中文摘要：

以烟台市为研究区域，首先分析了居民地的影像特征，研究了居民地和背景地物在TM各波段上的光谱特征及可分性。进一步分析各地物的谱间结构特征，发现居民地和道路的TM3、TM5的光谱值比较接近。这一特征可用TM5-TM3〈T1来表示。通过试验，当T1取值为5时，可以将其提取出来，利用形状指数的差异去除道路，得到居民地提取结果。最后分析建立了基于光谱特征和形状特征的简单决策树模型，对研究区域居民地信息进行了提取，并对结果进行了精度评价。结果表明，该方法可以将居民地信息提取出来，并且效果较好，其提取精度与通常的监督分类方法相比有了很大的提高。

英文摘要：

In this paper, taking Yantai city as a case study area, the extraction of residential information from TM images is discussed, Firstly, the characteristics of residential areas on the image are studied. Secondly, spectral characteristics of residential information and other land-use types are analyzed to find the possibility of extracting residential information from the background according to spectral characteristics. Thirdly, the unique feature which can distinguish residential area and roads from other land use types is that the TM3 and TM5 of residential area are very close by further study, It can bc represented by the formula, such as,. TMS-TM3〈T1, When T1 is given respectively with 5, the residential area and roads can bc extracted, Then the shape index of every spot is calculated and categorized choosing a threshold to remove the roads, Finally, the vector coverage of residential areas is overlaid on the original image to check effeteness of this model, and an accuracy assessment is given to the results by random samples, The results suggest that this model is simple and effective, and the accuracy of the cx traction is much higher than the supervised classification＇s.