Deep learning-based defect detection for critical power grid components

With the advancement of the intelligent transformation of power systems, insulator defect detection, as a critical component of transmission lines, is crucial for the safe operation of power grids. Traditional manual inspection is inefficient and risky, while existing deep learning models still face challenges in generalization ability and real-time performance under dynamic environmental interference and diverse defect morphologies. In this study, an improved algorithm based on the You Only Look Once version 8 (YOLOv8) framework is proposed by introducing the C2f_SCConv module to optimize the feature extraction process. This module combines residual connections with selective convolution operations to dynamically adjust convolution kernel weights, suppress background noise, enhance defect feature representation, and reduce computational complexity. Experiments comparing Faster Region-based Convolutional Neural Network(Faster R-CNN), Single Shot MultiBox Detector (SSD), YOLOseries models verify the effectiveness of the improved algorithm. The results show that the proposed method outperforms YOLOv8s in precision, recall, and [email protected], achieving a balance between parameter count and detection speed for engineering applications, and providing an efficient and accurate solution for drone-based power inspection.