A Surface Defect Detection Method Based on Domain Correlation

Detecting surface defects is a crucial aspect of the product production process. To achieve automated detection of these defects, numerous algorithms leveraging machine vision technology have been developed. However, regardless of whether traditional statistical learning algorithms or deep learning algorithms are used at this stage, many defect samples are needed for modeling. Focusing on this problem, this paper proposes a defect detection algorithm that is based on neighborhood correlation. On the basis of the theory of visual perception, the method establishes the neighborhood excitatory region of each pixel in the area to be detected. Then, several activation points within the excitation region are sought according to the correlation. Finally, a Gaussian model of each pixel is built on the basis of the statistical characteristics, and defect detection is performed accordingly. Experiments on multiple datasets show that this method only needs to learn that a positive sample image can quickly and effectively detect small defects on the product surface.