Compressive Sensing Based Classification Of Macerals Of Inertinite In Coal

As one of the important components of the coal microstructures,the composition of inertinite in coal petrology based blending directly influences the process properties of the mixed coal.So,classification and recognition of macerals of inertinite is of significant for the efficient and comprehensive utilization of coal.According to the image feature of inertinite of coal maceral,this paper firstly extracts coefficients of each scale layer from curvelet transform as the initial feature set;then reduces the dimension of sparse high frequency coefficient according to the thought of compressive sensing algorithm;and the high frequency coefficients after dimensionality reduction is combined with those of the reserved Coarse layer low to build the final feature set;finally,a combined SVM based classifier is constructed to verify the validity of dimensionality reduction method and the final feature set,and the results are satisfactory.The major work of this dissertation is as follows:(1)After reading and consulting large amount of related literatures,the research status of domestic and abroad,including coal macerals analysis,feature reduction algorithm and image classification algorithm,are sumrized.According the standard of classification of coal of China,the structural characteristics of each maceral image of inertinite are analyzed in detail.(2)According to the characteristics and differences of structure of each maceral of inertinite,microscopic image of macerals are decomposed with Curvelet Transform,coefficients of each scale layer after transform are composed as the initial feature set,and the initial feature set is analyzed.(3)In view of the fact that the dimension of the initial feature set is higher,with the aid of the thought of compressive sensing algorithm,dimension of sparse high frequency coefficients after curvelet transform are reduced,which form the final feature set by combing with low frequency coefficients of the reserved Coarse layer,so as to conserve enough useful information while redundancy data and dimensions of coefficient set are reduced.By analyzing the relationship between the selected dimension and the effective information amount,the value range of row of themeasurement matrix(same as the dimension of feature data)corresponding to high classification accuracy is obtained.(4)A combined support vector machine(SVM)based classifier is built,and the macerals of inertinite are automatically classified.To verify the validity of compressive sensing based dimensionality reduction method proposed in this dissertation,some dimensionality reduction algorithms,such as PCA(principal component analysis),KPCA(kernel principal component analysis)are also employed,and classification results with these algorithms are compared.To analyze the effect of different feature set on the classification accuracy,frequency coefficients after wavelet transform and low frequency coefficients of the Coarse layer are also employed respectively as the input data of the classifier,and the results of the classification are compared too.The specificity and innovation of this dissertation lies in: According to the differences in texture structure of macerals in inertinite,curvelet transform is introduced to the extraction and representation of texture feature of inertinite macerals;with the aid of the thought of compressive sensing algorithm,a new dimensionality reduction method is proposed,and is employed to reduce the dimension of sparse high frequency coefficients after Curvelet Transform,so as to complete the automatic classification of inertinite macerals with fewer coefficient set.