利用粗糙集理论进行遥感分类信息提取

Remote Sensing Classification Information Extraction Based on Rough Set Theory

以辽宁省双台子河口湿地为研究对象,以Landsat 8和HJ-1-A/HJ-1-B的多时相遥感影像为数据源,根据研究区现状,将研究区分为旱地、芦苇、水田、碱蓬、混合植被、水面、滩涂、居民点、养殖塘九个类型。利用时间序列的归一化植被指数提取植被与非植被的分类阈值,采用粗糙集理论和多时相遥感影像,对植被和非植被分别进行分类规则的获取,建立了研究区决策树分类模型。为了进行精度评价,利用相同的训练点又进行了同样基于像元的最大似然法分类。最后利用混淆矩阵对上述两种方法进行了精度评估,基于粗糙集的决策树分类法与最大似然法总体分类精度分别为93.70%和91.62%,Kappa系数分别为0.92和0.90,两项指标值基于粗糙集理论法均比最大似然法有所提高。这为构建决策树分类模型进行湿地地表分类信息提取提供了一条新的研究思路。

Using Shuangtaizi estuarine wetland as the research area,and Landsat 8and HJ-1-A/HJ-1-B remote sensing data as the data sources,this study was conducted for land-cover information extraction.According to the status of the land,the study area was divided into 9categories,including upland,reed,paddy field,Suaeda,mixed vegetation,water body,beach,residential land,culture pond.First,the study area was divided into vegetation and non-vegetation using the time-series normalized difference vegetation index（NDVI）.Then,the classification rules of vegetation and nonvegetation were extracted based on the rough set theory and on multi-temporal remote sensing data.Finally,a decision tree classification model was established.For the purpose of an accurate evaluation,the maximum likelihood classification was conducted based on the pixels using the same training samples;and the confusion matrix and kappa coefficient were calculated.The results showed that the overall accuracy both of the deeision tree and the maximum likelihood classification reached up to93.70% and 91.62% with a kappa coefficient of 0.92 and 0.90 respectively.The two evaluation index values were improved.It provided a novel research idea for a wetland classification information extraction based on remote sensing images.