高分辨率SAR图像自动分割的连续多标记凸松弛方法研究

项目名称： 高分辨率SAR图像自动分割的连续多标记凸松弛方法研究

项目编号： No.61463005

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 方江雄

作者单位： 东华理工大学

项目金额： 46万元

中文摘要： 高分辨率SAR图像解译技术在生产生活中已得到了广泛的应用。SAR图像分割是SAR图像自动解译的基础和前提，其分割质量的好坏直接影响到目标解译等工作。然而，现有的多相图像分割模型，分割高分辨率SAR图像仍存在乘性斑点噪声污染、非凸能量泛函的局部极优、预先未知的目标区域数和计算效率低等问题。本项目针对此类问题，提出高分辨率SAR图像自动分割的连续多标记凸松弛方法。研究内容如下：(1)研究连续多标记凸松弛SAR图像分割模型。建立基于贝叶斯理论的连续多标记分割框架，开发适用于计算该分割模型的连续多标记凸松弛方法，并对分割方法进行性能评估。(2)研究基于非监督先验的连续多标记凸松弛分割模型。在连续多标记凸松弛分割模型基础上，通过分析非监督分割模型，进一步建立基于非监督先验的连续多标记分割框架，开发适用于计算该分割模型的连续多标记凸松弛方法，从而更好地评价分割算法的性能。

中文关键词： 高分辨率SAR图像；自动图像分割；连续多标记；凸松弛方法

英文摘要： Interpretation of high resolution SAR image has been widely applied in our daily life.SAR image segmentation, which is a basic part of automatic interpretation of SAR images, whose segmentation performance directly affects the works such as the target interpretation. However, there are some problems for the existing multiphase SAR image segmentation models, such as multiplicative speckle noise, the unconvex and local minimum of the energy functional, the unknown number of segmented regions and low computational efficiency. To solve these problems, we propose the continuous multi-label convex relaxation approach for the automatic segmentation of high resolution SAR images. The research includes as follows: (1) A continuous multi-label convex relaxation formulation for the high resolution SAR image segmentation is proposed. Firstly, the continuous multi-label segmentation framework based on Bayesian theory is established. Then, the continous multi-label convex relaxation method applied to solve the segmentation model is developed. Finally, the performance for the segmentation method is evaluated. (2)A continuous multi-label convex relaxation formulation with unsupervised prior for the high SAR image segmentation is proposed. By analyzing unsupervised segmentation model based on the multi-label convex relaxation formulation, the continuous multi-label segmentation framework and the continous multi-label convex relaxation method are established, and the performance for the segmentation method is evaluated.

英文关键词： High Resolution SAR Images;Automatic Image Segmentation;Continuous Multi-label;Convex Relaxation Approach