Research On Detection And Classification Of Web Defects

Textile quality control research is an important research direction of textile automation.Fabric defects automatic detection and control has been a weak link in the field of textileautomation. Many domestic and foreign scholars in the research field of textile automationhas been quite effective. Internationally, there have been many manufacturers introduced theirown textile machine vision-based automatic defect detection system; there are somecommercial applications also. However, most of the hardware and software of these systemsis very expensive, small and medium enterprises are unbearable on the price. Nowadays, inour country, textile automatic detection system is at the stage about theoretical study of thetransition phase to the commercialization of the system, which has good potential fordevelopment. This is of great significance to the development of China’s textile automationand textile quality control. The paper is focus on the machine vision-based textile defectsdetection system.In this paper, a comprehensive analysis of the existing fabric of domestic and foreignautomatic defect detection system and auto-detection algorithm are investigated, thencomputer vision-based textile line detection system hardware platform are designed.Computer based image acquisition and processing system is the core of this platform, whichis combined with the structure of the traditional cloth inspection machine, and together withthe high-resolution linear array CCD camera for image acquisition. Fabric defect detectionalgorithm is the main focus of this research. Combination of effective fabric defect detectionalgorithm, improved and integrated, is proposed based the "joint fabric defect detection" of"method library" method. This approach integrates Gabor-Gauss method, backgroundanalysis, and multi-scale wavelet method as a defect detection method of the combined libraryof fabric defect detection. The precondition of this approach is a fast computer processingspeed and the appropriate velocity of the moving cloth. On this basis, the preliminarygraphical user interface (GUI) is designed in order to facilitate human-computer interaction.Then, the effective extraction of22features are extracted by using co-occurrence gray levelmatrix and Tamura texture feature extraction method from five typical fabric defects, andtreatment with PCA for dimensionality reduction. Finally, the feature data set is classifiedwith scaled conjugated BP neural network. Sample classification accuracy can reach96%,which is very useful for further classification research.Synthesis process and experimental results, this test fabric defect detection andclassification methods have good performance, adaptability, and meet the expected researchtask goal basically. Test results are very useful for further research of this subject.