This is the core code of our work published in ISPRS J. 😘
This article utilizes the Semantic Distribution Alignment Loss to align the semantic features of SAR and Optical modalities, facilitating the fusion of complementary information.
The fundamental concept behind this loss is that two distributions are considered identical if their statistics are the same.
This loss quantifies the difference between feature distributions by mapping the features into the Reproducing Kernel Hilbert Space (RKHS).
For further details of this loss, please refer to "A Kernel Method for the Two-Sample-Problem."
If the project is helpful to you, please consider citing us.
@article{gretton2006kernel,
title={A kernel method for the two-sample-problem},
author={Gretton, Arthur and Borgwardt, Karsten and Rasch, Malte and Sch{\"o}lkopf, Bernhard and Smola, Alex},
journal={Advances in neural information processing systems},
volume={19},
year={2006}
}
@article{li2023aligning,
title={Aligning semantic distribution in fusing optical and SAR images for land use classification},
author={Li, Wangbin and Sun, Kaimin and Li, Wenzhuo and Wei, Jinjiang and Miao, Shunxia and Gao, Song and Zhou, Qinhui},
journal={ISPRS Journal of Photogrammetry and Remote Sensing},
volume={199},
pages={272--288},
year={2023},
publisher={Elsevier}
}