一种基于PCA与RBF-SVM的煤岩显微组分镜质组分类方法

A classification method of vitrinite for coal macerals based on the PCA and RBF-SVM

在分析煤岩镜质组显微组分特点的基础上,针对其结构复杂、特征量多且相互交织从而影响分类准确性等问题,提出一种基于主成分分析(PCA)的煤岩显微组分镜质组分类方法。首先根据镜质组显微图像中各组分呈现的条状、团块、颗粒等纹理特点和亮度差异,采用基于灰度共生矩阵的能量、熵、惯性矩、局部平稳性等纹理特征量和基于灰度分布统计的亮度比、均值、均方差、三阶矩偏度等亮度相关特征量对其进行描述,构成初始特征量集;再采用主成分分析法对初始特征量集进行进一步的抽取;最后构建基于径向基函数的支持向量机(RBF-SVM),采用积累贡献率较大的主成分作为分类参量实现镜质组的自动分类。实验结果表明:纹理和灰度统计特征可有效刻画煤岩镜质组显微组分;采用PCA对初始特征进行抽取之后,用于分类的特征空间维数大幅度降低,分类算法的泛化能力增强,分类的准确率显著提高。

On the basis of analyzing the characteristics of macerals in vitrinite of coal,in view of the fact that the struc- tures of macerals are very complex and there are too many features that mix and makes classification difficult, a macer- al classification method based on principal component analysis （PCA） is proposed. Firstly, according to the texture characteristics （ strip, crumb, grain, etc） and intensity difference, macerals are represented with texture related features as energy, entropy, moment, local smooth based on gray level co-occurrence matrix and intensity related features as contrast, mean, standard deviation, 3-order moment deviation based on the gray-level statistics of coal microscopic ima- ges, and a primary feature set is generated. Then, by using PCA, primary features are further selected and extracted. Fi- nally, a Support Vector Machine based on radial basis function （RBF-SVM） is built, and macerals are classified ac- cording to those principal components with greater cumulative contribution. Experimental results show that texture can present macerals feature of vitrinite effectively;with features extracted by PCA, the dimensions of feature space are greatly reduced, the generalization ability of classification algorithm is improved, and the accuracy rate of classification is obviously increased.