摘要
高光谱图像(Hyperspectral imagery,HSI)分类是高光谱遥感对地观测技术的一项重要内容,在军事及民用领域都有着重要的应用.然而,高光谱图像的高维特性、波段间高度相关性、光谱混合等使得高光谱图像分类面临巨大挑战.近年来,随着深度学习新技术的出现,基于深度学习的高光谱图像分类在方法和性能上得到了突破性的进展,为其研究提供了新的契机.本文首先介绍了高光谱图像分类的背景、研究现状及几个常用的数据集,并简要概述了几种典型的深度学习模型,最后详细介绍了当前的一些基于深度学习的高光谱图像分类方法,总结了深度学习在高光谱图像分类领域中的主要作用和存在的问题,并对未来的研究方向进行了展望.
Hyperspectral imagery（HSI） classification occupies an important place in the earth observation technology of hyperspectral remote sensing, and it is widely used in both military and civil fields. However, due to HSI s characteristics including high dimensionality in data, high correlation between spectrum and mixing in spectrum, HSI classification faces great challenges. In recent years, as new deep learning technology emerges, the HSI classification methods based on deep learning have achieved some breakthroughs in methodology and performance and provided new opportunities for the research of HSI classification. In this paper, we review the research background,actuality of HSI classification technologies and several common datasets. Then, we provide a brief overview of several typical deep learning models.Finally, we introduce some deep learning based HSI classification methods in detail, summarize the main function and existing problems of deep learning in HSI classification, and present some prospects for future work.
作者
张号逵
李映
姜晔楠
ZHANG Hao-Kui;LI Ying;JIANG Ye-Nan(Shaanxi Provincial Key Laboratory of Speech and Image Information Processing, School of Computer Science, Northwestern Polytechnical University, Xi＇an 710129)
出处
《自动化学报》
EI
CSCD
北大核心
2018年第6期961-977,共17页
Acta Automatica Sinica
基金
国家重点研发计划项目(2016YFB0502502)
预研领域基金课题(614023804016HK03002)
陕西省国际科技合作计划项目(2017KW-006)
西北工业大学博士论文创新基金(CX201816)资助~~
关键词
深度学习
高光谱图像分类
卷积神经网络
栈式自编码网络
深度置信网络
Deep learning
hyperspectral imagery （HSI） classification
convolutional neural network （CNN）
stacked autoencoder
deep belief network
