Chip Surface Defect Detection Algorithm Based on Improved YOLOv3

Abstract

Chip package testing is a key process to eliminate bad products in the electronics industry. Aiming at the problems of YOLOv3 chip detection in complex environments with low accuracy and large number of model parameters, an improved YOLOv3 automatic detection method based on EMO, called EMO-YOLOV3, was proposed. This method uses EMO to replace the backbone of Darknet53 in YOLOv3, inherits the efficiency of CNN to model short-range dependencies and the dynamic modeling capability of Transformer to learn long-distance interactions. The results show that the model has a very good detection effect on 7 kinds of environment, such as character defect, scratch defect and braid damage. Compared with the original YOLOv3 model, the mAP of chip surface defect detection is improved by about 3.4% while the number of model parameters is reduced. Therefore, it is considered that this method can be used for real-time automatic detection of chip surface defects.