Coal Rock Macerals Classification By MSVM Based On United Texture Feature

Coal rock macerals analysis technology is an important method to analyze coal rock, and provides an important basis for coal coking, gasifying industry. It is a pity that traditional analysis system can not meet requirements of objectivity and real-time with its low automation degree, high cost, heavy professional skill degree and evident subjectivity. Based on this, digital image processing technology is introduced in the coal analysis field to classify and recognize coal rock macerals, ameliorating the traditional analysis system to improve its automation degree and analysis efficiency in the field of coal rock, and making some reference. In the paper, these following aspects are done.(1) Removing noises such as cratches and holes in the microscopic images. Scratches and holes in the process of preparation as inevitable noises will affect the identification correctness by using digital image processing techniques, so these noises should be removed. In order to remove scratches effectively, after analyzing a large number of microscopic images provided by the coking plant, finding the scratches are straight lines with characteristics of different degree, different shades, different length. Based on this, this paper compares method based on gray image mathematical morphology, method based on average replacement of adjacent area, method based on traditional FMM(Fast Marching Method) algorithm and improved FMM algorithm to repaint scratches, founding that the improved FMM algorithm can not only repaint scratches effectively, but also can protect simulation of the original image texture, much better than other methods. To fill the holes in the image, this paper uses a mathematical morphology reconstruction method based on gray image and finds that it can fill the holes effectively by experimental verification.(2) Extracting coal macerals texture feature. Coal macerals texture is a typical natural texture with great randomness and complexity. RILBP(Rotation Local Binary Pattern) can keep texture feature does not change under the condition of image rotation, and have better anti-interference ability to light and dark noise, but with higher dimension feature space, GLCM(Grey Level Co-occurrence Matrix) has insufficient ability of anti-interference and rotational resistance, but with lower dimension feature space. Based on this, this paper proposes a RILBP and GLCM fusion method for texture feature extraction, in the hope to have resistance to rotation and interference but lower dimension feature space. Contrasting three texture feature extraction methods, finding that RILBP-GLCM method is better than RILPB, GLCM with lower volatility.(3) Classficating cola rock macerals. MSVM(Multi-classification Support Vector Machine, MSVM) is used to classify cola rock macerals, containing Vitrinite, Inertinite, Exinite, Minerals and Background. 1134 coal rock microscopic images are collected, among which 401 are used as training samples, and 733 are used as test samples, the texture feature vectors extracted by RILBP, GLCM, RILBP-GLCM as the MSVM training and test samples, and obtaining their classification results. The results show that RILBP-GLCM has the highest classification accuracy is 92.7694% when grayscale compressed at level 16, higher than the 31.3652% of RILBP and 89.4952% of GLCM algorithm. And this proves the effectiveness by using RILBP-GLCM method to extract coal rock macerals texture feature.