LAUSR

Spatio-temporal fusion aims to integrate mul-ti-source remote sensing images with complementary high spatial and temporal resolutions, so as to obtain time-series high spatial resolution fused images. Currently, deep learning (DL)-based spatio-temporal fusion methods have received broad attention. However, On the one hand, most of existing DL-based methods train the model in a band-by-band manner, ignoring the correla-tions among bands. On the other hand, the general coarse spa-tio-temporal changes of the low spatial resolution images (e.g. MODIS) calculated at the pixel domain cannot completely cover the fine spatio-temporal changes of the high spatial resolution images (e.g. Landsat), due to the complex surface features and the general large spatial resolution ratio between the fine and coarse images. Besides, existing DL-based spatio-temporal fusion meth-ods are insufficient in exploring multi-scale information by only stacking convolutional kernels with different sizes. To alleviate the above challenges, we propose a Progressive Spatio-Temporal Attention Fusion model in a multi-band training manner based on Generative Adversarial Network (PSTAF-GAN). Specifically, we design a flexible multi-scale feature extraction architecture to extract multi-scale feature hierarchies. Then, the spatio-temporal changes are calculated on the feature domain in different feature hierarchies. Besides, a spatio-temporal attention fusion architec-ture is proposed to fuse the spatio-temporal changes and the ground details in a coarse-to-fine manner, which can explore multi-scale information more sufficient and gradually recover the target image. The results of quantitative and qualitative experi-ments on two publicly available benchmark datasets show that the proposed PSTAF-GAN can achieve the best performance com-pared with the state-of-the-art methods.