面向高光谱图像处理的函数型数据学习方法研究

项目名称： 面向高光谱图像处理的函数型数据学习方法研究

项目编号： No.61472155

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 李红

作者单位： 华中科技大学

项目金额： 78万元

中文摘要： 高光谱图像数据在提供丰富光谱信息的同时，所具有的高维、强相关、冗余性等特征，给传统处理方法带来了极大困难。本项目拟在对函数型数据分析理论及方法特性探索基础上，通过将高光谱图像像元表示为函数型数据，充分利用高光谱图像丰富的光谱信息，同时有效融合蕴含在按顺序排列的波长内部的信息和相应的光谱信息，更好地揭示高光谱图像数据的内在本质特征；结合信息理论学习，构造高效的函数型数据极小化误差信息熵方法，处理高光谱图像中普遍存在非高斯分布噪声问题；分析各类型基函数特点，建立有效的高光谱图像像元的函数型数据表示模型；充分利用函数型数据的本质特征，揭示像元光谱曲线的内在特性，建立混合像元分解模型；探索适合于高光谱图像分类的函数相似性度量及分类器，设计鲁棒的函数型数据特征提取及分类方法，构建高光谱图像处理方法与应用框架。函数型数据分析理论及其分析方法的引入将为高光谱图像处理模式的丰富及完善开拓出新的广袤空间。

中文关键词： 函数型数据分析；高光谱图像；极小化误差信息熵；函数型数据特征提取；混合像元分解

英文摘要： Hyperspectral image data provides abundant spectral information, at the same time, with the high dimensionality, strong correlation and high redundancy, etc, brings great difficulties to the traditional processing methods. This project plans to base on exploring the theory and methods of functional data analysis, by means of representing pixels of hyperspectral images as functional data, make full use of abundant spectral information, and effectively incorporate the information that is inherent in wavelength order and corresponding spectral information, to better reveal the essence structural characteristics of the hyperspectral image data; combine information theoretical learning, construct efficient minimum error entropy method of functional data, solve the problem that non-Gaussian distribution noise often exists in hyperspectral images; analyze the characteristics of the various types of basis functions, build functional data representation models for pixels of hyperspectral images; make full use of the essence characteristics of functional data, reveal inherent property of sprectral curves of the pixels, build spectral unmixing models; explore functional similarity measures and classifiers suited to hyperspectral image classification, design robust methods for functional data features extraction and classification，construct framework for the methods and applications of hyperspectral image processing. The introduction of the theory and analysis methods of functional data analysis would greatly contribute to develop a vast new world regarding to the abundance and perfection of hyperspectral image processing ideology.

英文关键词： functional data analysis;hyperspectral image;minimum error entropy;functional data feature extraction;pixel unmixing