煤岩显微组分的鉴别需综合显微组分的反射率、颜色、突起、形态、大小、细胞结构等特征。基于图像分析的煤岩显微组分自动识别方法能够降低人工的劳动强度并提高测试速度与可重复性, 因而其为煤岩显微组分分析的未来发展趋势。煤岩显微组分种类较多且不同显微组分的结构特征差异较大, 目前对其光学性质的定性描述很难体现许多显微组分独特的形貌及结构特征, 因而识别准确度较低, 急需对煤岩显微组分的形态特征和纹理特征进行深入研究和表征。针对煤岩显微组分的识别特征, 分析油浸反射光下煤岩显微组分的灰度、形态及纹理特征等图像学特征, 对15 种有机显微组分及3 种无机显微组分选取1 693 幅能够表征显微组分特征的代表性图像, 构建烟煤显微组分特征数据集, 利用灰度共生矩阵对灰度均值、灰度方差、灰度歪斜度、灰度峰态、灰度能量、灰度熵等6 种灰度特征参数以及纹理对比度、纹理相关性、纹理角二阶矩和纹理同质性等4 种纹理特征进行提取, 利用K 最近邻( KNN) 和支持向量机( SVM) 2 种机器学习分类算法进行分类。结果发现, 显微组分的灰度特征及纹理特征差异较大, 支持向量机分类算法对显微组分的识别效果优于K 最近邻, 基于灰度+纹理特征的支持向量机算法可以对煤中18 种显微组分实现大于60%的识别精度。

The identification of coal macerals requires a comprehensive consideration of their reflectance, color,protrusions, morphology, size, and cellular structure characteristics. The automatic identification method of coalmacerals based on image analysis can reduce manual labor intensity and improve testing speed and repeatability,making it the future development trend of coal macerals analysis. There are many types of coal macerals, and thestructural characteristics of different macerals vary greatly. Currently, qualitative description of their optical propertiesis difficult to reflect the unique morphology and structural characteristics of many coal macerals, resulting in low rec⁃ognition accuracy. It is urgent to conduct in⁃depth research and characterization of the morphological and texture features of coal macerals. According to the recognition characteristics of coal macerals, the characteristics of gray value,morphology and texture of coal macerals under oil immersion were analyzed. 1 693 representative images that cancharacterize the characteristics of macerals were selected from 15 organic macerals and 3 inorganic macerals. The pa⁃rameters of gray and texture characteristics were statistically analyzed by gray level co-occurrence matrix. Gray pa⁃rameters such as mean, variance, skewness, kurtosis, energy and entropy, and texture parameters such as con⁃trast, correlation, angular second moment and homogemeity were extracted. Two machine learning classification al⁃gorithms, K nearest neighbor (KNN) and support vector machine (SVM), were used to classify 1 693 petrographicimages of 18 macerals respectively. The results have shown that the gray and texture features of macerals are quite dif⁃ferent. The recognition effect of support vector machine algorithm is better than k nearest neighbor. The SVM algo⁃rithm based on gray + texture features can achieve recognition accuracy of more than 60% for 18 macerals, which isbetter than that of KNN algorithm.