Woven Fabric Linear Research Of Detection Based On AR Model

The computer vision technology has been more and more widely applied into textile industry with great development of computers and image processing technology nowadays. Quality detection of fabric apparent defect, especially with the method of computer vision technology has been a hot spot that attracts many scholars from different countries. As a result of labor cost increases, automation inspection has become an orientation. There are many ways to do defect detection, but a class of defects during tests is harder to be detected, all of which have following common characteristics:more slender shape, smaller size. Not too much research on detection study for this type of woven fabric defect, and the test results are not very satisfactory.First In this article defects are divided into four categories based on the appearance such as shape, size:linear defects, a large area of strip defects, patchy defects and other defects. Linear defects refers to the appearance of the slender shape of the defect type, which is continuous or intermittent reflected in the fabric. These defects occupies a certain percentage in all defect types, while the linear defect detection results are not satisfactory by using the automatic detection algorithm previously, it provides this subject a clearer target scope and a reliable basis for further study.AR model is well suitable for detecting defects of this kind According to correlational research. Firstly, fabric images of normal texture and defective texture were segmented to many child windows of same size,making the window of the fabric image to be a matrix, L(i,j), which had child windows of 8×8. Then, according to the feature that fabric texture has periodicity and orientation, a methodological method of feature extraction was studied, using which gray vale of every child windows was calculated using specific method to obtain two sequences in the light of both vertically and horizontally, and then this two sequences were connected end to end to form the third sequence. After that, three characteristics called variance characteristics, CV characteristics and range characteristics were obtained using the above method, in which algorithms of mean value, variance and range were used respectively. Linear defects were the main research object, and preliminary experimental investigation indicated that as the characteristic of linear defect detection, variance features sequence is superior in linear defect detection.To get a better test results and higher detection rate, optimize the order of AR model. First, reduce the order of a broader range of the plain, twill two different textures gradually, then the order in a more reasonable range of 3 to 6 bands, respectively, by testing the order of the, tests showed that order 4 AR model is suit to two texture optimization. with Burg algorithm and followed the normal texture and texture defects were estimated with the corresponding spectral data. Finally, find defects and their locations by comparing the correlation of spectral estimation of normal and abnormal texture, and mark them in the image of the original fabric.Finally, this thesis finish a certain amount of experiments and verificate respectively the linear defects of warp direction, linear defects of direction of weft and non-linear defects, and this method are effective not only for good linear defect detection, but also for non- linear defects.