结合空间信息的混合像元分解方法研究

Conventional spectral unmixing methods only exploit spatial information of remotely sensed imagery, processing a pixel independently of its neighboring pixels. However, both spatial and spectral information are contained in remotely sensed imagery. Recent studies have demonstrated that the integration of spatial and spectral information can improve the accuracy of unmixing results. Based on the idea of incorporating spatial information, this proposal aims to provide a suite of novel spatial-spectral unmixing methods. First, a spatial statistical preprocessing method is developed for the detection of homogenous regions prior to endmember extraction. A number of spatial statistics are used to quantify the degree of spatial autocorrelation of pixel spectra. Homogenous regions are detected and pixels that fall in such regions are used for endmember extraction. Second, spatially-interpolated endmembers are investigated for spectral unmixing in order to account for endmember variability. For each pixel, endmember spectra are obtained by spatial interpolation with pure pixels present in its spatial neighborhood. Third, a linear spatial-spectral mixture model is developed to incorporate spectral interference arising from adjacency effect in abundance estimation. An iterative optimization algorithm is developed to estimate fractional abundances of endmembers and coefficients representing the intensity of adjacency effect. With this proposal, we hope to further demonstrate the effectiveness of incorporating spatial information in spectral unmixing and stimulate broader use of spatial information.

传统混合像元分解方法仅使用遥感图像的光谱信息，对像元的处理独立于其空间邻域中的其他像元。然而，空间信息和光谱信息同时存在于遥感图像中。近年的研究表明综合利用空间信息与光谱信息对提高混合像元分解精度具有重要意义。本项目以结合空间信息为核心思想，提出基于空间统计的图像预处理方法，利用空间统计量描述像元光谱的空间自相关，获取图像同质区域用于端元提取；提出基于空间插值的端元光谱合成方法，使用混合像元邻域内的纯净像元通过空间插值合成端元光谱，以模拟端元可变性并进行光谱解混；提出线性空间-光谱混合模型，在传统线性混合模型中考虑由邻近效应带来的光谱干扰，开发迭代优化算法，同时解算端元丰度和代表邻近效应强度的系数。期望通过本项目研究，进一步验证结合空间信息的混合像元分解方法的有效性，推动空间信息的广泛使用。

混合像元分解是一项将混合像元的光谱分解为一系列端元光谱以及与之相应的丰度信息的技术。端元光谱是影像中存在的纯净地物的光谱，而每个像元的丰度信息是存在于该像元的端元在像元中所占的面积百分比。许多混合像元分解方法将图像中的像元视为独立存在，因此在解混过程中仅使用影像的光谱信息。然而，最近的研究表明图像像素间的空间自相关为混合像元分解提供了有用信息，综合使用空间信息和光谱信息可以有效地提高解混精度。围绕混合像元分解技术的三个基本环节，即端元提取、端元组合选择以及丰度估计，本项目提出了四项空间-光谱解混的新算法：基于空间统计的图像预处理方法、基于空间插值的端元光谱合成方法、线性空间-光谱混合模型和基于空间约束的端元遴选方法，并使用模拟影像和真实影像证明了所提方法的优势。本项目的实施进一步验证了结合空间信息的混合像元分解方法的有效性，有助于激发更多的空间-光谱解混研究。