Fabric Defect Detection Technology Automatically Based OpenCV

Fabric defect detection is a key step in the quality control of textile products. At the present time, the defect detection relies chiefly on artificial visual inspection, which has a lower defection efficiency and accuracy rate (about fifty percent) due to the influences of subjective and objective factors. Therefore, it is very important to develop a quick and accurate automatic system for detecting fabric defects. In this thesis, the open source computer vision (in short OpenCV), wavelet analysis theory, and BP neural network have been synthetically applied to study the technology of fabric defect detection based on the machine vision.First of all, OpenCV’s library function PyrDown (scaling function) and Canny (edge defection function) have been applied to detect the boundary of defects automatically and to obtain the defect areas, which improves the detection rate of common defects effectively.Secondly, by analyzing existing methods for extracting texture characteristics, a method to decompose the image into warp sub-image and weft sub-image is presented on the basis of wavelet analysis&Mallat algorithm, and the self-correlation function is used to split the windows of warp sub-image and weft sub-image into many sub-windows. According to the characteristics of the fabric texture, sub-window’s energy, variance, range, entropy and inverse are chosen as the foundation for defect recognition and classification.Thirdly, aiming at the technical problems to automatically identify kinds of defects, by analyzing the principles of the artificial neural network and the structural designing of BP neural network, an optimal BP neural network has been designed, which contains how to determine the neuron number of input-layer, output-layer and hidden-layer, choosing the transfer function and valuing the learning rate.Finally, on the basis of theoretical analysis and method research, a set of automatic defect detection software has been developed based on Visual C++6.0software development environment and OpenCV open vision, which includes such several modules as image acquisition, image preprocessing, defect detection, wavelet decomposition, feature extraction and defect classification. The detecting experiments on scraped thread, roving breakage, anti-wire, rust, smudge, loom fly, color contamination, long strip defect, white streak and buckle-off monochromatic tabby cloth have been done, and the results have verified the effectiveness and practical value of the method proposed here.