Deep Learning-based Methods for Scratch Detection on Filter Surface
During the manufacturing process of optical filters,scratches will appear on the surface of the filters.The use of manual inspection suffers from the problems of low efficiency and miss detection and misdetection,and this paper pro-pose a method to detect the scratches on filters and calculate the size of the scratches to address this problem.Firstly,the CBAM attention mechanism is added to the jump connection part of U-NET to enhance the expression ability of the shallow features,which makes the identified scratches more complete and accurate,and then the multi-scale feature fusion module is used to fuse the feature maps of each layer in the decoding stage,and the accuracy of this algorithm in identi-fying scratches is up to 95.17%,and then,the skeletonization algorithm is used to extract the centre line of the scratched area,and the length of the centre line is calculated as the length of the scratches.The centre line is extracted from the scratched area using the skeletonisation algorithm,the length of the centre line is calculated as the length of the scratch,and finally the width of the scratch is calculated using the method of maximum internal circle diameter.
optical filtersU-netquantitative analysis of scratches
NETL
NSTL
万方数据
