Open AccessHeterogeneous remote-sensing image matching method based on deep learning
Author(s) -
Huitai Hou,
Qin Xu,
Chaozhen Lan,
Zhixiang Cui,
Jianqi Qin,
Lianxia Wang
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1684/1/012110
Subject(s) - robustness (evolution) , computer science , ransac , convolutional neural network , artificial intelligence , matching (statistics) , adaptability , pattern recognition (psychology) , deep learning , graph , blossom algorithm , artificial neural network , image (mathematics) , computer vision , mathematics , theoretical computer science , statistics , ecology , biochemistry , chemistry , biology , gene
In this paper, a deep learning matching method is proposed to address the difficulty in matching heterogeneous remote sensing images, which is caused by their differences in imaging modes, time phases, and resolutions. A heterogeneous image is inputted into the convolutional neural network (CNN) to extract deep features and, then, a graph neural network (GNN) is used for matching. Finally, correct matching points are retained while ensuring the effective elimination of mismatches. The algorithm adopted in this study was tested using multiple sets of heterogeneous remote-sensing images and compared with D2-Net+NN+RANSAC and Superpoint+SuperGlue algorithms. The results show that the algorithm used in this study possesses strong adaptability and robustness and is an optimal algorithm for the robust matching of remote-sensing images with different sources and large differences.