Research On Classification Method Of Coking Coal Macerals

Coking coal maceral refer to the basic organic constituent units of coking coal under microscope. The component of micro-structure of coking coal is closely related to it's properties, such as coking behavior, cohesiveness, CO adsorption capacity, thermal fragmentation and so on. Therefore, to implement the automatic classification and identification of macerals in coking coal is of imporment significance for predicting the nature of coking coal, utilizing coking coal resources reasonablely and optimizing coal blending structure. Based on analyzing the characteristics of macerals in coking coal micrographs, the texture features of coking coal macerals are firstly extracted, and the texture features are analyzed and the effective feature quantities are selected. Then, classification scheme is designed to classify the coking coal maceral. The main content of the dissertation is as follows:(1) Based on large amont of relevant literature, research status of coking coal maceral analysis and image classification are reviewed.(2) The characteristics and differences among different coal maceral are analyzed in detail. 41 dimensional features are extracted based on the gray-scale statistical distributions,the gray-scale co-variance matrices, the gray scale length, local binary mode, wavelet decomposition and Tamura texture, and the effectiveness of each feature to the coking coal macerals classification is analyzed by statistical approach.(3) According the results of feature analysis, the effective features are selected to construct feature set, and the classification scheme of the coking coal maceral is built with the support vector machine. The selected features in the feature set are directly used to classify the main three categories of inertia, vitrinite, exinite and some typical macerals in each category.(4) In view of the problem that the dimension of the feature space of the coking coal maceral is too high that makes classification difficult, a dimension reduction method named PCA-SLPP is proposed based on manifold learning, which is employing the advantage of Principal Component Analysis (PCA) and that of Supervised Locality Preserving Projections(SLPP) algorithm. By experiments of coking coal macerals classification, the effectiveness of the reduced dimension method is verified.(5) Combining the advantages of principal component analysis and extreme learning machine, and by introducing the singular value decomposition into the extreme learning machine, a classification method named PSVD-ELM based on improved extreme learning machine is proposed. This classifying method is employed to the coking coal maceral classification, and the results are compared with those from other classification methods,effectiveness of the new method is verified.The special and innovation of the dissertation lie in: By take factors of the global structure, locality preserving and label information of feature data into consideration, the dimension of feature space is reduced, and the accuracy of the classification is improved simultaneously. Combing advantages of the principal component analysis and the extreme learning machine and introducing the singular value decomposition into the extreme learning machine, macerals of coking coal are classified effectively. Compared with those of commomly used methods, the proposed method has not only a better performance of feature extraction, but also a faster speed and a higher accuracy of classification.