Insulator Defect Detection Based on Deep Learning
Li Liu, Jing Fu, Li Yan, Hong Xie · 2023
Insulators hold great significance in ensuring electrical insulation and power support in transmission lines. However, detecting insulator defects in UAV power images is challenging due to variations in insulator scale, dense image texture, and complex transmission line backgrounds. To address these challenges, we propose an advanced network model based on YOLOv5, incorporating Coordinate Attention (CA) and Transformer encoder blocks at multiple scales. The integration of CA mechanism enhances the accuracy of insulator defect detection, while replacing the C3 network with Transformer encoder blocks improves the identification of various insulator defects in dense scenarios. Additionally, a small target-oriented detection head is added to detect tiny insulator defects. Experiments demonstrate a significant improvement of our optimized model over the original YOLOv5, with an increase of3. 5% in mAP accuracy. Comparative analysis with the previous state-of-the-art detection model further validates the effectiveness and reliability of our improved model for insulator defect detection.