An insulator defect detection method based on scale perception and attention guidance

Abstract

This application discloses a scale-aware and attention-guided insulator defect detection method, belonging to the field of insulator defect detection technology. The method includes: introducing adaptive rectangular convolution into the feature extraction backbone network of YOLOv12n, using the height and width of the convolution kernel as learnable parameters and dynamically adjusting sampling points to adapt to defects of different scales and shapes; constructing a deformable interactive attention module, generating an attention mask through channel compression and modulation coefficient reconstruction to suppress background interference; and designing a detail-preserving context fusion module, using a gating mechanism to dynamically fuse multi-scale features along channels in groups, preserving details and enhancing semantics. Experimental results show that this invention improves detection accuracy while achieving model lightweighting, meeting the needs of insulator defect detection in practical industrial scenarios.

Images