Self-supervised 3D industrial defect detection method based on reconstruction mode

The invention discloses a self-supervised 3D industrial defect detection method based on a reconstruction mode. The method comprises the following steps: S1, generating an abnormal sample from normal samples by using an abnormal simulation strategy; s2, inputting a defect image generated by simulation into a reconstruction network to obtain a reconstructed image, and calculating SSIM Loss and L2Loss of the normal image and the reconstructed image; s3, splicing the input image and the reconstructed image, inputting the spliced input image and the reconstructed image into a discrimination network to obtain a defect result, and calculating Focal Loss; s4, adding the SSIM Loss and the L2Loss in the step S2 and the Focal Loss in the step S3 to obtain a final loss L; and S5, optimizing the network by using an optimizer to reduce the loss L. According to the method, a self-supervision method based on a reconstruction network is adopted, supervised learning is carried out through automatically generated labels, the n