Description: This work uses machine vision to detect defects on the wafer surface in the semiconductor industry to improve the yield of semiconductor products and improve the production speed of enterprises. Firstly, the industrial linear array camera is used to scan the whole wafer to obtain the complete wafer original image. Secondly, the grayscale characteristics of the wafer original image are extracted to correct the wafer image pose. Again, the single sample grain image and the complete wafer image are used for multi-template matching, and the best matching coordinates are selected by Kmeans and nonlinear difference method to complete the segmentation of grain samples. Finally, the image enhancement is carried out on the segmented samples, the enhanced image traversal extracts the outer contour feature and the contour grayscale feature, generates a one-dimensional array, carries out Kmeans and hierarchical clustering analysis, filters out discrete samples, marks discrete samples as defect samples, generates defect wafer maps, and completes wafer defect detection. This method can be applied to the wafer production manufacturing process, instead of the current manual inspection method.