Surface defect detection of metal parts: Use of multimodal illuminations and hyperspectral imaging algorithms

The illumination technique used in the industrial inspection systems can affect the quality of the acquired images. The choice of this technique plays an important role in the designing process of the system. Using multiple modalities can then avoid a difficult choice and increase the performances. We propose to consider multi-modal data by constructing pseudo-spectral cubes and by using hyperspectral imagery algorithms to detect 3D surface defects on metal parts. In this paper, four basic lighting modalities (white source light and monochromatic source light including non-polarized and polarized light) are proposed and tested to illuminate the metal parts, and different detection algorithms have been compared. We show that the spectral angle mapper is able to detect the defects with a non-supervised approach, and gives good detection performances.