典型煤岩反射光谱无监督感知方法研究

Research on unsupervised sensing methods of typical coal and rock based on reflectance spectroscopy

  • 摘要: 针对现有煤岩反射光谱有监督识别方法存在煤岩位置变化时识别效果差的问题,为研究基于反射光谱的煤岩自主适应性识别,提出了基于聚类距离改进型模糊C均值聚类(FCM)算法的典型煤岩反射光谱无监督感知方法。以兴隆庄煤矿气煤、泥岩、粉砂岩、泥质灰岩4种典型煤岩样品为研究对象,测定了每种试样多个背向反射角下的近红外波段的反射光谱曲线,分析了4种煤岩反射光谱最具差异性的特征波段,选取2 150~2 400 nm作为4种煤岩反射光谱差异性特征波段,在特征波段内,对气煤-泥岩、气煤-粉砂岩、气煤-泥质灰岩光谱组合进行煤岩反射光谱无监督识别研究。研究结果表明:4种试样表面的背向光谱反射率均呈现出随背向反射角增大而先增大后减小的整体趋势,背向反射角增大时,泥岩、粉砂岩和泥质灰岩的各吸收谷深度变化较小,只有微弱的减小,气煤的各吸收谷深度减小相对明显;采用改进型FCM(RFCM,CFCM)方法将光谱数据快速聚类,由最终聚类隶属度概率矩阵判定光谱数据类别,进而判定不同位置煤岩类别;相较于FCM,改进型FCM对各煤岩组合的识别率均大于90%,其中CFCM对各煤岩组合聚类识别迭代次数最少,总耗时均小于0.1 s,为优先选择方法,为反射光谱技术应用于煤岩界面不同位置煤岩的高效适应性判定提供了参考。

     

    Abstract: In view of problem of poor recognition effect of existing supervised recognition methods of coal and rock based on reflectance spectroscopy when positions of coal and rock change, in order to study self-adaptive recognition of typical coal and rock based on reflectance spectroscopy, an unsupervised sensing methods of typical coal and rock based on reflectance spectroscopy and fuzzy C-means clustering (FCM) algorithms with improved clustering distances was proposed. Four typical types of coal and rock samples of Xinglongzhuang Coal Mine including gas coal, mudstone, siltstone and argillaceous limestone were studied and spectral reflectance curves of each sample were measured in near infrared band at multiple back reflection angles. The characteristic band with the most different spectral curves of the four types was analyzed and 2 150-2 400 nm were selected as the characteristic bands with the differences of the four types. In the characteristic band, the unsupervised recognition of reflectance spectra of coal and rock was studied for each coal-rock spectra combination of gas coal-mudstone, gas coal-siltstone and gas coal-argillaceous limestone. The results showed that with increasing of back reflection angle, back spectral reflectance of surfaces of all the four types increased first and then decreased. Meanwhile, the depth of absorption valleys of mudstone, siltstone and argillaceous limestone slightly decreased, and the decrease of the depth of absorption valleys of gas coal was relatively obvious. The improved FCM (RFCM, CFCM) methods were used to cluster spectral data quickly, and classifications of the spectral data were determined by membership probability matrix of the final clustering to recognize classifications of coal and rock at different positions. Comparing with FCM, the recognition rates of each coal-rock combination were both more than 90% using the two improved FCM methods. Among them, CFCM took the least number of iteration to cluster and recognize each coal-rock combination, and its total time consumptions were all less than 0.1 s. CFCM is the preferred method and provides a reference for the application of reflectance spectroscopy technology to the highly efficient and adaptive recognition of coal and rock at different positions of coal-rock interface.

     

/

返回文章
返回