中文摘要：

针对遥感影像对多种地物进行分割时分割结果不够精确、适用性较差、效率较低的问题,提出了一种使用多星形先验与图割算法相结合的方法,实现遥感影像各类地物高效分割。算法利用均值漂移算法进行预分割,在图割算法中引入多星形先验信息,并基于前景背景交替迭代思想,实现多种地物一次分割。通过引入多星形先验,利用形状信息提高分割的准确性;利用一次交互获取多种地物的种子点,通过前景与背景种子点的交替迭代,改进图割理论的前景背景分割,提高分割效率。定量分析与实验结果表明：该算法对分割人工建筑区、植被、道路及水系更具准确性、高效性及普适性。

英文摘要：

To solve the problem that segmentation of multiple objects in remote sensing image is lack of accuracy, applicability and efficiency, this paper proposes an algorithm based on Multiple Star Prior and Graph Cuts and realizes segmentation of types of remote sensing features. This algorithm uses the Mean-Shift to execute pre-segmentation,introduces the energy multiple star information in Graph Cuts, and is on the basis of background-foreground alternating iteration to segment all features at a time. By introducing a star a priori, using shape information the segmentation accuracy is improved;an interaction is used to obtain a variety of ground seeds,improving the Graph Cuts through foreground and background alternate iteration of the seeds. Quantitative analysis and experimental results show that the algorithm is more precise, efficient and applicable for the segmentation of constructed built-up areas,vegetation, roads and water.