基于上下文感知和异质特征集成的SAR图像分割与评价

项目名称： 基于上下文感知和异质特征集成的SAR图像分割与评价

项目编号： No.61501352

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 余航

作者单位： 西安电子科技大学

项目金额： 19万元

中文摘要： SAR图像分割是SAR图像理解与解译的基础和关键技术之一。本项目针对大规模、复杂场景SAR图像分割和评价时难以有效利用多种图像特征的问题，构造上下文感知规则和异质特征集成方法，建立SAR图像的准确完备表示框架，以更加准确的表示SAR图像目标，并得到辨别度高和鲁棒性强的SAR图像目标相似度测度。在该表示框架下，研究基于上下文感知和异质特征集成的模糊聚类算法和区域合并算法，并设计空间域（区域合并）-特征域（模糊聚类）联合的自适应协同策略和数据交换接口，通过结合基于特征域和基于空间域两类分割算法的优势，解决现有算法面对大规模、复杂场景SAR图像时，难以同时兼顾分割准确性和分割效率的难题；提出适用范围广、精确性高的SAR图像分割结果定量评价指标，有利于改善基于人工的分割结果评价方法标准不统一、容易产生误判的现状。本项目对SAR图像理解与解译领域的应用基础研究具有重要的科学价值。

中文关键词： 上下文感知；异质特征集成；图像分割；分割评价；SAR图像处理

英文摘要： SAR image segmentation is the foundation and one of the key technologies for the understanding and interpretation of SAR images. One problem for SAR segmentation algorithms and the quantitative evaluation of segmentation results is to effectively make use of multiple features. To deal with this problem, this proposal will construct context-aware rules and design heterogeneous features ensemble strategies, which realizes the complete and accurate description of SAR images from different aspects and levels, and improves the accuracy and generalizability of the similarity measures between objects in SAR images. Novel fuzzy clustering algorithms and region merging algorithms based on context-aware and heterogeneous features ensemble will be proposed, both advantages of which will next be combined by an adaptive cooperative strategy and data exchange interface associating space domain and feature domain. This can alleviate the dilemma that traditional segmentation algorithms cannot simultaneously maintain the accuracy and the efficiency when dealing with large-scale and complex SAR images. Considering manual evaluation of SAR segmentation results is a subjective process and easy to make mistake, context-aware and heterogeneous features ensemble based quantitative validation index of segmentation results with wide suitability and high accuracy will be constructed to improve the unsubjective evaluation level of SAR image segmentation. This proposal has important scientific value for the applied basic research of the understanding and interpretation of SAR images.

英文关键词： Context-aware;Heterogeneous Features Ensemble;Image Segmentation;Segmentation Evaluation;Synthetic Aperture Radar Image Processing