融合低秩和形态学的高光谱影像特征提取  

Feature Extraction of Hyperspectral Remote Sensing Image Based on Low Rank and Morphology

作　　者：江楠 张国明[2] 王俊淑[3] 韦玉春[3] Jiang Nan;Zhang Guoming;Wang Junshu;Wei Yuchun(Ecology and Environment Administration for Pearl River Basin and South China Sea,Ministry of Ecology and Environment,Guangzhou 510611,China;Health Statistics and Information Center of Jiangsu Province,Nanjing 210008,China;Key Laboratory for Virtual Geographic Environment of Ministry of Education,Nanjing Normal University,Nanjing 210023,China)

机构地区：[1]生态环境部珠江流域南海海域生态环境监督管理局,广东广州510611 [2]江苏省卫生统计信息中心,江苏南京210008 [3]南京师范大学虚拟地理环境教育部重点实验室,江苏南京210023

出　　处：《南京师范大学学报（工程技术版）》2020年第2期52-58,共7页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：国家自然科学基金项目(41471283);江苏省自然科学基金项目(BK20171037);江苏省高校自然科学研究面上项目(17KJB420003)。

摘　　要：高光谱遥感影像具有较高的光谱分辨率,能够精细刻画地物的反射光谱,具有很高的地物分类与识别能力.但高维波段之间通常具有较高的相关性,冗余度高,为影像处理和分析带来负担.针对高光谱影像特点的特征提取和选择为有效提取信息提供了保障.提出一种融合低秩和形态学的特征提取方法(MSEMP),利用低秩来精简高光谱影像中的冗余信息,获取秩最小的光谱紧致表达,并在此基础上利用多形态多尺度结构元素提取形态学剖面,获取影像空间特征.实验对AVIRIS和ROSIS传感器的两组数据进行测试,通过MSEMP提取特征后进行分类实验,可以获得较高的分类结果,证明了低秩和形态学相结合的特征提取方法的有效性.Hyperspectral remote sensing images with high spectral resolution can describe the reflection spectrum of ground objects in detail,and represent a good ability to classify and identify the ground objects.However,there is usually a high correlation and redundancy among bands,which brings burden to image processing and analysis.Feature extraction and selection of hyperspectral images provide a guarantee for the effective information extraction.In this paper,a feature extraction method,MSEMP,which integrates low rank and morphological profiles is proposed.Low-rank is utilized to simplify the redundant information and obtain the spectral compact expression with the minimum rank of hyperspectral data.Based on low rank representation of hyperspectral image,morphological profiles are extracted by using multi-shaped and multi-scale elements.The proposed algorithm is tested on AVIRIS and ROSIS data,and experimental results show that the classification accuracy based on MSEMP is higher compared with other methods.It indicates that MSEMP is an efficient feature extraction method.

关 键 词：高光谱遥感影像 形态学 低秩 特征提取 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]