吕亚飞1, 3,,,
张筱晗1, 4,
刘瑜1,
崔晨浩5,
顾祥岐1
1.海军航空大学信息融合研究所 烟台 264000
2.清华大学电子工程系 北京 100084
3.91977部队 北京 100089
4.61646部队 北京 100089
5.32144部队 渭南 714000
基金项目:国家自然科学基金(61790550, 61790554, 61531020, 61671463)
详细信息
作者简介:徐从安:男,1987年生,博士,研究方向为遥感图像智能处理、多目标跟踪
吕亚飞:男,1992年生,博士,研究方向为遥感图像智能处理、跨模态检索
张筱晗:女,1992年生,博士,研究方向为遥感图像智能处理、目标检测
刘瑜:男,1986年生,副教授,研究方向为智能数据处理
崔晨浩:男,1991年生,研究方向为雷达数据处理
顾祥岐:男,1995年生,博士生,研究方向为雷达数据处理、信息融合
通讯作者:吕亚飞 YFei_Lv@163.com, xcatougao@163.com
中图分类号:TP751.1; TP183计量
文章访问数:594
HTML全文浏览量:186
PDF下载量:75
被引次数:0
出版历程
收稿日期:2020-07-10
修回日期:2020-12-07
网络出版日期:2020-12-15
刊出日期:2021-03-22
A Discriminative Feature Representation Method Based on Dual Attention Mechanism for Remote Sensing Image Scene Classification
Cong'an XU1, 2,Yafei Lü1, 3,,,
Xiaohan ZHANG1, 4,
Yu LIU1,
Chenhao CUI5,
Xiangqi GU1
1. Information Fusion Institute, Naval Aviation University, Yantai 264000, China
2. Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
3. 91977 Troop, Beijing 100089, China
4. 61646 Troop, Beijing 100089, China
5. 32144 Troop, Weinan, 714000, China
Funds:The National Natural Science Foundation of China (61790550, 61790554, 61531020, 61671463)
摘要
摘要:针对遥感图像场景分类面临的类内差异性大、类间相似性高导致的部分场景出现分类混淆的问题,该文提出了一种基于双重注意力机制的强鉴别性特征表示方法。针对不同通道所代表特征的重要性程度以及不同局部区域的显著性程度不同,在卷积神经网络提取的高层特征基础上,分别设计了一个通道维和空间维注意力模块,利用循环神经网络的上下文信息提取能力,依次学习、输出不同通道和不同局部区域的重要性权重,更加关注图像中的显著性特征和显著性区域,而忽略非显著性特征和区域,以提高特征表示的鉴别能力。所提双重注意力模块可以与任意卷积神经网络相连,整个网络结构可以端到端训练。通过在两个公开数据集AID和NWPU45上进行大量的对比实验,验证了所提方法的有效性,与现有方法对比,分类准确率取得了明显的提升。
关键词:遥感图像处理/
场景分类/
注意力机制/
特征表示
Abstract:Considering the problem of low classification accuracy caused by large intra-class differences and high inter-class similarity in remote sensing image scene classification, a discriminative feature representation method based on dual attention mechanism is proposed. Due to the difference in the importance of the features contained in different channels and the significance of different local regions, the channel-wise and spatial-wise attention module are designed, based on the high-level features extracted by the Convolutional Neural Networks. Relying on the ability to extract contextual information, the Recurrent Neural Network is adopted to learn and output the importance weights of different channels and different local regions, paying more attention to the salient features and salient regions, while ignoring non-salience features and regions, to enhance the discriminative ability of feature representation. The proposed dual attention module can be connected to the last convolutional layer of any convolutional neural network, and the network structure can be trained end-to-end. Comparative experiments are conducted on the two public data sets AID and NWPU45. Compared with the existing methods, the classification accuracy has been significantly improved, and the effectiveness of the proposed method can be verified.
Key words:Remote sensing image process/
Scene classification/
Attention mechanism/
Feature representation
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=8b30aa1b-a784-40d0-a7c4-f23c72e0b94f