设为首页 联系我们 加入收藏

当前位置: 网站首页 期刊分类目录 2019第3期 正文

深度迁移学习在高光谱遥感图像分类中的研究现状与展望

作者:时间:2019-05-13点击数:

PDF全文下载:  201903001.pdf

 

文章编号: 16726987201903000111 DOI 10.16351/j.16726987.2019.03.001


刘小波13 尹旭13* 刘海波2 汪敏13 颜丙云2

 

(1.中国地质大学(武汉) 自动化学院,湖北 武汉 4300742.青岛科技大学 自动化与电子工程学院,山东 青岛 2660613.复杂系统先进控制与智能自动化湖北省重点实验室,湖北 武汉 430074)

 

摘要: 高光谱遥感通过利用许多窄电磁波波段获取包含丰富的空间、辐射和光谱信息,在对地观测研究领域扮演着重要角色。随着深度学习的迅速发展,深度神经网络及深度森林等算法在高光谱遥感图像分类任务中得到广泛应用,但同时也产生了一系列困难,如对训练样本数量需求高、模型训练耗时以及分类代价大等问题。将深度学习与迁移学习结合,能够有效解决上述问题,在高光谱遥感图像分类领域得到初步应用。本工作首先介绍高光谱遥感图像分类的相关背景,之后介绍深度学习在高光谱遥感图像分类中的应用,并指出其具有的优势与不足,最后介绍深度迁移学习在高光谱遥感图像分类中的应用,并对当前研究存在的问题进行总结与展望。

 

关键词: 高光谱; 遥感; 深度学习; 深度迁移学习; 图像分类

 

中图分类号: TP 751.1文献标志码: A

 

引用格式: 刘小波, 尹旭, 刘海波, . 深度迁移学习在高光谱遥感图像分类中的研究现状与展望\[J\]. 青岛科技大学学报(自然科学版), 2019 403): 111.

LIU Xiaobo, YIN Xu, LIU Haibo,  et al. Classification of hyperspectral remote sensing image based on deep transfer learning: A review\[J\]. Journal of Qingdao University of Science and TechnologyNatural Science Edition), 2019, 40(3) 111.

 

Classification of Hyperspectral Remote Sensing Image Based on Deep Transfer Learning: A Review

 

LIU Xiaobo13, YIN Xu13, LIU Haibo2,  WANG Min13, YAN Bingyun2

 

(1.School of Automation, China University of Geosciences, Wuhan 430074, China; 2.College of Automation and Electronic Engineering, Qingdao University of Science and Technology, Qingdao 266061, China; 3.Hubei Key Laboratory of Advanced Control and Intelligent Automation for Complex Systems, Wuhan 430074, China)

 

Abstract: Hyperspectral remote sensing (HRS) technology plays an important role in the field of earth observation by using many narrow electromagnetic wave bands to detect ground information containing abundant space, radiation and spectrum. With the rapid development of deep learning, algorithms such as deep neural networks (DNNs) and deep forest have been widely adopted in the classification tasks of HRS image, but at the same time they have brought a series of difficulties, such as the demand for a large number of training samples, the timeconsuming training process, and the high training cost. Combining the advantages of deep learning and transfer learning, effectively solves the above problems, which has already highlighted excellent performance in HRS image classification tasks. This paper first introduces the background of HRS image classification, and then introduces the application of deep learning in HRS image classification, and compares its advantages and disadvantages. Finally, it introduces the application of deep transfer learning in the classification of HRS images, and forecasts the problems existing in the current research.

 

Key words: hyperspectral; remote sensing; deep learning; deep transfer learning; image classification

 

收稿日期:  20181227

基金项目: 国家自然科学基金项目(61603355,61873249); 中国地质大学(武汉)中央高校基本科研业务费资助项目 (CUGL17022); 湖北省自然科学基金面上项目 (2018CFB528).

作者简介: 刘小波(1978) , , 副教授.*通信联系人.

 

Copyright © 2011-2017 青岛科技大学学报 (自然科学版)