Research And Development Of Fabric Defect Detection

Textile industry is the traditional mainstay of the national economy. Its productivity scalecontinues to expand, whereas the defects of fabric seriously affect its quality. There arealready some advanced detection systems overseas, yet many small and medium-sizedChinese companies still use manual inspection, with a low rate of efficiency and high rate offalse and missing detection, so it’s urgent to develop the fabric detection system which canpractically meet the actual needs of the industrial field. In this paper, considering the fabricimage and its texture features, the technology of fabric defect detection is intensivelyresearched.Aiming at the difficulty of the background interfering when detecting the fabric, thispaper studied the technology of feature-enhancing and solved the problem of inaccuracy inextracting the fabric texture parameters. Two accurate algorithms were proposed. The formerbased on the polar coordinate transformation of Fourier spectrum, through the Fouriertransform of the image to get the direction, then with the approach of polar coordinatetransformation to separate the angle of direction; and the latter based on DMF, first measurethe DMFs of row and column, and then design a distance algorithm to sort the weight of everyextremum so as to eliminate the impact of interference points, thus extracting the period oftexture cycle with accuracy.An algorithm was put forward based on an improved version of the Gabor to extractfeatures. The characteristics of Gabor filters and a practical two-directional Gabor filter wereintensively studied; this algorithm takes the direction and periods of fabric texture as thedirection and size of the Gabor filter, effectively highlighting the texture features. The featuresand parameters of GLCM were explicitly discussed and experimented, the GLCM was used toextract features, which were processed and used to construct categorizer.As for the difficulty of pattern fabric detecting, an algorithm based on the offset series ofelement was proposed. Firstly the precise periods of element were used to extract a flawlesselement; the offset series of this element were constructed to analyze the difference betweenthe flawless fabric block and the fabric block to be detected. Then a double classificationmechanism which not only can meet the requirement of accurate detection but also canimprove detection speed was proposed. Finally, a comparative analysis of the test images wasmade.Experiments show that the algorithm based on the improved Gabor filter and the GLCMhas a better result, with a98%detection rate, a zero false alarm rate, and a200ms detectingtime. With the above data satisfied, the algorithm could be used to realize the detection ofmesh defects in the process of knitting. The algorithm based on the offset series of elementand double classification mechanism has high detection rate and low false alarm rate; theperformance of the algorithm is completely prior to the area subtraction and the motif-basedalgorithm, which can realize the pattern fabric defect detection with excellent performance.