When using traditional automatic optical inspection to detect plastic defects, it is difficult to quantify the defects due to the variety and changing positions of the defects. It is easy to encounter the problem of insufficient defect samples, which makes it difficult to quantify the defects. This leads to insufficient detection accuracy. If manual inspection is maintained, the detection speed is relatively slow and the quality is inconsistent. There are still many difficulties in identification.By using SolVision's Segmentation technology, a defect database is established for the shape and color of rubber product defects, and then AI is used to learn the characteristics of the defects. This can identify defects of various types and positions. With the increase of learning images, the ability of AI visual inspection is continuously optimized, which significantly improves the accuracy of rubber defect identification and effectively solves the problem of unstable detection of rubber product defects.