基于正则化方法的遥感图像混合像元分解

Unmixed of Mixed Pixels in Remote Sensing Imaging Based on the Regularization Method

由于传感器的分辨率的限制,在低空间分辨率遥感图像中存在着大量的混合像元。混合像元所表示的并不是单一地面物体类别的光谱反射值,而是多种类别的反射光谱的组合。混合像元的混合模型可以分为线性混合模型和非线性混合模型。线性混合模型是最常用的一种解混合方法,对于线性混合模型的求解算法进行了研究,根据最小二乘原理,提出了基于正则化方法的线性混合模型求解算法,对实际遥感TM图像进行了解混合运算,求得了端元丰度图像和伪彩色合成图像。

There are lots of mixed pixels in the low spatial resolution remote sensing images due to the limitation of the resolution of the satelite sensor. What the mixed pixel represented is not the reflectation value of a singal ground cover class,but the reflectation mixed value of multi ground cover classes. Usually,there are linear mixed model and nonlinear mixed model,and the linear mixed model is the mostly used model in factual. This paper devoted to solve the linear mixed model problem and the Least Square method. The regularization approach is used to construct the Largarange function to solve the linear mixed model,and then the real remote sensing TM image is used to test this new method. The endmember abundance images and the speudo colour images are ob-tained.