It will inevitably produce a number of defective goods in the production process of button batteries, so it is necessary to rely on information technology to design a set of reasonable algorithm to automatically detect the unqualified products. We use image acquisition equipment to get the button cell surface image, and then to filter mixture noise, OSTU optimal threshold segmentation, character of image segmentation, extraction of defect mode, BP neural network classification of defects.A measure of standard of the algorithm in each step is simple and efficient, so it can lay the foundation for the algorithm to do real-time detection in production practices under meeting the detect precision in the premise condition.