高光谱图像多端元处理技术研究

With respect to the current development trend and urgent requirement of processing techniques on hyperspectral imagery, this project will illustrate multiple endmembers from the two aspects: extraction and application. For the extraction, we will develop multiple endmembers cluster extraction algorithm by the classical single endmember extaction algorithm assisted with a new integrated distance criterion and SVM-based unmixing algorithm; For the application, we will reseach the four areas under a unified framework for the application of multiple endmembers: 1) Classification. We will propose a SVM-based semi-supervised classificatin algorithm based spatial-spectral information using multiple endmembers as a small-size training set and adding the spatial information; at the same time, sparse representation classification will be researched based on spatial-spectral combined multiple endmembers modeling; 2) Spectral unmixing.TWSVM-based multiple endmembers modeling of spectral unmixing will be researched to take parallel and Non parallel data distributions into consideration, with multiple endmembers information included; 3) spatial resolution improvement algorihtm. We will relate the super-resolution model to multiple endmembers and to extend the classical inversion model achive a improved spatial resolution algorhtm which will catch more eyes on classes of interest; 4) Visualization. We research the visualization method based on multiple endmembers information to meet with the requirments of keeping spectral distance and separability will be utilized to develop color display approaches with higher accuracy and more levels. The success of this project will be of great significance as well as application value to the effective exploration and utilization of hyperspectral imagery.

针对高光谱图像处理技术的现状和需求，课题拟研究多端元信息的提取和利用方法。关于多端元信息提取，利用经典单端元提取方法，辅以新型综合距离准则和SVM光谱解混技术，建立有效的多端元提取方法。关于多端元信息利用，研究统一框架下的以下四个方面：1)分类。以多端元信息作为小样本训练信息，同时挖掘空间信息，建立空谱信息结合的SVM半监督分类方法，同时研究多端元模式下结合空间信息的稀疏表达分类技术；2)混合像素分解。以TWSVM承载多端元表达模型，研究同时兼顾平行与非平行分布特性数据的混合像素分解方法；3)空间分辨率提高。研究多端元模式下的超分辨率模型，拓展经典模型反演算法，获得具有保护重点类别的特殊分辨率提高方法；4)可视化。研究整合利用多端元信息，且满足光谱距离保持特性及可分性要求的可视化方法。课题内容对于提高高光谱图像的分析效果和科学决策意义重大。

针对高光谱图像处理技术的现状和需求，课题拟研究多端元信息的提取和利用方法。关于多端元信息提取，利用经典单端元提取方法，辅以新型综合距离准则和SVM光谱解混技术，建立有效的多端元提取方法。关于多端元信息利用，研究了统一框架下的以下四个方面：1)分类。以多端元信息作为小样本训练信息，同时挖掘空间信息，建立了空谱信息结合的SVM半监督分类方法，同时研究了多端元模式下结合空间信息的稀疏表达分类技术，这些分类技术充分利用多端元信息，较之主流方法取得了很好的分类效果；2)混合像素分解。以TWSVM承载多端元表达模型，研究同时兼顾平行与非平行分布特性数据的混合像素分解方法，该方法在处理非平行结构数据解混时取得了较好的效果；3)空间分辨率提高。研究了多端元模式下的超分辨率模型，拓展了经典模型反演算法，该方法超分辨效果良好，尤其是对于感兴趣类别起到了重点保护的作用；4)可视化。研究了整合利用多端元信息，且满足光谱距离保持特性及可分性要求的可视化方法，研究取得了多层次、可满足不同用户需求的可视化显示方案。课题内容对于提高高光谱图像的分析效果和科学决策意义重大。