WTAYOLO: A Detection Model for Weak Scratches on Metal Surfaces

Scratches on the surface of lithium batteries significantly affect their electrochemical performance and safety in operation. Therefore, stringent scratch detection protocols for lithium batteries are crucial for ensuring product quality. However, due to the reflective properties of lithium battery materials and the morphological characteristics of scratches, scratch detection algorithms still face significant challenges. To address this, this study proposes the WTAYO model for detecting weak scratches on metal surfaces, integrating photometric stereo technology to enhance the accuracy of scratch detection. The photometric stereo system effectively reduces the interference resulting from surface reflections. Additionally, the WTAConv module introduced in this paper expands the receptive field, thereby enhancing the ability to extract scratch features from images. It also strengthens important features through channel and spatial attention mechanisms while suppressing redundant information, thereby significantly enhancing the model's performance in detecting subtle scratches. Experimental results show that, without increasing the number of parameters, the mAP50 of WTAYOLO improved to 86.8%, reflecting a 5.1% improvement over YOLOv11n. This method strikes an effective balance between segmentation accuracy and model cost-effectiveness, providing an effective solution for industrial applications.