基于DS证据理论的多幅遥感影像融合去噪  

Remote Sensing Image De-noising by Fusing Multiple Images Based on DS Evidence Theory

作　　者：黄冬梅[1] 朱贵鲜 张明华[1] 王丽琳[2] 苏诚[2] 

机构地区：[1]上海海洋大学信息学院,上海201306 [2]国家海洋局东海预报中心,上海200137

出　　处：《海洋科学进展》2017年第3期414-427,共14页Advances in Marine Science

基　　金：上海市科学技术委员会科研计划项目--地方院校能力建设(15590501900);极地海洋环境监测示范应用系统开发项目--极地海洋环境监测示范应用系统开发(201405031-05);国家自然科学基金项目--基于多模态深度学习的弱特征多源海洋遥感影像协同分类模型研究(41671431);国家自然科学基金青年基金项目--一种面向多模态遥感信息的质量抽样检验方案研究(41501419)

摘　　要：针对目前单幅遥感影像去噪方法会过度扼杀影像边缘信息或噪声去除不理想问题,提出了一种基于DS(Dempster-Shafer)证据理论的多幅遥感影像融合去噪方法。DS证据理论结合多源信息,能够有效处理不确定性问题,适用于噪声具有随机性和不确定性情景。本文采用Landsat8卫星获取的上海周边海域遥感数据,依据DS证据理论处理不确定问题的优势,并充分利用多幅遥感影像的有效信息,设计4个模型,即两状态高斯混合模型、两个噪声分析模型、边缘分析模型。这4个模型用以获取每个像素与噪声相关的概率,作为证据理论的4个证据,用于决策去噪方案。结果表明,本文提出的多幅遥感影像融合去噪方法在保证去除噪声的情况下能较好地保持影像边缘和纹理细节信息。The existing methods for de-noising a single remote sensing image tend to blur the image edges or give an imperfect de-noising result.To solve these problems,the de-noising method by fusing multiple remote sensing images based on DS evidence theory is proposed in this paper.The DS evidence theory is capable of utilizing multiple-source information and effectively dealing with cognitive uncertainty,and thus applicable to the cases of noise with uncertainty and randomness.Using the advantages of DS evidence theory,the paper aims to make full use of effective information of multiple remote sensing images taken by Landsat 8in the sea-area around Shanghai.Four models are implemented to determine the probability of each pixel related with the noise,including the two-state Gaussian Mixture Model,edge analysis model,and two Noise Analysis Models based on spatial correlation.These four models work as the four evidences of DS theory to support the decision-making of de-noising scheme.Experimental results show that the method keeps the details of the image edge and texture information well while the noise is removed.

关 键 词：遥感影像 DS证据理论 影像融合 影像去噪 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]