基于GF-1数据复杂地区地物类型提取探究

GF-1 Data-Based Extraction of Land Cover of Complex Areas

土地利用/土地覆盖作为人类活动对地球环境影响的一个重要指标,其在全球变化和土地覆盖变化监测中发挥着重要作用.以重庆市永川区为例,以国产GF-1 PMS为数据源,分别采用随机森林(Random Forest,RF)、支持向量机法(Support Vector Machine,SVM)和人工神经网络(Artificial Neural Network,ANN)算法实现研究区内地表覆盖类型的提取.对比结果表明:在训练样本相同的条件下,RF算法地物分类精度要优于SVM和ANN算法,尤其是林地、道路和大棚三类地物RF算法分类精度与SVM和ANN算法分类精度之间差异较大.但RF分类结果中依旧存在像元错分和漏分现象,本研究利用易混淆像元在NDVI和形状上的差异,实现了RF分类结果中易混淆像元的修正,提高了地物分类的精度.

As an important indicator of the impact of human activities on the earth s environment,land use&land cover data play a crucial role in monitoring global change and land cover change.In a study reported in this paper,Yongchuandistrict of Chongqing municipalitywas used as an example and,based on domestic GF-1 PMS data,the land cover types in the study region were extracted with RF(random forest),SVM(support vector machine)and ANN(artificial neural network)algorithms.The results showed that with the same training samples,RF algorithm gave more accurateland cover classification than SVM and ANN algorithms.Especially,the difference was statisticallysignificant between the classification accuracy of RF algorithm and that of SVM and ANN algorithms for forest land,roads and greenhouses.However,misclassification and missing of pixels remained in RF classification results.In this study,the differences of confusable pixels in NDVI(Normalized Difference Vegetation Index)and shape were utilized to realize the correction of confusable pixels in RF classification results,and the classification accuracy of ground objects was improved.