高光谱遥感图像张量分析的理论与方法研究

相对于高光谱遥感图像提供的巨大数据量，当前的数据处理技术尤显不足。如何从高光谱遥感图像中有效地提取有用信息，如何实现图像处理与光谱分析的有机结合，近年来格外受到重视。我们考虑，高光谱遥感图像外在数据表达为三阶张量形式，而内在光谱、纹理、几何和光照特性之间存在着复杂结构关系和信息冗余，它具有不确定性、结构稀疏性、空谱结构的特点。本项目从上述特点出发，探索和发展概率统计框架下结合结构稀疏特性的张量分析新理论和新方法，为解决高光谱遥感图像实际应用中的图像恢复、降维、压缩、像元分类、混合像元分解、目标探测典型问题，提供全新的技术手段和支撑。该项目的实现，将突破目前为止普遍采用的基于矩阵或矢量的高光谱遥感图像处理框架，实现空间信息和光谱特征为整体的更有效的高光谱遥感图像处理与分析。尤其是，在目前大数据量的高光谱遥感图像仍缺乏有效处理和分析工具情况下，可为其提供一种较为通用的有效处理方法和手段。

Existing data processing techniques are incompetent in dealing with large amount of data provided by hyperspectral remote sensing images. Recently, great attention has been paid to how to extract useful information effectively from the hyperspectral remote sensing images and how to integrate image processing with spectral analysis. We consider that hyperspectral imagery is a three-order tensor from its exterior representation and there exist complex structural relationships and information redundances among intrinsic spectral, textural, geometrical and illuminating characteristics, and it also exhibits particular stochastical charactistics, structural sparsity, and spatial/spectral structure. Based on the above points, our project will explore and develop novel theories and methods of tensor analysis by combining the structural sparsity within the framework of probability and statistics. Furthermore, it will offer completely new technological means and supports to solve the practical problems in hyperspectral image processing such as image restoration, dimensionality reduction, compression, pixel classification, decomposition of mixed pixels, and target detection. Once accomplished, the project will break through the currently widely-used frame of hyperspectral image processing based on matrix or vector representation, and it can effectively realize processing hyperspectral images with spatial information and spectral features as a whole. Especially, for the current situation of lacking effective tools to deal with such large hyperspectral data set, the project will provide more effective and general techniques to solve these tasks.