Defect detection via dual attention-guided YOLO for low-contrast infrared thermography images

Infrared thermal imaging technology, with its non-intuitive, visual, and real-time advantages, shows broad application potential in power equipment defect detection. However, infrared images suffer from challenges such as low contrast, blurred edges, and noise interference, which hinder defect identification. To enhance the accuracy and efficiency of infrared image-based defect detection, this paper proposes an improved YOLO object detection algorithm for defect detection from infrared thermography images using a dual attention mechanism (spatial attention and channel attention). By introducing the residual convolutional block attention (RCBA) module, we strengthen feature extraction capabilities by enhancing key regions and channels, thereby optimizing the feature extraction performance of YOLOv8. Experimental results demonstrate that this method achieves higher recognition accuracy and robustness in detecting representitive defects of power equipment, such as flashover, bubble defects, impurities etc.