| The exinite is one of the three major micro-components of coal maceral.Although it has less storage,but higher hydrogen content,hydrocarbon production capacity and volatility,and is widely used in oil and gas generation.Therefore,the automatic classification and identification of exinite in coal is of great research signifance.Based on the analysis and comparison of the characteristics of microscopic images of exinite in coal,aiming at the problems of complex and diverse structure and large amount of information redundancy features obtained by traditional methods,convolutional neural network(CNN)model is employed to extract feature of exinite microscopic images,and a support vector machine classifier,which is suitable for small sample classification problem,is constructed for the classification.Bying comparing with the classification results,that features of exinite are extracted with traditional methods,the validity of convolutional neural network model for feature extraction of microscopic images of exinite is verified.The main work of the dissertation is as follows:(1)On the basis of extensive consulting relevant literatures,the research status of the coal macerals identification algorithm,image classification algorithm and the convolutional neural network algorithm at home and abroad are reviewed.(2)Characteristicss of microscopic images of exinite in coal are analyzed in respect of color,shape and texture,and total of 19-dimensional features based on the gray-scale co-variance matrices,Tamura texture and wavelet decomposition are extracted and analyzed.(3)Based on the basic principles and algorithms,the validity of convolutional neural networks for feature extraction of microscopic images of exinite in coal are explored.By constructing different convolutional neural network models and selecting different network depths,the influence of network model and network depth on the feature extraction of exinite is analyzed.(4)According to the feature matrix of the ideal feature layer extracted by the deep neural network,a support vector machine classifier is constructed to classify the micro-components of exinite in coal.By comparing with the classification result with that from the 19 dimensional traditional features,the effectiveness of the method by using convolutional neural network for feature extraction of exinite is verify.The specifity and innovation of this dissertation lies in: convolutional neural network directly takes the original image as input,performs high-level abstract modeling on sample data,constructs complex high-dimensional feature space,automatically learns features from training samples,and realizes feature hiding.Compared with the traditional feature extraction method,this method has advances of strong feature extraction and expressiveness,high classification accuracy and the algorithm is simple. |
