Woven Fabric Defect Detection Based On SVD

In recent years, with the increase of labor costs, People are increasingly concerned about production automation. Introduction of information technology in the textile production process, not only can reduce labor costs, but also improve production efficiency. In the textile production process, woven fabric defect detection is a very important step. Therefore woven fabric defects effectively detect significance for improving textile quality, lower costs.However, the existing algorithms for different machine fabric texture poor adaptability and real-time, difficult to adapt to the actual production needs. We propose a woven fabric defect detection algorithms which based on singular value decomposition (SVD) of woven fabric defect detection. Singular value decomposition in image processing has the following advantages:(1) Singular values of the image with good stability;(2) Singular value decomposition algorithm can reduce the computational complexity while can extracted the main image features. Therefore, when dealing with images, can achieve faster speed and higher accuracy.Firstly, the normal fabric samples for vertical and horizontal image sub-window operation direction of projection. Then combining into a sequence of two projectors projection sequence and then wait until all sub-joint projection sequence corresponding to the detection window composed of a matrix of fabric image and the matrix singular value decomposition, extraction of basis vectors. Finally, getting the base vector treated fabric samples for the detection reconstructed by the reconstruction error to determine whether the defects and determine its location. In order to detect the effect of an intuitive understanding of the final paper will compare the effects of the detection algorithm based on AR model and algorithm to detect the effect. The following main topics of this research will be described.(1) The fabric sample image acquisition methods and explore how to reduce the amount of computation.The child window in the direction of vertical and horizontal pixel gray value projection, and end to end joint projection sequences. Without loss of energy in the image information of the fabric and greatly reduce the amount of computation.(2) Basis vectors obtained by the use of the singular value decomposition of the test samples to reconstruct.By singular value decomposition, texture features only reflect the normal basis vectors from the normal sample. Then treat the samples tested were reconstructed using the basis vectors. By comparing reconstruction error reconstruct the original sample to sample and determine whether the defect samples.(3) Preferably the size of the window.The window size has a great influence on the final results of the detection, the window size must be preferable. Normally select the size of the child window, you should consider the size and proportion of defects in the child window.(4) Preferably the number of basis vectorsthe number of basis vectors K reconstruction error. The more the number of basis vectors, the reconstruction error is smaller, has a great influence on the detection results. Exploration during the test, hoping to find such a number of basis vectors, in this case, the number of basis vectors, both good texture area will normally be reconstructed, at the same time without the defective part of the reconstruction of a good texture.(5) Contrast detection algorithm based on the effect of AR modelAnd testing the effect of AR model algorithm were compared the false detection rate, undetected rate and real-time indicators. To detect the effect of the proposed algorithm has a more intuitive understanding.After the6991samples’ experiment, SVD algorithm enables the false detection rate of less than10%, and detection rate greater than90%. Comparing with AR model algorithm, SVD algorithm is better on the detection accuracy and real-time algorithm.