Hyperspectral Pansharpening Using Noisy Panchromatic Image
Hyperspectral Pansharpening Using Noisy Panchromatic Image
复制标题
DOI：
10.23919/apsipa.2018.8659523
复制
发表时间：
2018-11
期刊：
影响因子：
--
通讯作者：
Saori Takeyama;Shunsuke Ono;I. Kumazawa
中科院分区：
文献类型：
--
作者：
Saori Takeyama;Shunsuke Ono;I. Kumazawa
获取高分辨率的高光谱图像是非常困难的。为了解决这个问题，高光谱全色锐化技术得到了广泛的研究。这些技术从一对观察到的低分辨率HS图像（低HS图像）和观察到的高分辨率全色（观察到的PAN）图像估计高空间和光谱分辨率的HS图像（高HS图像）。给定的HS和PAN图像往往含有噪声，但现有的方法大多不会考虑它，使结果在这样的情况下的伪影，噪声和频谱失真。为了解决这个问题，我们提出了一种新的高光谱泛锐化方法，同时考虑噪声在给定的HS和PAN图像。我们的方法不仅估计高HS图像，而且同时估计干净的PAN图像，从而获得高质量和鲁棒的估计。该方法有效地利用了观测信息和先验知识，并将其归结为一个非光滑凸优化问题，通过原-对偶分裂方法有效地求解.我们的实验表明，我们的方法比现有的高光谱泛锐化方法的优势。
Capturing high-resolution hyperspectral (HS) images is very difficult. To solve this problem, hyperspectral pansharpening techniques have been widely studied. These techniques estimate an HS image of high spatial and spectral resolution (high HS image) from a pair of an observed low resolution HS image (low HS image) and an observed high resolution panchromatic (observed PAN) image. Given HS and PAN images often contain noise, but most of the existing methods would not consider it, so that the results have artifacts, noise and spectral distortion in such a situation. To tackle this issue, we propose a new hyperspectral pansharpening method considering noise in both given HS and PAN images. Our method estimates not only a high HS image but also a clean PAN image simultaneously, leading to high quality and robust estimation. The proposed method effectively exploits observed information and a-priori knowledge, and it is reduced to a nonsmooth convex optimization problem, which is efficiently solved by a primal-dual splitting method. Our experiments demonstrate the advantages of our method over existing hyperspectral pansharpening methods.