基于约束最小二乘的光学遥感图像融合

Pan-sharpening Algorithm Based on Constrained Least Squares of Optical Remote Sensing Images

在光学遥感图像融合方法中,最小二乘法常被用于求解多光谱图像拟合低分辨率全色图像的线性回归系数,但是回归系数常常出现负数,导致其物理意义不明确。针对这种实际情况,提出了基于约束最小二乘的低分辨率全色图像构造方法。通过IKONOS-2全色与多光谱图像的融合实验,结果表明:该方法所求得的回归系数具有明确物理意义,符合实际情况,并且与光谱响应函数法、最小二乘法相比,其融合质量基本保持一致,并且由于该方法不需要先验知识,故其实用性较强。

In the method of panchromatic （Pan） sharpening of optical multispectral images by simulating low-resolution panchromatic image using linear regression, least squares was always adopted to get the regression coefficients, which were false for minus values, between low-resolution multispectral images and low-resolution panchromatic image. To solve the problem, a new method was proposed based on constrained least squares. Validating experiment was carried out on IKONOS-2 images, and showed that resultant regression coefficients were more acceptable for their physical meanings, and that the discribed method performed comparable with spectral response function or least squares based methods. Furthermore, independence of prior knowledge leads to its practicable potential.