Research On Automated Fabric Defect Detection Using Independent Component Analysis

Automatic Fabric defects detection is an important research area in which computer vision is used to solve the products quality inspection problem of textile and garment during production. This technology makes comprehensive use of digital signal processing, image recognition, computer intelligent and other multi-disciplinary knowledge, so it has high theoretical research value. The fabric defect detection, based on computer vision technology, which is of high practical value, can improve the degree of enterprise automation, reduce the low efficiency, high miss rate, affected by environmental factors and inspector’s subjective factors which caused by artificial detection, thereby improving economic efficiency and competitiveness of the enterprise.It is because that type of fabric defect is various and defect detection algorithm is greatly influenced by many factors, such as the environment, illumination and others, there have been various limitations of the automatic detection algorithm. This paper proposed a global optical ICA automatic algorithm which is based on independent component analysis(ICA) method and combined with swarm intelligence optimization technique. Based on this new algorithm, the fabric defect detection scheme is designed.Independent component analysis method is widely used in the field of signal processing. It has the blind signal separation processing capacity for the source signals that only depends on the observed signal. The estimated signals which are independent each other, original, mixed by unknown factors can be estimated from the observed signal through the high order statistics of the data. As the feature vector of fabric defects can be mainly reflected by higher order statistic vectors of the image data, the fabric defect detection algorithm which is based on ICA model can extract hidden fabric defect feature information in the background of fabric texture. But the typical ICA algorithms have some defects, such as easily falling into local extreme point, therefore, the swarm intelligence optimization algorithm is employed for the global optimization of ICA iterative algorithm in this topic.The main results of the work are as following:First of all, the performance of four kinds of classic ICA algorithm are studied from two aspects—signal processing and image processing. The simulation experiments are measured by run time, PI value and kurtosis value before and after the separation under the MATLAB software. The experimental results demonstrate that Fast ICA algorithm shows excellent performance in signal and image processing. It also confirmed that ICA had technical feasibility in fabric defect detection.Secondly, since the ICA algorithm itself has a few shortcomings, we apply particle swarm algorithm to improve it, then put forward one of global optimization ICA algorithm. The problem of local extremum points was solved, the quick convergence and global optimization of the ICA algorithm were realized.Thirdly, a kind of fabric defect detection scheme is designed which is based on the global optimization ICA algorithm, and the several common defect types of detection are realized. The fabric image is preprocessed at first, then the ICA global optimization algorithm is used to get the global optimal de-mixing matrix. Finally, convolution algorithm and threshold segmentation are given to complete the defect detection. An analogue simulation research on some parameters is conducted. It is concluded that the independence criterion of the ICA algorithm can be used as evaluation function to search for the essential characteristics of the fabric, which used for defect detection. The testing results of the scheme is greatly influenced by control constant and the size of window and is not influenced by the number of the window too much.Finally, based on the MATLAB software development platform, the visual interface of fabric defect detection system is designed. The users can preprocess the image, show the detection of fabric defect area after determining a testing fabric image in this interface. And fabric defect detection effect can be observed from different control constant, the number of windows, window size and the fitness function.