Cascaded Temporal and Spatial Attention Network for solar
 adaptive optics image restoration

Context. Atmospheric turbulence severely degrades the quality of images observed through a ground-based telescope. An adaptive optics (AO) system only partially improves image by correcting certain level wavefronts, making post-facto processing necessary. Several deep learning-based methods have recently been applied in solar AO post-processing. However, further research is still needed to get better while enhancing model robustness and using inter-frame intra-frame information. Aims. We propose an end-to-end network that can handle anisoplanatism leveraging attention mechanisms, pixel-wise filters, cascaded architecture. Methods. developed attention-based neural named Cascaded Temporal Spatial Attention Network (CTSAN) for restoration. CTSAN consists four modules: optical flow estimation PWC-Net explicit alignment, temporal spatial dynamic feature fusion, sharpness prior sharp extraction, encoder-decoder architecture reconstruction. also used hard example mining strategy create loss function order focus on regions are difficult restore, improve stability. Results. other two state-of-the-art (SOTA) supervised learning restoration trained real 705 nm photospheric 656 chromospheric corresponding Speckle images. Then all quantitatively qualitatively evaluated five testing sets. Compared SOTA methods, restore clearer images, shows stability generalization performance when restoring lowest contrast image.