Fabric Defect Detection And Intelligent Recognition Based On Vision Inspection Technology

Textile plays an irreplaceable role in people's daily life,national defense and medical field.Fabric defect appearing in weaving process damages appearance and quality of products and failing to detect defect(or yarn break)in time will causes a big waste of raw materials.This research mainly focuses on the study of digital image processing technology,machine learning and their applications in fabric defect detection.Key point in the development of detection algorithm is to enhance its versatility and improve its precision.A fabric defect detection system was developed and its applicability in textile industry was explored.The main research work is as follows:The sparseness of yarn distribution of eyelet fabric makes the existing detection algorithms,which are mainly designed for conventional fabric,have a poor performance on eyelet fabric defect detection.In order to solve this problem,an adaptive image segmentation method suited to eyelet fabric detection was proposed.After spectrum analysis was carried out on the raw image,a specific frequency domain filter was designed to weaken the intensity of normal texture background and made the feature of the defective region become more distinct.Subsequently,gray level distribution of the filtered image was analyzed to calculate image segmentation parameters which were to be used in region growing method.Finally,considering the potential existence of pseudo defects caused by segmentation algorithm,morphological operations were used to obtain the ultimate defective region by filtering pseudo defects.The algorithm makes good use of the advantages of both of frequency domain filtering and region growing method.The combination of them makes contour of extracted region relatively clearer and more accurate and the algorithm is immune to illumination changes.In view of the fact that there is little research on the detection of eyelet fabric,the algorithm proposed in the present work has a certain reference value for the future research on defect detection of fabric with sparse texture.In order to detect various kinds and forms of plain weave fabric defects,an algorithm based on support vector machine(SVM)was investigated in this work.Compared with traditional machine learning methods,support vector machine,which is based on statistical learning theory,has a better learning ability in the situation of nonlinear,small samples and high dimension.At the beginning,due to the difficulty in obtaining enough representative defective samples during production,abnormal detection method was adopted and one-class SVM was used as classifier.At this step,only normal fabric samples were used in the process of training and the algorithm has an excellent performance in testing accuracy.After enough defective samples were identified by one-class SVM,the identified samples including positive samples and negative samples were used as training set of multi-classes SVM.A new classifier was built and detection rate of it was further improved.In the study,two groups of feature vectors were determined through experiments,and each of them has its advantage in the detection of different kinds of defect.The feature space and parameters of SVM were optimized to obtain the best detection result.A method of combining the detection result of two groups of feature vectors was proposed in this part to reduce both false positive probability(FPP)and false negative probability(FNP)of the algorithm.In order to realize fabric defects detection and intelligent recognition,the design scheme of detection system including hardware,software and algorithm design was discussed to meet the demand of industrial application.It controls the movement of motors according to the real-time feedback of detection results.The development of human-computer interaction interface makes it possible for users to setup the basic parameters of camera,to select image acquisition modes and to control the speed of motor in a convenient way.The research in this thesis provides theoretical and experimental basis for fabric defect detection in textile industrial application.