基于鲁棒估计的遥感图像融合方法  

A Robust Estimation Method for Remote Sensing Image Fusion

作　　者：夏晓天[1] 王斌[1,2] 张立明[1] 

机构地区：[1]复旦大学电子工程系,上海200433 [2]复旦大学波散射与遥感信息重点实验室,上海200433

出　　处：《复旦学报（自然科学版）》2013年第3期347-355,370,F0003,共11页Journal of Fudan University：Natural Science

基　　金：国家自然科学基金资助项目(61071134);高等学校博士学科点专项科研基金资助项目(20110071110018)

摘　　要：提出一种基于鲁棒估计的遥感图像融合方法.该方法首先建立了高分辨率的多光谱图像到低分辨率的多光谱图像和高分辨率的全色图像之间的观测模型,然后在最大后验概率框架下引入鲁棒估计以增强估计的鲁棒性,最后利用阶段非凸和逐次超松弛方法实现了低分辨率的多光谱图像和高分辨率的全色图像之间的融合.鲁棒估计的引入,大大减小了观测噪声对融合结果的影响,而且省去了目标函数中的正则项,使得融合过程更加简单方便.以QuickBird卫星数据为例的实验结果表明,与其他几种常见方法相比,本方法不仅能够提高多光谱图像的空间分辨率,对光谱信息的保持也具有更好的效果.A remote sensing image fusion algorithm for multispectral and panchromatic images based on robust estimation is proposed.Firstly,the observation model from high-resolution multispectral images to low-resolution multispectral images and high-resolution panchromatic image is constructed.Then robust estimation is introduced into the maximum a posteriori（MAP） framework to enhance the robustness.Finally,the graduate non-convexity and the simultaneous over-relaxation approaches are used to fuse the low-resolution multispectral images and the high-resolution panchromatic image.By introducing robust estimation,the observation noise effect on the estimation results is reduced greatly.Moreover,the regularization term is removed to make the fusion process easier and more convenient.The proposed algorithm is tested on QuickBird images.The experimental results show that compared to several common methods,the proposed algorithm not only improves spatial resolution of the multispectral images but also keeps the spectral information better.

关 键 词：遥感图像融合 最大后验概率 鲁棒估计 阶段非凸 逐次超松弛 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]