Printed Surface Defect Detection Model Based on Positive Samples

For a long time, the detection and extraction of printed surface defects has been hot issue in print industry. Nowadays, defect large number products still relies on traditional image processing algorithms such as scale invariant feature transform (SIFT) oriented fast rotated brief (ORB), researchers need to design for specific products. At present, based object have applied but lots labeling samples with defects. Besides, there are many kinds surface, so it is difficult enumerate all Most unsupervised learning positive use generative adversarial networks (GAN) variational auto-encoders (VAE) algorithms, these methods not effective complex surface. Aiming at problems, In this paper, an algorithm proposed innovatively. We propose kind network matching network. This divided into full convolution points extraction, graph attention using self cross attention. Though key network, we can get robustness images, solve problem deviation because different camera positions influence production lines. Just one sample needed benchmark detect The paper proved “The First ZhengTu Cup Campus Machine Vision AI Competition” got excellent results finals. working company apply production.