Hybrid Wafer Defect Classification Based on Channel Attention Mechanism and Multi-Path Depth Convolution
Accurate classification of wafer map defects plays a crucial role in improving manufacturing processes.Compared with single defects,mixed defects have complex characteristics and various types,which are more consistent with real industrial manufacturing conditions.In order to effectively identify and classify hybrid defects,a method combining channel attention mecha-nism and multi-path deep convolutional neural network is proposed.This method adds a channel attention mechanism to the multi-path deep convolutional neural network branch to focus on the detailed features of the hybrid wafer map.Experimental results on a real data set of 38 types of defects show that the model is superior to some existing deep learning models in terms of accuracy,with an average accuracy rate of up to 97.67%,and can effectively classify mixed defects in wafer images.
computer visionwafer defect recognitiondeep learningchannel attention mechanismmulti-path deep con-volution
万方数据
维普
