A Dual Attention Network for Multimodal Remote Sensing Image Matching

Multimodal remote sensing image matching is a crucial task for many remote sensing applications. Due to different imaging mechanisms and conditions, multimodal images often suffer from significant nonlinear radiation distortions (NRD). To address this issue, this paper proposes a robust feature matching framework, where the learnable descriptors can be invariant to contrast changes. Firstly, we created a multimodal remote sensing image dataset for training the algorithm, as there is a lack of datasets in this field. Secondly, we use the phase congruency algorithm to preserve the invariant features between multimodal image pairs and optimize the parameters to make them more suitable for multimodal remote sensing image matching tasks. Finally, we design a matching network that introduces both position attention and channel attention modules to model the spatial and dimensional dependencies. Experimental results show that our method exhibits significant advantages over popular algorithms on publicly available datasets.