摘要
针对向神经网络模型中引入注意力模块,导致模型复杂度进一步提高的问题,该文提出了结合轻量双注意力模块的孪生卷积网络变化检测方法。通过在孪生卷积网络模型基础上,加入构建的轻量双注意力模块,达到了向模型中引入较少参数量并且提升变化检测性能的效果。在现有的两个公开数据集上,该文算法分别达到了95.2%和98.4%的总体检测精度,优于其他现有方法。
In order to tackle the problem that the introduction of attention modules would increase the complexity of the neural network,a Siamese convolution network combined with lightweight dual attention modules was proposed for change detection.The lightweight dual attention module could improve the accuracy of change detection with less parameters added to network.The overall accuracy of the proposed method reached 95.2%and 98.4%,respectively,on two public change detection datasets,and it indicated the effectiveness of our method.
作者
罗卿莉
陈相
魏钜杰
LUO Qingli;CHEN Xiang;WEI Jujie(State Key Laboratory of Precision Measuring Technology and Instruments,Tianjin University,Tianjin 300072,China;Chinese Academy of Surveying and Mapping,Beijing 100036,China)
出处
《测绘科学》
CSCD
北大核心
2022年第8期213-219,共7页
Science of Surveying and Mapping
基金
国家自然科学基金项目(41601446,41801284)
天津市自然科学基金重点项目(21JCZDJC00670)
城市轨道交通数字化建设与测评技术国家工程实验室开放课题基金项目(2021ZH04)
天津市交通运输科技发展计划项目(2020-02)
关键词
变化检测
孪生卷积神经网络
轻量双注意力模块
遥感影像
change detection
siamese convolution network
lightweight dual attention module
remote sensing image