摘要
针对从多时相高分辨率遥感影像上做建筑物变化检测出现的细节特征丢失、变化检测结果模糊的问题,该文提出了一种基于多尺度特征孪生神经网络的遥感影像建筑物变化检测算法。以孪生神经网络为基础网络,将Inception v2结构加入网络特征提取层中,获得遥感影像多尺度特征,并对其进行多特征融合,更好地还原建筑目标的细节信息。与全卷积孪生神经网络网络相比,在WHU和LEVIR-CD建筑变化数据集下的实验结果表明,overall accuracy、F1-score和Kappa系数指标都有所提高,本文的多尺度孪生神经网络方法更利于建筑物变化检测,并获得了良好的测试结果。
Aiming at the problems of loss of detailed features and fuzzy change detection results in building change detection from multi-temporal high-resolution remote sensing images,this paper proposes a remote sensing image building change detection algorithm based on multi-scale feature Siamese network.Based on the Siamese network,the Inception v2 structure is added to the network feature extraction layer to obtain multi-scale features of remote sensing images,and multi-feature fusion is performed to better restore the detailed information of the architectural target.Compared with the fully convolutional Siamese network,the experimental results under the WHU and LEVIR-CD building change data sets show that the overall accuracy(OA),F1-score(F1)and Kappa coefficient(Kappa)indicators have improved,the multi-scale Siamese network method in this paper is more conducive to building change detection,and has obtained good test results.
作者
林娜
孙鹏林
王玉莹
黄韬
LIN Na;SUN Penglin;WANG Yuying;HUANG Tao(School of Smart City,Chongqing Jiaotong University,Chongqing 400074,China;Chongqing Geomatics and Remote Sensing Center,Chongqing 401147,China)
出处
《测绘科学》
CSCD
北大核心
2022年第5期185-192,共8页
Science of Surveying and Mapping
基金
重庆市教委科技项目(KJQN201800747,KJQN202103410)
关键词
变化检测
多尺度特征卷积
孪生神经网络
深度学习
change detection
multi-scale feature convolution
Siamese network
deep learning
