Research On Defect Detection For Woven Fabric Based On Sparse Representation And Dictionary Learning

Effective detection and control of textile surface defects is one of the key links to control costs and improving product competitiveness in modern textile enterprises.With the rapid development of artificial intelligence and computer science,computer vision has been more and more widely used in the field of industrial surface inspection.Instead of artificial vision by computer can not only improve the speed of detection,computer instead of artificial vision can not only improve the detection reduce labor costs,but also provide credible reference data for the evaluation of the fabric quality grade through the fabric defect automatic detection system.Therefore,there is very import significance to develop an efficient,reliable and accurate automatic fabric defect detection algorithm.In recent years,with the rapid development of compressed sensing theory,sparse representation and dictionary learning have been widely used in computer vision and pattern recognition.Based on sparse representation and dictionary learning theory,this paper designs two approximate representation models of woven fabric defect images,and the two model are used to realize the design of woven fabric defect detection algorithm.The specific research content and related work are summarized as follows:1)The problems existing in the detection of woven fabric defects are analyzied,and the existing flaws in the existing detection algorithms are summarized based on the feature extraction method and the non-feature extraction method.After that,according to the sparse representation and dictionary learning methods,the approximate representation model of the woven fabric image is designed in this paper.Finally,it is feasible to obtain an approximate expression method through experimental analysis,and provide a basis for further detection of woven fabric.2)The overall flow of woven fabric defect detection algorithm based on sparse representation is explained.First,the histogram equalization pretreatment is performed on the defect image.Then,the defect image is decomposed according to the objective function of the approximate sparse representation model to obtain the decomposed defect components.Finally,the decomposed defect components are added to the threshold decomposition to achieve the purpose of defect detection.The experimental results show that this method has short detection time and high efficiency,and the average detection rate can reach 96.8% for fabric defect images including star pattern,square pattern,dot type,fine grid and stripe,and net color.3)A dictionary-based learning defect detection algorithm for woven fabrics is introduced.The detection algorithm flow step is: First,the image is subjected to preprocessed by gray-scale and mean-filtering operations.Then the normal image joint matrix is used for dictionary learning,and the defect dictionary joint matrix is reconstructed by using a learning dictionary.Finally the reconstruction error is solved.The division mark of the segmentation threshold on the defect area is calculate to achieve the purpose of defect detection.Experiments show that this algorithm for defect detection can not only solve the problem of simple texture fabric defect detection has good fabric surface consistency,but also have good detection effect for complex texture fabric images.Through quantitative analysis of the results,we can see that the detection rate of the dictionary learning algorithm for simple textures and complex textures defects reaches 97.41% and 97.18%,respectively,and the detection effect is ideal.