Metafoemer Super-Resolution Network with Recursive Gated Attention for the Meteorological Satellite Cloud Image
Lun-Qian Wang, Xinghua Wang, Weilin Liu, Bo Xia, Hao Ding, Jinglin Zhang · 2023
Increasing the resolution of satellite cloud images with super-resolution (SR) methods can help meteorological systems process geographic information more accurately. In this paper, we verify several SR methods on the natural color cloud image (NCCI) dataset. On one hand, we find that the residuals between the interpolated up-sampled image and the ground truth are beneficial to the SR effect of the satellite cloud image. On the other hand, we propose the MetaFormer super-resolution network (MetaSR), which adopts the MetaFormer architecture to extract satellite cloud features more efficiently. The recursive gated convolution and the channel-space attention are also embedded in MetaSR to achieve arbitrary-order spatial interactions and feature filtering. The experimental results indicate that MetaSR achieves excellent SR results for satellite cloud images on the NCCI dataset. In detail, the MetaSR takes only 34% parameters of EDSR and improves the PSNR by about 0.045dB on NCCI for ×4 SR.