MSF: A Multi-Scale Fusion Generative Adversarial Network for SAR-to-Optical Image Translation

This paper proposes an image translation method based on multi-scale fusion GAN (MFS) network. In MFS network, there are two modules: optical image generation sub-network (OGS), optical image generation sub-network (OGS) and SAR image regressive sub-network (SRS). Firstly, we design a multi-scale fusion generator (MFG) to perform SAR-to-optical image translation and SAR image regression in OGS and SRS. MFG can extract multi-scale features at different scales, and fuse shallow and deep features, which enriches the semantic information of features. Combined with the PatchGAN discriminator, SAR-to-Optical image translation can be effectively realized. Finally, we conduct experiments on the SEN1-2 dataset, and the results show that our method performs better compared to the existing methods.