Abstract: Aiming at the current manual detection of mobile phone battery surface quality, a software program for non-destructive testing system for battery surface defects is designed. First, the surface of the battery is subjected to pre-processing operations such as tilt correction, ROI(Region Of Interest) extraction and character gray value modification. The adaptive threshold luminance based on gray density distribution and gray level difference is proposed to perform traversal of the sub-images of the ROI. The defective sub-images of the coincident regions are merged and the regions without obvious defects are filtered. Then, SVM(support vector machine) multi-class classification method is used to extract the binary image pixel distribution regularities as training feature and identify the battery surface defect types. Finally, the visual interface of the software is developed to determine the optimal variable parameters of the scheme. The recognition rate is as high as 95% by the experiment.