基于灰色理论的SAR图像分割及其效果评价方法研究

项目名称： 基于灰色理论的SAR图像分割及其效果评价方法研究

项目编号： No.60803088

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 机械、仪表工业

项目作者： 马苗

作者单位： 陕西师范大学

项目金额： 18万元

中文摘要： SAR图像分割是SAR图像分析的关键步骤，也是SAR图像解译中尚未完全解决的重要问题。现有SAR图像分割方法的本质多是基于概率统计和模糊数学，其实现依赖于大样本、特定概率分布或隶属度等先验知识，这在实际应用中很难完全满足。本项目以小样本数据分析与建模为特色的灰色理论为工具，深入研究了斑点噪声特性，噪声抑制、SAR图像分割和分割性能评价方法。包括：①#20998;析了斑点噪声的特性，利用Gaussian-Hermite矩和直方图特征，提出确定图像中的斑点噪声强度的方法和不同类型噪声的灰色识别方法；在此基础上，探讨了基于灰色关联分析的SAR图像增强方法、改进算法，以及相似性灰关联的强噪声图像增强方法；②#23450;义了灰熵模型，将多尺度变换、群体智能优化算法、熵与形态学等知识相结合，提出多种快速高效的SAR图像分割方法和线与边缘的检测方法。③#30740;究了SAR图像分割效果评价中的小样本数据序列形成方法，提出基于灰色聚类分析和灰色关联分析的分割效果评价模型，并形成了相关评价方法。

中文关键词： 灰色理论；灰色系统；图像分割；效果评价

英文摘要： SAR image segmentation not only is a key step of SAR image analysis, but also is an open problem of SAR image understanding. Most of existing methods are based on probability statistic and fuzzy math in essence, which depend on a large number of samples and priori knowledge such as some special distribution or membership degree. It is difficult for all kinds of application to completely meet with all these demands. Taking grey theory as tool, this project research speckle noise, noise suppression,SAR image segmentation and its performance evaluation. The main contributions consists of three parts: ①after the property of speckle noise was deeply analyzed, two evaluation methods of speckle intensity are proposed, respectively based on Gaussian-Hermite Moment and histogram characteristic,and a type recognition of noise is suggested via grey relational degree. Moreover, based on grey relational analysis, several noise suppression methods are presented for SAR images and strong noise pollution images.②a grey entropy model is defined. a bench of efficient methods on SAR image segmentation and Line & edge detection are proposed, which Combine multiscale transform, swarm intelligent based optimization algorithm, entropy and morphology.③he method to form data sequence with few samples in SAR image segmentation performance assessment is studied, and some comprehensive evaluation models and methods are developed via grey clustering analysis and grey relational analysis.

英文关键词： grey theory;grey system;image segmentation;performance evaluation