IDD-YOLO: An Insulator Detection Algorithm for Distribution Networks

To address the low detection accuracy and high miss rate of insulators in UAV aerial images, caused by complex backgrounds, scale variation, and occlusion, this paper proposes IDD-YOLO, a high-precision insulator detection model based on an in-depth optimization of YOLOv8n. First, the C2f-MLCA (C2f with Multi-Layer Channel Attention) module enhances the network’s focus on insulator features and suppresses background noise. Second, the DGST (Dynamic Gated Spatial-Topology) module in the neck adaptively controls multi-scale feature fusion, improving robustness to morphological variation and occlusion. Additionally, the SPPF-LSKA (SPPF with Large Selective Kernel Attention) module enlarges the effective receptive field with minimal computational overhead. On a public insulator dataset, IDD-YOLO achieves a mAP@0.5 of 97.4%, surpassing YOLOv8n by 2.0 percentage points while maintaining 92 FPS, demonstrating excellent balance between accuracy and efficiency for intelligent power grid inspection.