基于机器学习的矿物智能识别方法研究进展与展望

Research Advances and Prospective in Mineral Intelligent Identification Based on Machine Learning

矿物智能识别是地球科学与信息科学的重要交叉方向,显示出强大的生命力.本文在调研国内外研究动态的基础上,把矿物智能识别划分为4个阶段,即矿物采集、数据获取、模型构建、分类判别;根据测试方法和获得的数据类型,把矿物智能识别分为基于化学成分、显微光学图片、光谱分析的3条基本路线;总结了应用于矿物智能识别的机器学习方法和技术,包括统计学习、线性回归模型、距离度量模型、树结构模型、神经网络模型及其与样本问题相关的新技术.在此基础上,提出消除地质学与人工智能的鸿沟、建设可用于学习的高质量矿物数据集、完善适合矿物智能识别的机器学习方法、增进模型可解释性、加强工业推广的实践是该领域未来的重点发展方向.

Mineral intelligent identification is a developing interdisciplinary research field between earth science and information science,where machine learning shows great vitality.This paper divides the procedure of mineral intelligent identification into four stages,including mineral collection,data acquisition,model building and category discriminant.Based on the test methods and data types,the mineral intelligent identification can be achieved by three different research routes,namely,chemical-compositionbased,microscopic-optical-image-based and spectral-image-based.Various methods of machine learning for mineral intelligent recognition are reviewed in detail including statistical learning,similarity measurement,decision tree,artificial neural network and few new technologies related to testing sample.We suggest that the future directions in this field are to eliminate the gap between geology and artificial intelligence,to build high-quality mineral datasets that can be learned by the machine,to explore and consummate machine learning methods suitable for mineral intelligence identification,to increase the ability of model explanation,and to strengthen the practice of industrial application.