基于光谱—空间特征的ASTER影像岩性分类研究——以甘肃北山白峡尼山地区为例  

Study on lithologic classification of ASTER image based on spectral—spatial features——A case study of Baixiani Mountain,Beishan Mounntains,Gansu Province

作　　者：梅佳成 刘磊[1,2] 尹春涛 张群佳 王乐 MEI Jiacheng;LIU Lei;YIN Chuntao;ZHANG Qunjia;WANG Le(School of Earth Science and Resources,Chang’an University,Xi’an,710054;Key Laboratory of Strategic Mineral Resources of the Upper Yellow River,Ministry of Natural Resources,Lanzhou,730046)

机构地区：[1]长安大学地球科学与资源学院,西安710054 [2]自然资源部黄河上游战略性矿产资源重点实验室,兰州730046

出　　处：《地质论评》2024年第1期239-250,共12页Geological Review

基　　金：陕西省自然科学基础研究计划(编号:2023-JC-ZD-18);自然资源部黄河上游战略性矿产资源重点实验室开放课题资助项目(编号:YSMRKF202203);中央高校基本科研业务费专项资金(编号:300102353501)的成果。

摘　　要：遥感岩性制图是地质填图中的重要工作,基于光谱特征的岩性分类易受到色调、纹理等因素影响导致精度不佳。前人进行岩性自动分类研究多关注影像的光谱特征,而忽略空间特征,笔者等基于甘肃北山白峡尼山地区ASTER影像,将支持向量机、极限学习机两种机器学习分类方法与基于空间特征的快速漂移算法相结合进行岩性分类。结果表明支持向量机分类总体精度为89.17%;极限学习机不但具有需调节参数少的优势,且分类精度和速度均优于支持向量机,分类总体精度达96.70%;利用快速漂移算法提取的影像空间特征可有效减少错分区,提升岩性分类效果。研究证实将基于光谱特征的极限学习机和基于空间特征的快速漂移算法结合的岩性分类方法具有客观、高效、高精度等优势,可为后续地质填图和找矿勘查工作提供可靠数据支撑,在遥感岩性分类领域具有较高的推广价值。Objectives:Geological mapping is a basic work for geology.However,the working areas for geological mapping are mostly mountainous areas with high elevation and steep terrain which is difficult for field work.It’s urgent to develop a semi-automatic to automatic lithologic mapping method using remote sensing data by combining the spectral features and spatial features of each lithologic unit.Therefore,this study,taking Baixiani Mountain,Beishan Mountains,Gansu Province,as the working area,utilized two machine learning methods to test the feasibility for automatically lithologic mapping.Methods:Two methods,support vector machine and extreme learning machine,combined with a spatial feature extraction method,quick shift algorithm,were used to process ASTER remote sensing data for lithologic classification.Results:The overall accuracy of support vector machine classification was 89.17%,while the extreme learning machine not only had the advantage of requiring fewer adjustable parameters,but also had higher classification accuracy and speed than the support vector machine,with an overall accuracy of 96.70%.The use of image spatial features extracted by the quick shift algorithm effectively reduced misclassification areas and improved lithological classification.Conclusions:The study confirmed that the lithological classification method combining extreme learning machine based on spectral features and quick shift algorithm based on spatial features has advantages such as objectivity,efficiency,and high accuracy,and can provide reliable data support for subsequent geological mapping and mineral exploration work,with high promotion value in the field of remote sensing lithological classification.

关 键 词：岩性分类 支持向量机 极限学习机 快速漂移 机器学习 北山 

分 类 号：P627[天文地球—地质矿产勘探]