Nondestructive testing surface defect of titanium alloy based on improved YOLOv5s

A titanium alloy surface defect detection approach based on improved YOLOv5 is proposed to resolve the problem of low accuracy and inefficiency caused by small target defects and complex background interference. First, an optical inspection system integrating an industrial camera and PC is designed to acquire high-quality titanium alloy images. Second, a titanium alloy surface defect dataset is established through systematic collection and data augmentation. Finally, the C3 module is replaced with C2f module to enhance feature fusion capability for small defects, while the Squeeze-and-Excitation (SE) channel attention mechanism is embedded in the backbone network to dynamically amplify defect-related features and suppress background noise. Experimental results demonstrate that the proposed YOLOv5sSC model achieves a 2.9% improvement in mAP@0.5 on the NEU-DET dataset compared to the baseline model while maintaining a real-time detection speed of 88.3 FPS. On the self-built titanium alloy dataset, mAP@0.5 increases by 2.6%, validating the superior balance of the improved algorithm between lightweight design and detection accuracy in complex industrial scene.