基于NTF和HHT的空间碎片天基高光谱探测与识别方法

项目名称： 基于NTF和HHT的空间碎片天基高光谱探测与识别方法

项目编号： No.61201310

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 张淼

作者单位： 哈尔滨工业大学

项目金额： 25万元

中文摘要： 空间碎片的存在严重威胁着在轨航天器的安全，其中难以通过地基观测的中型碎片无疑更为危险。本项目旨在对轨道上的中型碎片进行天基探测，并对其物质组成进行分类识别。为了实现准确度高、漏检率低的空间碎片探测方法，用高光谱成像仪而非可见光相机作为遥感元件，进而提出两种新的探测方法：利用单一物质成分碎片所具有的特定光谱信息，结合非负张量分解(NTF)及相关分析技术进行探测；利用多物质成分碎片发生旋转所产生的光谱信息变化，结合希尔伯特-黄变换(HHT)分析光谱序列的周期特征来进行探测。此外，根据已探明空间碎片的高光谱数据，利用基于波段加权核的支持向量机进行物质组成分类，收集空间碎片识别信息以备编目之用。本项目借助高光谱数据扩展了以往的探测、识别手段，并针对高光谱数据量大、维数高、冗余信息多的特点引入NTF、HHT 等算法，兼有数据压缩、自适应特征提取等优势，以期推动空间碎片的探测与识别向更高水平迈进。

中文关键词： 高光谱图像；模式分类；目标探测；希尔伯特黄变换；非负张量分解

英文摘要： The existence of space debris seriously threatens safe of on-orbit spacecraft, the medium-sized debris which is hard to find by ground-based observation is obviously more dangerous. The goal of project is to find medium-sized debris in an orbit by space-based detection and classify and identify the substance construction of debris. In order to accomplish high accuracy and low miss rate detection, we use hyper spectral imager instead of space-based visible light camera as sensor and come up to two new ways of detection: One way of detection is to use the particular spectrum information of component debris of single substance and combine Nonnegative Tensor Factorization (NTF) and relevant analysis technics, the other way of detection is to use variation of spectrum information due to rotation of component debris of multi-substance and combine Hilbert-Huang Transform (HHT) to analyze periodic features of spectrum sequence. Furthermore, according to hyperspectral data of detected space debris, using support vector machine based on spectral band kernels to classify constitution of substance and collect identifying information of space debris for use of catalogue. The project uses hyperspectral data to expand past methods of detection and identification and introduce some algorithms as NTF, HHT according to the featu

英文关键词： Hyperspectral Image；Pattern Classification；Target Detection；Hilbert-Huang Transform；Nonnegative Tensor Decomposition