A method for multi-view surface defect detection of Si₃N₄ ceramic bearing balls integrating features enhanced by the Gabor salient domain

Abstract

Si3N4 ceramic bearing balls exhibit wear, pits, scratches, and delamination defects on the surface during manufacturing processes. Current Si3N4 ceramic ball detection methods mainly focus on a single view input, which leads to insufficient fusion of boundary, color, and shape features, consequently resulting in a low detection accuracy. In this research, propose multi-view surface defect detection of Si3N4 ceramic bearing balls integrating features enhanced by the Gabor salient domain (GSMF). Firstly, color, shape, and boundary information of defects are extracted from different angles, distances, and GSMF enhancement views. Secondly, by designing a salient domain enhancement module, GSMF enhancement boundary features are extracted, addressing the feature loss problem that results in scarce border information during decoding. By improving the co-attention of multi-view to prevent memory loss caused by long-distance transmission, more feature information is preserved. Finally, the accuracy of the detection method is validated through experimental tests.