高光谱图像多维稀疏字典构造及应用研究

项目名称： 高光谱图像多维稀疏字典构造及应用研究

项目编号： No.41471278

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 天文学、地球科学

项目作者： 尹继豪

作者单位： 北京航空航天大学

项目金额： 90万元

中文摘要： 高光谱图像具有图谱合一的特性，能够精细刻画地物光谱，为地物特征提取、分类、探测等技术的发展提供了大量有价值信息。但由于其数据量大、冗余度高，以致于高光谱成像、传输及数据处理速度慢，并在传输过程中容易丢失信息。本课题针对以上问题，结合新型信号分析与处理理论，探索高光谱图像的本质特征，加深对空间维与光谱维关系的认知，深入研究两者之间的稀疏特性，形成集数据分析、多维稀疏字典构造、混合范数重构为一体的高光谱数据处理理论体系，并对该体系展开地学应用研究。具体说来，首先，引入时间序列分析理论，并结合传统的信号处理理论，深入挖掘高光谱图像空间维与光谱维的联合特征，研究体现高光谱图像本质特征的稀疏表示；其次，根据获得的空间与光谱维的联合特征，建立多维稀疏字典；然后，基于多维稀疏字典和压缩感知理论，研究鲁棒性强的混合范数重构算法，获取重构图像；最后，将提出的算法体系应用于典型地物探测中，以验证其高效性。

中文关键词： 稀疏表示；压缩感知；高光谱遥感；目标探测；特征提取

英文摘要： Hyperspectral image can accurately depict the spectrum of the object, providing large quantities of useful information for feature extraction, classification, detection, etc. However, its huge data size, high redundancy make it hard to imaging, transmission, and processing, and its information is easy to lose on the transfer process. To solve the problems above, our group combine the new signal analysis and processing theory, explore the intrinsic characteristics of hyperspectral images, and deepen the cognition of the relationship between spatial and spectral dimension of hyperspectral image, and study the sparse characteristics between spatial and spectral dimension. Based on the research above, we will form a theoretical system which contains data analysis, multi-dimension sparse dictionary construct and hybrid norm reconstruction algorithms, and then do research on the application of geoscience. We mainly study as follows: Firstly, we will introduce the analysis of time series into the study and combine it with traditional signal processing theory, mining the union feature of the spatial and spectral dimension in hyperspectral image, study the sparse representation which can represent the intrinsic characteristics of hyperspectral image. Secondly, we will construct the multi-dimension sparse dictionary on the basis of the union feature, and then research the hybrid norm algorithms with strong robustness according to the multi-dimension sparse dictionary and compressive sensing theory. Finally, we will apply this algorithm system to typical object detection to prove its high efficiency.

英文关键词： Spare analysis;Compressed sensing;Hyperspectral remote sensing;Object detection;Feature extraction