Small sample chip appearance defect detection method and system

The invention discloses a small sample chip appearance defect detection method and system, and the method comprises the steps: firstly building an original data set, and then amplifying the original data set by using WGAN-GP to obtain an amplified data set; a chip appearance defect detection model is established, the chip appearance defect detection model is a YOLOv7 neural network model, and a CA attention mechanism is embedded in the chip appearance defect detection model; detecting a chip picture by using the trained chip appearance defect detection model, and outputting a picture with defects; performing laser marking on the corresponding chips according to the positions of the pictures with the defects to obtain the chips with the appearance defects; according to the method, an accurate amplification data set can be efficiently obtained, and small sample data enhancement is realized; a CA-YOLOv7 neural network model is established, and elastic network regularization is used for training, so that the meth