ArithFusion: An Arithmetic Deep Model for Temporal Remote Sensing Image Fusion

Different satellite images may consist of variable numbers channels which have different resolutions, and each has a unique revisit period. For example, the Landsat-8 30 m resolution in their multispectral channels, Sentinel-2 10 pan-sharp channel, National Agriculture Imagery Program (NAIP) aerial 1 resolution. In this study, we propose simple yet effective arithmetic deep model for multimodal temporal remote sensing image fusion. The proposed takes both low- high-resolution at t1 together with low-resolution future time t2 from same location as inputs fuses them to generate t2. We an operation applied two points feature space take care changes. evaluated on three modality pairs fusion, including downsampled WorldView-2/original WorldView-2, Landsat-8/Sentinel-2, Sentinel-2/NAIP. Experimental results show that our outperforms traditional algorithms recent learning-based models by large margins most scenarios, achieving sharp fused while appropriately addressing