基于多时相Landsat8数据的南方丘陵区典型地物信息提取——以秦淮河流域为例  

Land Use Classification of Typical Ground Objects in Qinhuai River Basin Based on Multi Temporal Landsat8 OLI Data

作　　者：宋明明[1] 郑文龙[1] 卞国栋 都金康[1] 

机构地区：[1]南京大学地理与海洋科学学院江苏省地理信息资源开发与利用协同创新中心江苏省地理信息技术重点实验室,南京210000

出　　处：《科学技术与工程》2016年第18期209-216,共8页Science Technology and Engineering

基　　金：国家自然科学基金项目(41371044)资助

摘　　要：以南方丘陵区秦淮河流域为研究区,采用多时相Landsat8遥感影像进行典型地物的光谱采样。根据典型地物光谱特征统计结果,针对各地类特点选取最优的指数提取模型,最终形成具有分层提取特点的决策树分类模型。结果表明:1借助植物的物候特征,采用多时相遥感影像进行水田与植被的划分能够取得较好效果;2秦淮河流域水田与旱地交错分布,水田主要分布于秦淮河河流洼地,夏季的水田与旱地光谱特征差异最明显,是遥感信息提取最好的时间,短波红外波段为水的吸收波段,在区分水田与旱地信息上具有明显优势;3归一化植被指数NDVI可区分农田信息与非农田信息,但由于部分绿化程度好的建筑用地在光谱信息上与农田存在一定的相似性,仍然存在一部分明显的误分现象,采用裸土指数BSI可以有效地将这部分建筑用地从农田信息中剔除;4建筑用地与裸地的混分问题是土地利用信息提取的难点之一,研究中尝试了多种指数模型与监督分类方法,结果显示监督分类方法在区分建筑用地与裸地信息方面效果最好;5决策树分类模型较单一的监督分类方法总体精度提高了23.1%。Multi-temporal Landsat 8 OLI image were utilized to extract thematic information of land use based on the study area-Qinhuai river basin which is located at southern hilly region of China. We first conduct the landuse spectrum sampling. Then the optimal index extraction model was selected for each class,based on the spectrum sampling statistical results. Finally,a decision tree classification model with hierarchical extraction features was formed. The research suggested that： 1By means of plant phenology,multi- temporal remote sensing data has a good effect in distinguishing pappy field and forest land; 2Pappy field and dry land staggeredly distributed in Qinhuai River basin. The pappy field is mainly located at Qinhuai River depression. The difference between pappy filed and dry land is most obious in summer,so it is the best time in extracting these information. Short wave infrared band is absorbed by the water body,We can use this band for distinguishing the pappy field and dry land; 3The farmland and non farmland can be distinguished by NDVI,but there is some building land with high vegetation coverage which is mis-classified with farm land. BSI can be used to remove these building land from farmland; 4Supervised classification has a good effect in distinguishing building land and bare land; 5The overall accuracy of Decision tree classification model is 23. 1% higher than supervised classification result.

关 键 词：Landsat8 多时相 分层提取 决策树 秦淮河 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]