MC-YOLO: A Lightweight Insulator Defect Detection Model Based on an Improved YOLOv8

Abstract

Due to the fast growth of China’s electric power industry, the total length of high-voltage transmission lines has been continuously increasing. As a key component of high-voltage transmission systems, insulators play a critical role, and achieving efficient and accurate defect detection for insulators is of great significance. To address the practical challenges of limited size, limited computational resources, restricted energy supply, and complex environmental conditions on Unmanned Aerial Vehicle (UAV) platforms, this paper proposes a lightweight insulator defect detection model, MC-YOLO, based on an improved YOLOv8 architecture. Specifically, the original backbone network is replaced with the lighter MobileNetV3 module, reducing the model parameters and GFLOPs to 21.3% and 20% of those of the original model, respectively. In addition, a Convolutional Block Attention Module (CBAM) is integrated into the network neck structure to effectively improve the extraction of key features, resulting in a 1.2 percentage points improvement in detection accuracy. Finally, the loss function is changed to Wise-IoU (WIoU) v3, which increases the localization capacity of the model and further increases the accuracy by 1.4 percentage points. Experimental results demonstrate that the proposed MC-YOLO model achieves a lightweight design while maintaining high detection performance, providing a viable technical solution for edge deployment in real-world engineering applications.