IF-YOLO: An Efficient and Accurate Detection Algorithm for Insulator Faults in Transmission Lines

Insulators are critical components of transmission lines, and regular inspection of insulator defects is essential for the safe operation of power systems. To address the issues of low detection accuracy and missed detections in UAV(Unmanned Aerial Vehicle)-based insulator defect detection, and to meet the real-time detection requirements of UAVs, an efficient and lightweight insulator defect detection algorithm named IF-YOLO (Insulator Fault-You Only Look Once) is proposed based on YOLOv10n. First, the Haar Wavelet Upsampling (HWU) was designed and integrated into the network architecture, combining it with the Haar Wavelet Downsampling (HWD) to address the issue of feature information loss caused by strided and transposed convolutions. Second, the GCA (Group Collaborative Attention) is proposed and combined with C2f (CSPDarknet53 to 2-Stage FPN) to enhance the model’s ability to extract features of small insulator defects. Third, a Pyramid Bottleneck structure is designed to increase the model’s receptive field, preventing the loss of edge feature information during iterations to further enhance detection accuracy. Then, an HSPP (Hybrid Spatial Pyramid Pooling) is designed to enhance detection performance on defects in complex backgrounds. Finally, SlideLoss is incorporated into the loss function to improve the model’s detection performance on difficult samples. Experimental results show that IF-YOLO achieves a detection accuracy of 94.6%, an improvement of 4.0% over YOLOv10n. Additionally, the model achieves 170.7 FPS (Frames per second), indicating its capability for precise real-time detection. Heatmap and visualization further confirm that IF-YOLO significantly enhances the feature extraction of small targets and resolves the issue of missed detections.