基于稀疏表示理论的高光谱遥感图像的特征提取与分类

项目名称： 基于稀疏表示理论的高光谱遥感图像的特征提取与分类

项目编号： No.61271435

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 吕科

作者单位： 中国科学院大学

项目金额： 86万元

中文摘要： 高光谱遥感的特点是谱分辨率的提高，但其高数据维给图像进一步处理带来了困难。由于高光谱遥感图像的数据量大，维数高，直接对图像进行处理，算法的复杂度非常高，对计算机的硬件性能也是一个挑战。针对目前多光谱遥感图像的不足和应用局限性，围绕高光谱遥感图像，提出一系列新的解决思路，研究一类新的高光谱图像特征提取和分类理论与方法，为科学研究和工程领域中图像识别研究提供一种新的实验手段，并进一步推动遥感图像处理的发展。 针对高光谱图像的特点，本项目采用基于稀疏表示的理论框架，研究高光谱图像的特征抽取和分类的关键技术。主要研究内容集中在三个方面：基于稀疏特征的流形学习特征提取算法研究；高分辨率影像结构特征提取与多尺度处理研究；基于遗传优化机制的支持向量机分类算法研究。

中文关键词： 稀疏表示；高光谱图像；模式识别；特征提取；分类

英文摘要： Hyperspectral remote sensing image has high spectrum resolution, while the high dimension of hyperspectral image is big problem for further processing. Because of large data quantity and high dimension, processing hyperspectral image directly has high algorithm complexity and is a chanllenge for computer hardware performance. Aim at the shortages and application limitations of multispectral remote sensing image, several new solve methods are proposed for hyperspectral image processing, moreover, some novel feature extraction and classification theories and approaches of hyperspectral image are discussed. These works provide a new experimental laboratory facility for image recognition in scientific research and engineering realms, and further promote the development of remote sensing image processing. According to the character of hyperspectral image, the key technologies of feature extraction and classification of hyperspectral image are discussed based on sparse representation theory. In this project, the research contents consist of three aspects as follows: manifold learning feature extraction algorithm research based on sparse feature; high resolution image structure feature extraction and multi-scale processing research; support vector machine classification algorithm research based on genetic optimization

英文关键词： Sparse Representation；Hyperspectral Image；Pattern Recognition；Feature Extraction；Classification