高光谱遥感图像的频域特征提取与分类研究

Classification is important to the application of hyperspectral imaging. And it is one of the most fundamental topics in the field of remote sensing. Hyperspectral remote sensing data provides much more abundant spectral information of ground objects than multispectral data, which makes the direct recognition and refined classification is possible. Frequency domain, being a commonly used transform domain, is capable of providing the important foundation for ground object recognition in hyperspectral remote sensing image. The proposal will develops a new method to extract the frequency feature set of ground object, and present a new solution to hyperspectral image classification based on frequency domain. Firstly, the characterization of ground objects in frequency domain is revealed by analysis of the frequency spectrum, power spectrum and phase spectrum of the spectral response curves of typical ground objects. Secondly, frequency feature is extracted based on the discrepancy of the characterization on different ground object. Then, the set of frequency features is chosen by the optimization method and experimental result. Thirdly, we construct the energy function based on the linear combination of frequency feature set which has been normalized and weighted. Finally, the semi-supervised classification of hyperspectral image is applied by optimizing the energy function. The research achievements will benefit hyperspectral image classification theoretically, and have great significance in the application of precision agriculture, environmental monitoring, mineral prospecting, military security, and so on.

高光谱图像分类对高光谱遥感的应用研究具有重要意义，也是当前遥感应用基础研究的热点内容之一。与多光谱遥感相比，高光谱遥感提供更为丰富的光谱信息，使得地物直接识别与精细分类成为可能。频域作为一种常用变换域，可以为高光谱遥感图像的地物识别提供重要依据。因此，本项目拟发展一套面向高光谱遥感图像的频域识别特征集提取方法，并提出一种高光谱遥感图像分类的频域解决方案。着重研究以下内容：1）通过分析典型地物光谱响应曲线的频谱、功率谱和相位谱，揭示高光谱图像中地物的频域表征规律；2）提取高光谱图像典型地物的频域特征，并利用数学优化方法和不同条件下的实验验证，生成频域识别特征集；3）基于该特征集，通过归一化和权重分配得到线性加权组合式，并构建能量函数，通过能量函数优化，实现高光谱遥感图像的半监督分类。研究成果对高光谱遥感图像分类有理论参照意义，在精细农业、环境监测、矿物勘探、军事安全等领域具有重要应用价值。

高光谱提供了比多光谱遥感更为丰富的光谱信息，进而可以进行更为精细的土地利用和地物识别应用。高光谱遥感卫星数据的分类对于高光谱遥感应用研究具有重要意义。本项目借助傅里叶变换将高光谱遥感卫星数据转换为频域的频谱信息，并由此发展了一套针对频域和高光谱遥感数据的信息提取相结合的遥感图像信息识别和分类的方案。本项目目前所完成的内容如下：1）本项目分析了典型地物光谱响应曲线的频谱、功率谱和相位谱，揭示高光谱图像中地物的频域表征规律；2）借助光谱相似性度量，验证了频域可以更好地应用于高光谱遥感图像的信息识别和分类中；3）通过一系列的频域分析手段，本项目给出了有效的高光谱遥感卫星数据的频域特征集，并有效地应用于了遥感图像分类应用中。研究成果对高光谱遥感图像分类具有重要理论参照意义，在遥感技术应用领域具有重要应用价值。