Research Of Defect Detection System Based Cord Fabric

Product quality is the lifeline of production enterprises, detection of defects is animportant link of textile enterprises. The traditional textile enterprises are using eyeto finish fabric defect detection, this method is boring and labor intensity, testingpersonnel is easy to be influenced by emotional factors, Generally exist errordetection rate and high miss rate, slow detection speed. So using machinevision instead of manual automatic detection technology has become a hot research athome and abroad.The study of object in this paper is cord fabric, It is a kind of periodic structuretexture cloth. The purpose of the study is to realize the fabric defect detection systembased on machine vision. The system includes hardware platform and software testingsystem. Firstly, selection of hardware to build system platform, a line scan camera,lens, image acquisition card, light source, measuring tools and industrial computer thesystem of platform mainly consists of six parts: camera, lens and light source consistsof Image acquisition system; length measuring tool provide encoder signal calculatethe length of cloth running, another important function is provide an external triggersignal for camera; Image capture card acquire image and stored in the industrialcomputer; the computer will be processing the image to obtain defect imforation andlocation.Software system includes the defect detection algorithm design and softwaremodule design. In defect detection algorithm design, mainly considering how toremove to repetitive, cyclical part of the cord fabric, and then threshold segmentationmethod to extract the region of interest, This paper describes three methods: based onGabor filter bank, based on morphological processing and based on Fourier analysisto cord fabric defect detection. and then compared with experimental data analysis todetermine the cord defects based on Fourier analysis detection method, this methodnot consider defect size and orientation, experiments show that the method canacquire higher detection accuracy rate and less undetected. Feature Selection is the important link of the defect classification algorithm, thispaper focuses on the structural characteristics and the characteristics of pixels toresearch, select the average gray, region of interest’s minimum bounding rectangleof length, width, length width ratio, area and angle as BP neural network input toclassify the defect, Recognition rate of over90%.