Insulator Defect Detection in Aerial Images Based on ADFE-Net

This study proposes an anchor-free insulator defect detection method based on an Adaptive Defect Feature Enhancement Network (ADFE-Net) for rapid and precise defect identification. The backbone network integrates dynamic snake convolution (DSConv) to enhance extraction of subtle defect features. A Feature Rotation Interaction Module (FRIM) is designed to strengthen semantic information exchange while suppressing background noise interference. Furthermore, a Single-output Weighted Bidirectional Feature Pyramid Network (SWBFPN) enables adaptive multi-scale feature fusion, particularly improving detection sensitivity for small defects. Experimental results show that ADFE-Net achieves 91.6% detection accuracy and 90.4% average precision, surpassing other state-of-the-art methods. When integrated with unmanned aerial vehicles (UAVs), this approach offers a promising solution for intelligent inspection in power grid systems.