Research On Tire X-ray Image Defects Detection Algorithms

As one of the most important parts of motor vehicles,tires directly affect the driving safety of motor vehicles.In order to ensure road traffic safety,tire manufacturers will conduct comprehensive quality inspections on tires entering the market.At present,most domestic tire manufacturers still use manual methods for tire quality inspection.In order to further improve production efficiency and save production costs,some tire defects detection algorithms based on image processing and deep learning have been applied in the actual production process.How to effectively combine and improve the existing detection algorithms to improve the accuracy rate is of great significance to reduce the production cost of tire manufacturers,improve tire quality and provide better road traffic safety guarantees.Based on the methods of image processing and deep learning,this thesis carry out research on tire X-ray image defects detection algorithms,and the specific research content is as follows:1)Aiming at the unique texture features of tire X-ray images,the preprocessing algorithms and feature extraction algorithms are improved.The preprocessing method mainly includes two image processing algorithms:In view of the unique structural distribution of tire X-ray images,an image area division algorithm based on threading method is proposed;Aiming at the grayscale distribution of tire X-ray images,a binarization algorithm based on grayscale adjustment is proposed.Aiming at the two most common types of defects in tires:impurities and thin lines,two dual-threshold feature extraction algorithms based on column comparison are proposed.Through further analysis,it is found that these two feature extraction algorithms are highly similar in nature and can detect tire images at the same time,which can save inspection time.Experiments show that the defects detection algorithms proposed by this thesis has detection accuracy rate and recall rate of more than 90%on the corresponding tire defect types in the data set provided by tire manufacturers.2)Aiming at the problem of low test accuracy of tire defects detection algorithms based on deep learning in small samples of large-resolution data sets,a frequency domain-based tire defect location algorithm is proposed to better integrate image processing methods and deep learning and realize automated tire defects detection.The tire X-ray image is a very regular image,and the existence of defects destroys its texture structure.If the image is converted from the spatial domain to the frequency domain,the defect area can be well located.After the defect is located,the neural network is used for further classification,and the defect type detection can be realized.Experiments show that this method has an accuracy rate of 96.5%and a recall rate of 99.0%on the tire X-ray image data set provided by the tire manufacturer.Compared with the defect detection algorithm based on image processing,the classification accuracy rate is greater.In summary,this thesis conducts an in-depth study of the key technologies in the automatic detection of tire defects from the two perspectives of image processing and deep learning.The proposed method has been experimentally verified in the actual production data set,and achieved high accuracy rate and recall rate.