基于多层特征上下文编码网络的遥感图像场景分类
投稿时间:2020-10-09  修订日期:2020-11-05  点此下载全文
引用本文:
摘要点击次数: 78
全文下载次数: 0
作者单位E-mail
李若瑶 复旦大学 电磁波信息科学教育部重点实验室 18210720036@fudan.edu.cn 
王斌 复旦大学 电磁波信息科学教育部重点实验室 wangbin@fudan.edu.cn 
张铂 复旦大学 电磁波信息科学教育部重点实验室  
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
中文摘要:遥感图像场景分类问题是目前遥感图像处理领域中的研究热点之一。卷积神经网络(Convolutional Neural Networks, CNNs)具有强的特征提取能力,已被广泛应用于遥感图像场景分类中。然而,目前的方法并没有充分考虑并利用CNN不同层间的互补信息和遥感图像的空间上下文信息,导致其相应的分类精度有待提高。针对上述问题,提议一种多层特征上下文编码网络,并将其用于解决遥感图像场景分类问题。所提议网络由两部分组成:1)密集连接的主干网络;2)多尺度上下文编码模块。前者用于融合CNN不同层的特征信息,后者用于对蕴含在多层特征中的空间上下文信息进行编码利用。在两个大规模遥感图像数据集上的实验结果表明,与现有的遥感图像场景分类方法相比,所提出的网络框架取得了显著的分类精度提升。
中文关键词:遥感图像  场景分类  卷积神经网络  多层特征上下文编码  空间上下文信息
 
Remote sensing image scene classification based on multilayer feature context encoding network
Abstract:Remote sensing image scene classification is one of the current hot topics in the field of remote sensing image processing. Since convolutional neural networks (CNNs) have powerful feature extraction capabilities, they have been widely applied in remote sensing image scene classification. However, the current methods have not fully considered and utilized the complementary information between different layers of CNN and the spatial context information of remote sensing images, resulting in that the corresponding classification accuracy needs to be improved. In order to address these issues, a multilayer feature context encoding (MFCE) network is proposed and utilized to solve the problem of scene classification for remote sensing images. The proposed network is composed of two parts: 1) A densely connected backbone; 2) A multiscale context encoding (MCE) module. The former is adopted to fuse the feature information of different layers of CNN, and the latter is utilized to encode and exploit the spatial context information that resides in the multilayer features. Experimental results on two large-scale remote sensing image datasets demonstrate that compared with the existing remote sensing image scene classification methods, the proposed network framework can achieve a significant gain in classification accuracy.
keywords:Remote sensing images, scene classification, convolutional neural network (CNN), multilayer feature context encoding (MFCE), spatial context information.
  HTML  查看/发表评论  下载PDF阅读器

版权所有:《红外与毫米波学报》编辑部