基于中等空间分辨率遥感影像的建筑用地信息提取

摘要

This paper introduces and analyses several domestic and international commonly used built -up land extracted index based on typical surface features′spectral response curves from medium spatial resolution remote sensing images ( such as Landsat TM images ) .Then we select Landsat TM image to extract built-up land information and use QuickBird and Google Earth images of the same period to assist in result validation . It turns out that , the radio resident-area index ( RRI ) has very low extraction accuracy and applicability , due to lacking radiometric correction .Secondly, the normal-ized difference building index ( NDBI ) and the difference built-up index ( DBI ) have a relatively high extraction accuracy , but being mixed with bare soil , contaminated water and other information . Thirdly, the index-based built-up index (IBI) and the enhanced index -based built-up index (EIBI) achieve the highest accuracy of 92%.EIBI intents to improve IBI but could not correct the effects of bare soil, because EIBI′s chosen weights may not be universal to achieve better results .Thus, we recommend to use IBI for the extraction built-up lands information .%本研究从中等空间分辨率遥感影像（如Landsat TM影像）的地物光谱响应曲线入手，介绍分析了国内外几种常用的建筑用地提取指数构建原理．然后选取Landsat TM 影像进行建筑用地提取实验，并用QuickBird 和Google Earth的同期影像辅以验证．实验得出，比值居民地指数RRI，由于其作者构建时并没有对影像进行辐射校正，从而影响了提取精度和模型适用性；归一化建筑指数NDBI和差值建筑覆盖指数DBI，提取精度相对较高，但是会混有裸土、污染水体等信息；指数型建筑用地指数IBI和增强的指数型建筑用地指数EIBI，提取精度最高，达到92％．虽然EIBI期望改进IBI未能很好抑制裸土信息的问题，但实际上所构建指数并没有较好的去除裸土信息，可能是其权重选择没有普适性，所以建议建筑信息提取优先使用IBI．