混合像元分解在高光谱遥感领域扮演着重要角色。项目针对目前的混合像元分解方法精度差、复杂度高等特点，结合压缩感知和非负矩阵分解理论，试图提出一种新的盲混合像元分解变分模型，该模型利用一组基对数据进行压缩，再进行像元分解，以避免大数据量带来的数据高复杂度问题。在基的选取上，项目将探讨不同基，如DCT、 Gabor等的性质，以得到最佳压缩效果。考虑到端元丰度的分段常数和稀疏特性，我们将用全变分和稀疏性度量函数来约束模型。在分析模型稳定性的基础上，项目会进一步研究快速有效的迭代算法求取端元和丰度。在模型和算法的验证方面，项目采用野外实测、卫星遥感等多源数据对方法的稳定性和精确性进行考查。混合像元分解方法的研究，对于矿物勘探、地形识别、环境监控等方面都有非常重要的理论意义和应用价值。

Hyperspectral imagery unmixing plays an important role in remotely sensed image processing. In this project, a variational model for blind hyperspectral imagery unmixing based on compressive sensing (CS) and nonnegative matrix factorization (NMF) will be proposed. First, we will use some bases to generate the compressed hyperspectral data, then we will build an energy model to compute the endmembers and their abundance in a pixel directly. The model is constrained by the total variation term and a sparsity measure function to keep the piecewise constant and the sparseness of abundance. The alternative minimization algorithm will be studied in order to compute the unmixed endmembers and their abundance fastly. We will conduct extensive numerical experiments to demonstrate the feasibility and efficiency of the proposed approach, using both synthetic and real data. The method will be very valuable in mineral prospecting, terrain identification, environmental monitoring, and so on.