基于随机森林和Landsat 8 OLI影像的脐橙果园种植信息提取  

Navel Orange Orchard Plantation Information Extraction Based on Random Forest and Landsat 8 OLI Images

作　　者：林娜[1] 王伟[1] 王斌 LIN Na

机构地区：[1]重庆交通大学土木工程学院,重庆400074 [2]重庆市地理信息和遥感应用中心,重庆401147

出　　处：《地理空间信息》2021年第11期96-100,I0007,I0008,共7页Geospatial Information

基　　金：重庆市教委科技资助项目(KJQN201800747)。

摘　　要：针对从中分辨率遥感影像中提取果园种植信息所面临的特征维数高、提取精度低等问题,采用随机森林算法进行特征优选,进而提取脐橙果园种植信息。以重庆奉节为研究区,首先选取春、秋、冬3季Landsat 8 OLI遥感数据生成光谱特征、植被指数特征和纹理特征,并结合地形因子构建初始特征集;再通过随机森林算法确定优选特征集;最后基于随机森林分类器,利用初始特征集和优选特征集进行脐橙果园的提取实验,并对各实验结果进行了精度评估和对比分析。研究结果表明,基于随机森林算法构建的优选特征集能在降低特征维度的情况下获得较高的精度,总体精度和Kappa系数分别为90.71%和0.89。Aiming at the problems of high feature dimension and low extraction accuracy in the process of extracting orchard plantation information from middle resolution remote sensing images,we extracted the navel orange orchard plantation information by using random forest algorithm with the feature optimization.Taking Fengjie Country,Chongqing City as study area,we generated spectral features,vegetation index features and texture features based on three Landsat 8 OLI images in spring,autumn and winter,and combining with terrain information,built initial feature sets at first.And then,we used the random forest algorithm to optimize feature set.Finally,based on random forest classifier,we used the initial feature set and optimized feature set to extract navel orange orchard plantation information,and estimated and compared the results of each experiment.The results show that the optimized feature set based on random forest algorithm can achieve high accuracy while reducing the feature dimension,with an overall accuracy of up to 90.71%and Kappa coefficient of 0.89.

关 键 词：随机森林 特征选择 遥感提取 脐橙果园 Landsat 8 OLI 

分 类 号：P237[天文地球—摄影测量与遥感]