Research On Automatic Defect Detection Of Tires Based On Semantic Segmentation And Low Rank Recovery

Tires are important components of automobiles,and they have the important role of bearing weight,transmitting traction and braking torque.Therefore,they have high requirements on the bearing capacity,cushioning capacity,wear resistance and durability of automobile tires.The quality of tires has an important impact on the safety of vehicle operation.The quality inspection of tire products should attract sufficient attention.Currently,visual inspection has been widely used to diagnose defects in tire products,but the method of visual diagnosis of tire defects is still incomplete.Because there are many defect categories and anisotropic textures in tires,automatic tire visual defect detection is particularly challenging.This paper analyzes the cord texture distribution and low rank characteristics of tire X-ray images,and proposes an automatic tire X-ray defect detection system based on semantic segmentation and low-rank matrix recovery.The texture of the tire X-ray image is anisotropic but the texture partition is obvious.Segmentation of the texture area is helpful to enhance the low-rank characteristics of the image,reduce the difficulty of defect detection,and improve the detection accuracy of the algorithm.In this paper,the classic semantic segmentation network is trained on the pre-processed tire X-ray image data set,and the neural network results are optimized to obtain a high-precision automatic texture region segmentation algorithm.The low-rank matrix restoration algorithm uses the low-rank subspace decomposition principle of the matrix to decompose the tire X-ray image matrix into a low-rank background matrix and a sparse matrix containing defects.Texture segmentation is conducive to improving the quality and computational efficiency of the low-rank decomposition algorithm.The segmented image blocks have strong low-rank characteristics.We researched and optimized the traditional Inexact Augmented Lagrangian multiplier method(IALM)for solving Robust Principal Component Analysis(RPCA),so that the sparse matrix obtained by image decomposition has less noise.Experiments have been conducted to test how the algorithm parameters will affect the decomposition accuracy and convergence speed of the algorithm,and select the best parameter value.The sparse matrix contains a large number of noise points and possible defects.In this paper,the postnoise reduction processing of the sparse matrix image is designed to eliminate the noise points and retain a more complete defect mask.Defect masks are the basis for detecting and locating defects and provide a reference for tire quality ratings.Experiments show that the method has high defect detection accuracy,low false alarm rate and real-time performance.A comparison experiment with existing tire X-ray detection methods was performed on the same data set.The experimental results validate the theoretical analysis and show that the method in this paper can perform effective automatic tire X-ray image defect detection with high accuracy and is robust to different tire styles and defects.In addition,the method can output a more complete defect mask for estimating defect area,defect location and tire rating.