Surface defect detection is a common engineering task, and traditional human visual inspection suffers from low efficiency, poor resolution, and a lack of traceable data. We propose a surface-topography detection system based on multicolor latitudinal and longitudinal light sources. By activating multicolor latitudinal and multicolor longitudinal light sources at the same position but at different times, decoupling and highlighting of the surface gradient magnitude and direction in the hue domain of the target surface can be achieved. We used a hot-rolled aluminum sheet as a target and calibrated the vision system, achieving recovery of the gradient magnitude and direction, which provided the necessary conditions for defect data analysis. We also used a hot-rolled aluminum sheet as the target for collecting co-located latitudinal and longitudinal images. In these images, various surface defects were highlighted relative to the background, with good differentiation between the different defects.
Read full abstract