项目名称: 基于深度流形学习的高光谱数据非线性特征提取方法研究
项目编号: No.61301206
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 无线电电子学、电信技术
项目作者: 陈雨时
作者单位: 哈尔滨工业大学
项目金额: 23万元
中文摘要: 高光谱遥感能够获取目标丰富的光谱信息,但这是以高数据维数作为代价的。针对高光谱数据的高数据维,解决这一问题的主流方法是特征提取。理论分析及具体实验均证明高光谱数据蕴含着相当程度非线性特性,而传统的线性特征提取方法忽略了非线性特性。因此,本研究针对高光谱高维数据中固有的非线性,以高光谱数据应用为导向。在深度学习的框架下,以流形学习为手段,对高光谱数据的非线性特征提取问题进行探索性和可应用性的研究。研究内容集中在以下三个方面:一、高光谱数据的非线性分析及其度量;二、流形学习理论及其非线性特征提取能力分析;三、基于深度流行学习的高光谱数据非线性特征提取。研究算法将形成高光谱非线性特征提取软件包,促进高光谱遥感在分类、检测、识别等多个领域内的实际应用,使高光谱数据得到充分、有效的利用。
中文关键词: 高光谱遥感;特征提取;深度学习;分类;
英文摘要: Hyperspectral remote sensing can get abundant spectral information of the target, but it has to pay the price of high dimensionality. To deal with the high dimensionality of hyperspectral data, the mainstream solution is feature extraction. Theoretical analysis and experiment results show than there are nonlinear characteristics in hyperspectral data, while the traditional feature extraction method ignore the nonlinear characteristics. The research will focus on the nonlinear characteristics of hyperspectral data, and we use the application of hyperspectral data as a guide. Under the framework of manifold learning, we use deep learning as mathematical tool to deal with problem of hyperspectral data nonlinear feature extraction. The research will focus on three parts: Analysis and measument of hyperspectral data nonlinear characteristics; manifold learning and it's capability of extract hyperspectral data nonlinear characteristics; deep manifold learning based hyperspectral data nonlinear feature extraction method. The research algorithms will form a hyperspectral data nonlinear feature extraction software package, and it will promote the real application of hyperspectral remote sensing in many research areas such as classification detection and recognition.
英文关键词: Hyperspectral remote sensing;feature extraction;deep learning;classification;