Abstract: To address the issues in edge bead removal (EBR) inspection, such as interference of circuit pattern and process residue, differences in wafer appearance and contour shape, and how to identify the specified layer in multi-layer EBR, a wafer edge detection method based on machine vision is proposed. First, a high-definition image of the wafer edge is captured with a line scan camera. Then a segmentation point is initialized for each column of pixels to search and approach the EBR iteratively until convergence. This method employs an iteration strategy to transform contour detection into pixel classification, and combines the Bayes classifier and a constraint mechanism to solve cluttered lines and various appearances, which can inspect all kinds of EBR effectively and provide a new perspective for similar detection problems. The experimental results prove that the inspection time takes less than 20 m/s and the deviation is within 30 μm for images of 5 000×600 pixels. Compared with the existing methods, the proposed algorithm can inspect the EBR of the wafers with complex patterns and meet industrial requirements in efficiency and precision, which has been applied to the Automatic Optical Inspection (AOI) of King-semi Co., Ltd.