Simple Semantic Segmentation Based The Study Of Internal Tire Defect Detection

According to the World Health Organization,40% of traffic accidents are caused by tire quality problems.Therefore,the quality inspection of tires before they leave the factory is crucial,which is directly related to the safety of drivers’ lives.In the tire industry,many factories still use the manual visual quality inspection method.This method is inefficient,subjective,labor-intensive,and has a high rate of missed inspections,which cannot meet the requirements of tire production automation.With the rapid development of deep learning and machine vision,researchers at home and abroad have proposed a series of tire defect automatic detection algorithms,but the low visual quality of tire X-ray images and tire laser scatter images,the small scale of defects and the large internal variation of defects make these algorithms have a series of problems such as low accuracy of tire defect recognition,relatively single adaptability,and poor compatibility of algorithms.In this paper,we use tire X-ray images and tire laser scatter images as the research objects to solve the classification,recognition and localization of tire defects based on deep neural networks and semantic segmentation methods,and confirm the feasibility of the proposed method through experiments.The main research contents and contributions of this paper are as follows.(1)Two kinds of tire datasets are self-made: tire X-ray image dataset and tire laser scattering image dataset.Among them,the tire X-ray image dataset includes impurity defects,air bubble defects,sidewall cracking defects and sidewall overlap defects.In addition,the tire laser scattergram dataset has a total of 2000 pieces,mainly divided into two categories of tire defects: normal and abnormal(air bubble defects).The data was expanded to 16,000 by the data augmentation method.(2)A tire laser scattergram classification network(CA-Res Net50)is proposed for the problem of low accuracy of tire laser scattergram recognition.The improved residual network model proposed in this paper is compared with the current commonly used classification network model on the same data set,and the experimental results prove that the testing accuracy of the proposed network model for tire laser scatter map is higher than other networks,and the recognition accuracy can reach 99.7%.(3)To solve the problem that it is difficult to accurately locate tire defects due to the anisotropy,complex multi-grain rationality and defect diversity of tire X-ray images,a tire X-ray image segmentation and localization model based on simple semantic segmentation(ESSNet)is proposed.The experimental results show that the model can accurately locate and segment the defects in tire images and achieve 86.24% m PA score on the test set.