空间优化的多端元混合像元分解研究

Spectral mixture analysis (SMA), which break through the spatial-resolution limitation of imaging spectrometer by signal processing techniques for high-resolution remote sensing applications, is one of the key problems in hyperspectral quantitative analysis. Based on the previous achievements in SMA of our research team, this project is carried out to analyze the influence of the spectral variability in SMA and try to alleviate spectral variability caused by inhomogeneous brightness for better performance of SMA. As a result, the model of spectral variability can be established by utilizing several spectral signatures or several distributions to represent an endmember, in which the local spectral variability is distinguished by spatial information. Correspondingly, a novel Multi-Endmember/Multi-Distribution Mixture Model (MEMM/MDMM) will be established and spatial optimized multi-endmember/multi-distribution mixture analysis algorithms will be proposed to alleviate spectral variability for better performance of SMA. In addition, the spatial distribution of pixels is also utilized to construct constraints for abundance. As a result, an Abundance-Spatial constrained Nonnegative Matrix Factorization Model (AS-NMFM) will be established for mixed pixels and Abundance-Spatial constrained mixed pixel blind unmixing algorithms will be proposed to achieve better performance of SMA by utilizing spatial information to alleviate spectral variability. Finally, a hyperspectral experimental system is established by utilizing the 220 bands imaging spectrometer in our laboratory and a novel performance testing system is constructed by utilizing different kind of hyperspectral images to verify the performan of the algorithms proposed in this project. This program is of great potentials in the applications such as deep-space exploration, environment projection, natural disaster detection and evaluation, National defense and military, and etc.

高光谱混合像元分解通过信号处理手段突破遥感器物理分辨率的限制达到高分辨的目的，是高光谱遥感定量分析应用的关键问题之一。为了减小光照不均匀性引起的光谱变化对混合像元分解精度的影响，该项目以课题组在混合像元分解方面的研究成果为基础，分析光谱变化对混合像元分解的影响规律，建立端元的光谱变化模型和混合像元的多端元（多分布）混合模型，使用空间信息标识端元的局部光谱变化，设计空间优化的多端元（多分布）混合像元分解算法；依据像元的空间相关性，建立丰度空间约束的非负矩阵分解模型并设计相应的混合像元盲分解算法，使用空间信息减小光谱变化对混合像元分解的影响；使用实验室装备的220波段成像光谱仪构建实验系统，设计基于多源高光谱数据的混合像元分解性能验证体系，对空间优化的多端元混合像元分解算法进行验证。其研究成果在深空探测、环境保护，精细农业、自然灾害的检测与评估、军事国防等领域有重要的学术价值和广泛的应用前景。

高光谱混合像元分解通过信号处理手段突破遥感器物理分辨率的限制达到高分辨的目的，是高光谱遥感定量分析应用的关键问题之一。为了减小光照不均匀性引起的光谱变化对混合像元分解精度的影响，该项目使用图像端元库对端元的光谱变化进行建模，建立空间优化的多端元混合模型，设计混合像元稀疏分解算法；依据像元的空间相关性，建立空间邻域保持的非负矩阵分解模型，设计基于Hopfield神经网络的混合像元非监督分解算法，使用空间信息减小光谱变化对混合像元分解的影响；基于GPU平台进行混合像元高速分解研究，并将项目成果应用于高光谱图像分类，空间分辨率提升以及视频总结研究，取得一系列标志性成果。项目的研究成果在深空探测、环境保护，精细农业、自然灾害的检测与评估、军事国防等领域有重要的学术价值和广泛的应用前景。