Wafer Surface Defect Detection Based on Improved Autoencoder and Deep Feature Extractor
In today's semiconductor defect detection field,it always faces the problem of insufficient defect samples and diversified defect samples,in order to solve the problem effectively,a wafer surface defect detection model based on improved self-encoder and deep feature ex-tractor is proposed by using the DFR model as the basic framework,which achieves a better feature extraction by using the pre-trained VGG19 model as a feature extractor,and subsequently image reconstruction using improved self-encoder to learn the normal features of the image.The anomaly scores are obtained by comparing the global differences between the input and generated images for defect detection,and the results show that for the homemade wafer dataset,the average AUC improves by 0.8 percentage points compared to the baseline model,and the accu-racy of defect detection reaches 0.997;for the MVTec AD dataset,the average AUC improves by 2.5 percentage points compared to the base-line model,and the accuracy of defect detection reaches 0.963.
国家科技期刊平台
NSTL
万方数据
