Tire flaw detection method and model based on self-attention mechanism and dual-field self-adaption

Abstract

The invention provides a tire flaw detection method and model based on a self-attention mechanism and dual-field self-adaption, and the method comprises the steps: cutting a tire X-ray image in a model training stage, and inputting the cut tire X-ray image into a tire flaw detection feature extraction module composed of ResNet50 and a VIT model; carrying out marginal probability distribution adaptation on the obtained source domain features and target domain features by using MK-MMD, and carrying out conditional probability distribution adaptation on the same category of samples in the source domain and the target domain by using LMMD; and a defect detection result is obtained through a defect classifier subsequently, and the model is trained in a back propagation mode. In the flaw detection stage, after a tire X-ray image is cut, the tire X-ray image is directly input into the tire flaw detection feature extraction module obtained in the model training stage; and inputting the obtained features into a defect classifier obtained in the model training stage to output a defect detection result. According to the invention, domain offset between images sampled by different X-ray machines can be overcome, and the method has good performance in tire flaw detection with deficient data.

Description

technical field [0001] The invention relates to the field of artificial intelligence transfer learning, in particular to a tire defect detection method based on a self-attention mechanism and dual domain self-adaptation. Background technique [0002] In the tire manufacturing process, due to the influence of the external environment such as the production process, the finished tire may contain many kinds of defects, which has a serious impact on the quality of the tire. Due to the data imbalance of defect samples and the uncertainty of the location and probability of defect occurrence, there is no perfect automatic defect detection technology at present. Most factories also mainly inspect tire X-ray images for defect detection by quality inspectors with naked eyes. This method is time-consuming and subjective. In addition, the background texture of the tire X-ray image is complex, and various types of defects are mixed in the background texture, making it difficult to disti...

AI Technical Summary

Problems solved by technology

At this time, it is necessary to re-collect defective samples for labeling and model training, which is time-consuming and cumbersome

The above phenomenon will reduce the flaw detection performance of the deep learning model trained on the original data (source domain) on the new data (target domain), resulting in the situation that the model cannot be reused

In addition, tire X-ray images have undesirable features such as low contrast and low brightness. The scale of tire defects varies greatly, the location of appearance is not fixed, and they are easily disturbed by background noise. These features bring challenges to tire defect detection.

Images