Research On Dictionary Learning In The Classification Of Fiberglass Fabric Defects

With the rapid development of the national industrialization,the demand for fiberglass fabrics in many industries is getting increasingly great.Therefore,it is increasingly important to control the quality of fiberglass fabrics.In the quality inspection of fiberglass fabrics,defect classification plays an important role.The study of accurate and reliable classification methods has important practical significance for the quality control of fiberglass fabrics.The traditional defect classification method relies heavily on the selection of defect features,while sparse representation and dictionary learning have strong representation ability for image features,and are widely used in pattern recognition and computer vision.This article will delve into how to use the dictionary to learn the classification of defects in fiberglass fabrics.The main work is as follows:(1)We have analyzed and summarized the existing problems of the fabric defect classification algorithm and introduced the sparse representation and dictionary learning in detail.Firstly,the original image is sparsely represented to obtain a dictionary,which is used to reconstruct the fiberglass fabrics defect image,and the reconstructed image is compared with the original image,the result shows that the learned dictionary can automatically extract the main defect features of the image.At the same time,the feasibility of dictionary learning to classify defects in fiberglass fabrics is proved.(2)Aiming at the problem that the dictionary has high redundancy and low discrimination in the traditional sparse representation classification method,HOG+LBP is used as a feature extractor to classify defect images in combination with a dictionary learning with structured incoherence and a dictionary learning model based on Fisher Discrimination,respectively.Firstly,the HOG+LBP feature extractor is used to preprocess and reduce the dimension of the defect image to obtain a feature matrix of each type of training images.Then,the dictionary learning model is used to learn a specific class dictionary for the feature matrix of each type of image,enhance the discriminability between different types of dictionaries;finally,the specific class dictionary is used to reconstruct the test images,and classify the test images according to the minimum method of reconstruction error.The experimental results show that the proposed two classification methods can effectively classify fiberglass fabrics defects.(3)A statistical classification algorithm for multiaxial fiberglass fabric defects based on dictionary learning is introduced.Firstly,the original defect image is preprocessed by the MR8 filter bank and a filter response matrix is constructed for each type of defect image.Secondly,the contribution rate of each image reconstruction by statistical dictionary atom is established.Two histogram features of the image are obtained: a coefficient histogram and a residual histogram.finally,combining the two histogram features,the classification is completed by the sparse representation classification method.finally,combining the two histogram features,the classification is completed by the sparse representation classification method.There are 17 figures,10 tables and 59 references in this paper.