The Research Of Non Negative Matrix Factorization Algorithm And Its Application In Fiber Classification

Fiber classification is the most important question in the import and export inspection departments,research institutions and enterprise.Also,it is a worldwide technical problem.This subject is a part of The National Natural Science Fund "The Research of Constrained Non-Negative Matrix Factorization Algorithm and its Application in Fiber Classification "(NO.61472075).The focus of this paper is the classification of fiber by Non-Negative Matrix Factorization Algorithm.In the process of fiber image acquisition,due to various limit,such as microscope,image acquisition system,imaging system,illumination,fiber image has so much noise.A suitable data representation can find potential structures in fiber images.In this paper,a method of fiber image representation based on Non-Negative Matrix Factorization is proposed.The fiber identification method,proposed in this paper,can extract the local feature of fiber image effectively and is more robust in the feature extraction,compared with SVD and PCA.In order to improve the classification of fiber,this paper has done the following research:(1)Image preprocessing: binaryzation,wavelet transform and laplace transform.(2)Feature extraction: For traditional fiber classification methods and machine learning algorithms,it is often necessary to rely on artificial experience to help the system select the key features.In this paper,by using the Non-Negative Matrix Factorization Algorithm,the original fiber image is directly used as the input of the system,and the key features of the fiber image are extracted automatically,so as to improve the classification of the system.(3)In this paper,the Non-Negative Matrix Factorization Algorithm is used to study the sparsity of fiber data decomposition,By applying the sparsity constraint to the Non-Negative Matrix Factorization Algorithm,the better decomposition results are obtained,and the local feature of the fiber is extracted better,and the recognitioncapability of the fiber image is further improved.(4)For the sparsity constraint of Non-Negative Matrix Factorization Algorithm is difficult to determine,A sparse adaptive algorithm based on genetic algorithm is proposed,and the recognition capability of the fiber image is further improved.