Research On Defect Detection For Flexographic Woven Labels Based On Machine Vision

In recent years,with rapid development of the apparel and textile industries,the surface quality inspection of the textile is particularly critical for mass production.As the important accessory of the textile,woven label describes the proportions of ingredients,the methods of storage and washing,etc.So,its quality inspection has become most important.However,due to its diversity,defect detection for flexographic woven labels in the industry mainly depends on human eyes without any automatic optical inspection devices.Therefore,visual inspection for the defects of flexographic woven labels is of great significance for the real industry.To solve the problem of defect detection for flexographic woven labels,this thesis designs a machine vision system to detect the defects in flexographic woven labels.The machine vision system consists of two parts.One is an image acquisition hardware system for acquiring flexographic woven labels,which is composed of light sources,a camera,a lens and the mechanical device.The other is a software system involving defect detection algorithms for flexographic woven labels.The main work of this thesis is described as follows.(1)Three kinds of image features are extracted from flexographic woven labels and then are combined together.Grayscale features are extracted by grayscale processing performed on flexographic woven label image.Canny operator is used for flexographic woven labels to obtain their gradient features.Entropy features are achieved by calculating the generalized entropy of flexographic woven label image.These extracted image features are combined together to form a sample function and construct a sample image with rich feature information.(2)The average eigenvector of the flexographic woven labels is calculated as a candidate eigenvector.Then,a priori knowledge of the flexographic woven label is constructed by comparing the reference eigenvector with the candidate eigenvector.Then,the prior knowledge is added into the objective function to construct the least squares regression model.(3)The augmented Lagrange multiplier method based on singular value decomposition is applied to iteratively solve the least squares regression model.The tested image of the flexographic woven label is decomposed into a foreground containing defects and a background without defects,and then the defects are highlighted by the differences between the sample image and the corresponding template image.The inspection of flexographic woven labels is implemented based on iteratively thresholding.Experimental results show that the proposed inspection method for flexographic woven labels can achieve the inspection performance with 0 omission rate and 1.7% error rate,which is superior to other existing detection algorithms.Also,it demonstrates a good perspective for engineering applications.