中文摘要：

传统的水平集c—V模型主要关注颜色特征，对纹理特征研究得较少，而在遥感图像分割过程中，纹理特征亦具有很重要的作用；并且对于不同类型的遥感目标，颜色信息和纹理信息在分割过程中所起作用亦不相同．提出一种自适应特征选择的水平集Chan—Vese模型（C—V模型），把多个颜色特征分量和纹理特征分量作为初始特征集，通过建立相似性函数的距离度量来自适应地选择特征．同时，根据不同图像特点自适应地调整不同特征分量的权重，最后把这些特征以及相应的权重融入到c—V模型中，进而对遥感图像进行分割．与以往的方法相比，本文方法不仅取得了较好的分割效果，而且所需时间消耗更少，效率更高．

英文摘要：

In recent years, remote sensing has become a main source of information for the observation of the Earth. It is of great value to extract various information from those remote sensing images, such as the distribution of different land covers and man-made targets. Image segmentation is the first, fundamental step in the analysis of remote sensing image. In this paper, the problem of image segmentation is tackled from the perspective of pattern recognition and feature selection issue is introduced and given more emphasis. An adaptive level set model with feature selection for remote sensing image segmentation is proposed. The traditional Chan-Vese model （C-- V Model） based on level set pays much attention to the color feature, but with less emphasis on texture features. In the processing of remote sensing image, sometimes texture feature is more important for the purpose of image segmentation. To solve the problem, this paper firstly utilizes the components of different color spaces and the texture features as the initial feature set. Then feature selection is performed through local similarity analysis. Meanwhile, the weights of different features are automatically adjusted according to the characteristics of different targets. The selected features, together with their weights, are introduced into the C- V model as inputs to segment the remote sensing image. Experimental results on various targets in remote sensing imageries show thatthe newly proposed algorithm not only outperforms the traditional model efficiently, but also reduces the computational time greatly.