Weak Feature Defect Detection of Insulators in Complex Images Based on Sequential Task Strategy

Detecting insulator defects from complex images poses a significant challenge using existing methods, especially for weak feature defects such as corrosion and notch. They exhibit limited spatial extent, complex morphology, and mutual interference. Therefore, this paper proposes a sequential task strategy for the detection of weak feature defects on insulators. Firstly, to address complex backgrounds, occlusions, and multi-scale issues, an insulator recognition method based on improved YOLOv5 is proposed, called feature refine fusion YOLOv5 (FRF-YOLOv5). This method realizes the accurate localization of insulators by introducing a spatially refined convolutional block attention module and a multi-scale feature fusion module. Then, DeepLabv3+ is cascaded, innovatively applied to the insulator defect detection, conducting pixel-level analysis on insulator re-gions and selectively replacing its backbone. Finally, the proposed method is validated. The results show it can effectively address the challenges of detecting insulator weak feature defects, and provide technical support for subsequent defect rating.