基于Hyperion高光谱数据和随机森林方法的岩性分类与分析  

Lithological classification and analysis using Hyperion hyperspectral data and Random Forest method

作　　者：柯元楚 史忠奎 李培军[1] 张西雅[2] KE YuanChu;SHI ZhongKui;LI PeiJun;EHANG XiYa(Institute of Remote Sensing and GIS,School of Earth and Space Sciences,Peking University,Beijing 100871,China;fnstitute of Urban Meteorology,China Meteorological Administration,Beijing 100089,China)

机构地区：[1]北京大学地球与空间科学学院遥感与地理信息系统研究所,北京100871 [2]中国气象局北京城市气象研究所,北京100089

出　　处：《岩石学报》2018年第7期2181-2188,共8页Acta Petrologica Sinica

基　　金：国家自然科学基金项目(41371329)资助

摘　　要：探索利用高光谱数据的岩性填图新方法是遥感地质应用领域的重要需求之一。本文运用随机森林方法和EO-1Hyperion高光谱数据,对新疆塔里木西北部柯坪地区的局部区域进行岩性分类,并对相关问题进行分析。分别利用光谱特征以及加入光谱一阶导数特征进行岩性分类,并对不同特征对岩性分类的重要性进行分析,同时与现有的基于光谱角制图方法(SAM)进行比较。结果表明,与SAM方法相比,随机森林方法得到了更高精度的岩性分类结果,是一种有效可行的岩性分类方法。根据特征重要性的排序,蓝绿光波段、短波红外波段以及相应的一阶导数特征对研究区Hyperion数据的沉积岩岩性分类贡献更大。Exploration of new methods for lithological mapping using hyperspectral data is one of critical needs in the field of geological remote sensing applications. In this paper,random forest method was used to classify EO-1 Hyperion hyperspectral data for lithological mapping over a portion of Kalpin,Northwest Tarim,Xinjiang,and the relevant problems were analyzed. The lithological classification was carried out by using spectral features alone and using both spectral features and first derivatives. The importance of different features to lithological classification was calculated and analyzed. The Random Forest method was compared with spectral angle mapper（ SAM）,a commonly used classification method. The results show that the Random Forest method achieved a significant improvement in accuracy upon lithological classification compared with SAM. Thus,the adopted method is a feasible lithological classification method. According to the ordering of feature importance, blue and green, shortwave infrared bands and their corresponding first derivatives contribute more to lithological classification of sedimentary rocks in the study area.

关 键 词：高光谱遥感数据 随机森林 光谱角制图 岩性分类 

分 类 号：P627[天文地球—地质矿产勘探]