized by using lightweight modules, which significantly reduces the model complexity. Secondly, the network adopts the bidirectional feature pyramid network structure to enhance the multi-scale feature fusion capability, and combines the SIoU loss function to improve detection accuracy and convergence speed. Finally, the synthetic fog algorithm is employed to effectively enhance the generalization performance of the model in complex environments. Experimental results show that the parameters, floating point operands and size of the proposed model are only 1.74×10⁶、3.5×10⁹ and 3.9 MB, and the mAP@0.5 reaches 92.8%.

Key words: lightweight, insulator defect detection, synthetic fog data enhancement, aerial image