Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Woven Fabric Linear Research of Detection Based on AR Model

Author ZhuJunLing
Tutor WangJun
School Donghua University
Course Digital Textile Engineering
Keywords Woven fabric defect detection Linear defects AR model Burg algorithm
CLC TP391.41
Type Master's thesis
Year 2012
Downloads 17
Quotes 0
Download Dissertation

With the computer and the development of image processing technology in today's society, the application of computer vision technology in the textile industry is also more extensive. Quality testing is especially apparent in the textile fabric defect detection of computer vision has been the hot spot of the research of scholars from various countries. With the increase in labor costs, automated detection has become a kind of trend. Defect detection in many ways, but during a test, a class of defects more difficult to detect, and they have certain similarities: the appearance of a more slender shape, small size. Study is relatively small for this type of machine fabric defect detection, and detection ineffective. Firstly, according to the detection method defects in accordance with the appearance of the shape, size is divided into four categories: linear defect, large area strip defects plaque defects and other defects. Linear defect refers to the appearance and shape of the elongated type defect, they reflect on the fabric is somewhat continuous, some discontinuity. And occupy a certain proportion of its weaving defect, the linear defect detection is less than ideal but in the automatic detection algorithm. The present paper provides a clearer target range. Related studies have shown that the AR model is more suitable for the detection of such defects. Will first obtain the normal texture and contains defects texture fabric image segmentation, the size of the sub-window according to a certain window divided to form a matrix L (i, j), (8 × 8 sub-window). Then use the fabric texture with the periodicity and orientation of this feature, a feature extraction method obtained through research: first image gray value of each sub-window within the specific operator in accordance with the longitudinal and transverse directions, respectively, to obtain row and column two sequences of two sequences end to end, and then put together to constitute a third sequence. According to this approach computing the mean, variance, range algorithm after a series of operations, the three characteristics: the variance feature, grayscale CV values ??characteristics, poor characteristics. Linear defect as the main object of study, a preliminary pilot study showed that the variance characteristic sequence as a linear defect detection characteristics superior. In order to obtain a better detection effect and higher detection rate, the preferred order of the AR model. First through the two kinds of plain, twill different textures in a relatively wide range of the order number range is gradually reduced, and then through the tests were in the range of more reasonable order 3-6 Order of two textures order optimization, The final test showed that the 4-order AR model order as both texture superior Secondly, based on the Burg algorithm of normal texture and estimated were obtained with a defect texture corresponding spectral data. Finally, the correlation between the detection of the spectral estimation of the comparison of normal and non-normal texture to a defect and its position, and the detected defect location marker in the image of the original fabric. Finally warp linear defects, zonal linear defect as well as non-linear defect test validation, this method of linear defects better detection effect, also seen for non-linear Defects also effective.

Related Dissertations
More Dissertations