Research and Application on the Improved SSD Chip Defect Inspection Algorithm

Abstract

Defect inspection is an important part of the light-emitting diode (LED) production process. In the production environment, most of the defects to be inspected are small targets below 32×32 pixels, which cannot be detected by current SSD algorithm effectively. Based on the concept of DSSD (Deconvolutional Single Shot Detector) algorithm, an improved SSD (Single Shot Detector) chip defect inspection method is proposed to accurately inspect LED chips surface defects. First, the original algorithm’s basic network VGGNet is replaced by ResNet, and the first pooling layer is removed; then a prediction module and a deconvolution module are added to fuse the semantic information between the high-level and low-level; finally, model hidden layer is pruned, and the last convolution layer and deconvolution module are removed. The experimental results show that compared with the original SSD algorithm, which enhances the accuracy of chip defect inspection.