基于可见光-近红外光谱的煤种分类方法

利用便携式地物光谱仪SVC HR-1024对92个烟煤和58个褐煤样本进行光谱测试,烟煤和褐煤在可见光-近红外波段光谱特征差异明显,褐煤的光谱反射率及其斜率均明显高于烟煤.在光谱特征分析的基础上,利用MAO模型法、随机森林法、BP神经网络法和ELM算法进行煤种分类.结果表明:MAO模型法和随机森林法的分类结果较优.若进行大面积、快速遥感识别时,对分类时间要求较高,应选择MAO模型法;若是小面积单一矿区分类,对分类准确率要求较高,选择随机森林法较为恰当.

Coal Classification Based on Visible and Near-Infrared Spectrum

The portable spectrometer SVC HR-1024 was used to carry out spectral tests on the 92 bituminite and 58 lignite coal samples from various coal mines. By comparing their spectral curves， the differences between bituminite and lignite samples can be observed visibly in spectral characteristics. The reflectance of lignite samples is obviously higher than that of bituminite samples， as well as the slope of spectral curves. On the basis of spectral characteristics analysis， the MAO model algorithm， random forests， BP neural networks and ELM-neural network were selected for the classification of bituminite and lignite samples. The results indicated that the MAO model algorithm and random forest algorithm outperform other algorithms on classification. For large-area and rapid recognition by remote sensing， the MAO model algorithm has a great advantage in the classification time. While the random forest algorithm can be used for classification in small mining areas.