The invention discloses a wafer defect detection method based on multi-scale corner feature extraction and belongs to the technical field of wafer defect detection. The wafer defect detection method can be used for determining wafer defects by searching a angular point with maximum response in a same image area in a scale space at each layer as a training feature vector of a neural network, so that the problem of the sensitivity to ambient light changes is avoided, and the wafer defects are effectively classified. The wafer defect detection method has the beneficial effects of high stability, high detection efficiency and low sensitivity.