面向线性光谱混合分解的邻域像元集螺线型构建方法

高时间分辨率遥感影像在地表景观破碎区域易形成混合像元,难以发挥其高时间维度优势。现有方式多是基于线性光谱混合模型,借助邻域像元所构成的像元集合组成线性方程组,求出组分光谱值的最小二乘解,提高其空间分辨率。然而,现有方法依赖窗口形式来构建邻域像元集合,在某些区域易造成方程组无解的欠定问题。本文在分析其问题原因的基础上,引入阿基米德螺线代替传统的矩形窗口,对邻域各像元依次遍历,构建空间邻近、组分相近的邻域像元集合来解决该问题。在GlobeLand 30数据上的试验表明,螺线型构建方法对5种混合尺度上多种类型地物均具有稳定的精度,与传统窗口构建方法相比,可从构建邻域像元集合方面将总体理论精度提高2%,分解结果精度提高近1个数量级。

A Spiral-based Construction of Adjacent Pixel Sets for Linear Spectral Unmixing

The problem of spectral mixing in fragmented landscape prevents the potentiality of high temporal resolution image from land surface detail dynamic monitoring.The general way is unmixing the spectrum to components based on linear spectral mixing model,with the aid of neighborhood pixels and the least squares method.However,constructing the neighborhood pixel set by a window leads to the underdetermined problem in some areas.This paper analyses the cause of the problem,and introduces spiral to construct optimal neighborhood pixel set as a solution.Experiment on GlobeLand 30 shows that the spiral method has a good applicability for each land surface type.Compared with the traditional method,the overall accuracy can be enhanced by 2％,the accuracy of unmixed results can be improved by nearly one order of magnitude.