Wafer defect detecting and positioning algorithm based on cascaded YOLO-GAN

The invention belongs to the field of deep learning computer machine vision and the field of semiconductor technology detection, and provides a wafer defect detection and positioning algorithm based on cascaded YOLO-GAN. In the wafer generating and manufacturing process, an original image wafer is sent into an improved YOLOv5-based wafer detection model and a BiseNet-based wafer semantic segmentation model respectively, and the position of a wafer target detection frame and a foreground mask of the wafer are obtained; the original image is input into a defect detection model based on an improved generative adversarial network, then a wafer image is reconstructed, and a wafer defect area is positioned; the position of a target detection frame of a wafer is used as a constraint, a connected domain of a defect image is analyzed, and a Softmax classifier is introduced to realize defect positioning and wafer defect subdivision. According to the invention, visual detection of wafers with different scales is realize