中文摘要：

山区居民地空间分布信息是山区灾害应急响应与评估的重要基础资料。我国大力发展的高分辨遥感系统,为快速获取山区居民地空间分布信息提供了数据保障,对灾害应急与评估具有重要的现实意义。针对高分辨率遥感影像居民地提取结果误分对象较多的情况,本文提出一种基于语义约束的高分一号（GF-1）遥感影像山区居民地提取方法,其基本思路是：根据山区居民地独特的分布规律,制定山谷线、山坡、雪线等语义约束条件,并将其与面向对象方法相结合进行居民地提取。以覆盖四川省康定县的一景GF-1影像为例进行实验,采取语义约束下的山区居民地提取,正确率为82.91%,漏分率仅为7.86%。与不使用语义约束的面向对象分类法直接提取结果相比,正确率提高了1.2倍。实验表明,本文提出的山区居民地提取方法能够有效改善山区居区地信息提取效果。

英文摘要：

The development of high-resolution remote sensing system in China provides a way to extract accurate spatial distribution of residential areas,which is very important for emergency response and assessment for disaster in the mountain areas. However,there is always unavoidable misclassification during the residential areas,when they are extracted from high-resolution remote sensing images. In order to improve accuracy of residential areas extraction,a new method based on semantic constraints was proposed in this paper. Three semantic constraints were established according to the distribution characteristics of residential areas i. e.,valley line,slope,and snow-covering conditions. The object-oriented method was used under these semantic constraints. In this paper,a mountain area（ 1000 km~2） in Kangding county,Sichuan was selected as case study. The results showed that ratio of correct extraction was 82. 91% by the proposed method,and ratio of miss extraction was only 7. 86%. The correct extraction ratio was increased by 1. 2 times as compared with that of the conventional object-oriented method.The results of case study indicated that this method could effectively improve extraction accuracy of residential areas in mountainous areas.