Progressive spatiotemporal image fusion with deep neural networks

• DOI: 10.1016/j.jag.2022.102745
• Publication Date: 2022-03-18T23:05:36Z
• Authors: Jiajun Cai; Bo Huang; Tung Fung
• Keywords: Moderate-resolution imaging spectroradiometer

Spatiotemporal image fusion (STIF) provides a feasible and effective solution for generating satellite images with high spatial temporal resolution. As deep learning-based algorithms show great potential in high-quality images, we propose novel learning model, namely progressive spatiotemporal network (DPSTFN), which is coupled pansharpening super-resolution processes to satisfy requirements of STIF based on Moderate Resolution Imaging Spectroradiometer (MODIS) Landsat data. First, process adopted make full use two MODIS bands 250 m Second, enhances the information that existed coarse-resolution alleviate enormous resolution gap between images. Third, combining aforementioned auxiliary processes, framework proposed generate deliberate robust results. Experiments are conducted using MODIS-Landsat datasets distinctive landforms evaluate performance DPSTFN. The results subjective objective evaluation our performs better than state-of-the-art traditional Fit-FC RASTFM, EDCSTFN StfNet.