WTAYOLO: A Detection Model for Weak Scratches on Metal Surfaces

WTAYOLO targets weak scratches on metal surfaces, where defect contrast is low and normal texture or illumination variation can hide the scratch signal.

This work appears in Proceedings of the 2024 7th International Conference on Sensors, Signal and Image Processing, 22-28.

Algorithm principle

The algorithm frames inspection as robust defect localization under weak visual contrast. It extracts features from the image or 3D signal, suppresses normal texture and illumination variation, and then isolates abnormal regions that correspond to scratches, surface flaws, or dimensional defects.

Visual material

Detailed method and result figures will be added when the original paper assets are available.

Results and impact

The work is relevant to industrial surface inspection pipelines where small scratches must be detected despite weak visual contrast.