Research On Fabric Defect Detection And Classification Method Based On Unsupervised Segmentation And ELM

As the earliest category of electronic commerce,clothing has become the largest and mature industry.Compared with the traditional model,online clothing sales have greater market value and economic benefits.However,under the Internet environment,the high return rate due to the quality of the clothing fabric not only brings losses to the consumers,but also impairs the reputation of the seller,which restricts the further development.Currently,there are two main reasons affecting the quality of clothing fabrics.Firstly,due to the immaturity of production techniques or yarn quality,many defects are generated in the fabric production process,such as hole,broken warp,broking weft,etc.Secondly,fabric is the basis of clothing.Defects such as crease,scratches,shearing,crease,dirt may appear in the fabric of the garment due to the sewing machine and human misoperation during the process of garment manufacturing.Thus,exploring the problem of ensuring the integrity of fabrics,this paper studies the detection and classification methods of fabric defects based on unsupervised segmentation and ELM,and achieves the goal of improving the quality of clothing and clothing sales,and targets practical application of the common seven types of woven fabric defect types.To resolve the problem of unsatisfied accuracy of fabric defect image segmentation,a novel unsupervised segmentation algorithm is proposed based on local patch approximation and dictionary learning.The algorithm can be divided into three steps: patch extraction,dictionary learning,and defect segmentation.Firstly,patch extraction from the grayscale image after grayscale processing in user input the original fabric defect image.Then,abnormal patches for dictionary learning were eliminated.Finally,a isomeric map is constructed based on the approximate difference between the defect patch and the normal patch,which can segment the defect area of input fabric image.Experimental results show that this method can improve the accuracy,efficiency and precision of segmentation for fabric defects,and can avoid the shortage of practical application.To resolve the problem of unsatisfied accuracy of fabric defect image classification,a novel fabric defect detection and classification method was presented based on unsupervised segmentation and ELM.Firstly,gray-level images of the original defect image input are processed for unsupervised segmentation.Secondly,the shape and texture features were extracted by using the image of the defective fabric and the image of the gray defect.Then,eigenvectors and labels of the fabric defects to be classified as the training set were used to train the ELM classifier.Finally,the trained ELM classifier and the Bayesian probability fusion method are used to classify the imported fabric defect images.Experimental results show that this method improves the accuracy and efficiency of fabric defect image classification,and can meet the needs of practical applications.