基于随机森林模型的不同空间分辨率遥感影像草地提取研究  

Research on grassland extraction from remote sensing images with different spatial resolutions based on random forest model

作　　者：宋梦龙 颜涛 张鹏 王力琪 程博 SONG Menglong;YAN Tao;ZHANG Peng;WANG Liqi;CHENG Bo(Inner Mongolia Remote Sensing Center Limited Liability Company,Hohhot 010010;AerospaceInformation Research Institute,Chinese Academy of Sciences,Beijing 100094;Lanzhou Jiaotong University Faculty of Geomatics,Lanzhou 730000,China)

机构地区：[1]内蒙古遥感中心有限责任公司,呼和浩特010010 [2]中国科学院空天信息创新研究院,北京100094 [3]兰州交通大学测绘与地理信息学院,兰州730000

出　　处：《草原与草业》2025年第1期12-17,共6页Grassland and Prataculture

基　　金：内蒙古自治区科技计划项目(2022YFSJ0008)资助。

摘　　要：遥感技术已成为快速有效获取草地覆盖信息的重要工具,在不同的研究中选择合适的空间分辨率影像至关重要。然而,基于随机森林(Random Forest,RF)模型的不同空间分辨率遥感影像草地提取研究依然存在不足。本文利用GF-6、Sentinel-2和Landsat 8遥感影像所形成的2~30 m间3个不同空间分辨率影像的数据集,依据不同地物的光谱特征,基于RF模型对内蒙古达尔罕茂明安联合旗典型性荒漠半荒漠化草原进行提取研究。研究结果表明,基于RF模型的分类方法,草地提取精度随着影像分辨率的升高总体呈现上升的趋势;国产高分影像的分类精度最佳,最高总体精度可达89.09%;不同空间分辨率的影像中,相同光谱特征的重要性会有所不同;高分影像在荒漠半荒漠化草原分类中具有较高的有效性和适用性。Remote sensing technology has become an essential tool for rapidly and effectively acquiring grassland cover information.Selecting appropriate spatial resolution imagery is crucial for various studies.However,research on grassland extraction using remote sensing imagery of different spatial resolutions based on the Random Forest(RF)model remains insufficient.This study utilized datasets with three different spatial resolutions ranging from 2 to 30 meters,derived from GF-6,Sentinel-2,and Landsat 8 imagery.Based on the spectral characteristics of different ground objects,the RF model was employed to extract typical desert and semi-desert grasslands in Darhan Mumingan Joint Banner,Inner Mongolia.The results indicated that the accuracy of grassland extraction based on the RF model generally improved with increasing image resolution;the classification accuracy of domestic highresolution imagery(GF-6)was the best,achieving the highest overall accuracy of 89.09%;the importance of the same features varied across images of different spatial resolutions;and high-resolution imagery demonstrated high effectiveness and applicability in classifying desert and semi-desert grasslands.

关 键 词：随机森林 空间分辨率 草地提取 荒漠半荒漠化草原 

分 类 号：P237[天文地球—摄影测量与遥感]