基于反射光谱与神经网络的砂岩型铀矿岩性智能分类研究

Lithology Classification Using Artificial Neural Network Based on Reflectance Spectrum

砂岩型铀矿是一种重要的铀矿资源。岩心是揭示砂岩型铀矿成因环境和深部地质信息的直接反应。当前利用神经网络结合光谱进行岩性智能识别的研究主要针对地表岩石,而针对钻孔岩心的工作则相对较少。为了补充相关工作,以松辽盆地钱家店地区的钻孔岩心为对象,开展了针对氧化-还原带不同颜色和不同粒度砂岩样品的光谱收集。同时,利用Visual Basic+C混合编程技术,开发了基于岩石反射光谱与神经网络技术的岩性学习与自动分类模块。利用该模块,对砂岩光谱进行了特征学习,并利用学习结果对岩心光谱开展了岩性自动分类实验。结果表明:基于光谱的神经网络技术可对砂岩型铀矿中氧化带与还原带砂岩进行识别,也可识别不同颜色和不同粒度的砂岩,其识别的准确度取决于不同岩石光谱差异大小。此外,岩心表面的附着物会对识别结果产生影响,应在光谱测量前予以剔除。

Sandstone-type uranium deposit is an important uranium resource.Borehole core is the basis for understanding the mineralization environment and direct refleetion of deep geological information.Since many studies used artificial neural network(ANN)and reflectance spectra to automatically classify rocks on surface but few focuses on cores.In order to fill the blanks,this paper collected the spectra of the sandstone samples with different lithologies of drill cores from different redox zones in Qianjiadian area,south Songliao basin,and developed a software package based on reflectance spectroscopy and ANN using Visual Basic and C languages.By utilizing this software,several ANNs were trained with spectra of sandstones.The ANNs were used to do automatic lithology classification,and the results showed that the spectrum-based neural network technology can identify sandstones from oxidation zone and reduction zone,and can also identify sandstones with different colors and different grain sizes.The identification accuracy depends on the spectral difference.Additionally,the covers on the core surface will affect the identification results,thus should be avoided before the spectra were measured.