Abstract:The study and prediction of space weather entails the analysis of solar images showing structures of the Sun's atmosphere. When imaged from the Earth's ground, images may be polluted by terrestrial clouds which hinder the detection of solar structures. We propose a new method to remove cloud shadows, based on a U-Net architecture, and compare classical supervision with conditional GAN. We evaluate our method on two different imaging modalities, using both real images and a new dataset of synthetic clouds. Quantitative assessments are obtained through image quality indices (RMSE, PSNR, SSIM, and FID). We demonstrate improved results with regards to the traditional cloud removal technique and a sparse coding baseline, on different cloud types and textures.

What problem does this paper attempt to address?

This paper attempts to solve the problem of removing cloud shadows from solar images taken from the ground. Specifically, the author focuses on how to clear the shadows caused by ground - level cloud cover, which can interfere with the detection and analysis of solar atmospheric structures (such as active regions and prominences). To achieve this goal, the author proposes a new method based on the U - Net architecture and compares it with the Conditional Generative Adversarial Network (C - GAN). ### Problem Background Solar images may be affected by clouds when taken on the Earth's surface, resulting in a decline in image quality and making it difficult to accurately identify key structures in the solar atmosphere. This influence is particularly significant in the field of space weather research, because solar activities (such as solar flares and coronal mass ejections) have an important impact on the Earth's environment. Therefore, removing cloud shadows is crucial for subsequent automatic solar structure detection and monitoring. ### Main Challenges 1. **Similarity between cloud shadows and solar structures**: The texture and shape of cloud shadows may be similar to some solar structures (such as filaments), which makes it difficult to distinguish between the two. 2. **Diversity of cloud shadows**: Cloud shadows can have different textures (such as streaky or fluffy), and their thickness and direction also vary, increasing the processing difficulty. 3. **Lack of cloud - free control images**: In practical applications, it is very difficult to find completely cloud - free control images for training deep - learning models. ### Solutions The author proposes two main methods to address the above challenges: 1. **Deep - learning method based on U - Net**: - Use the U - Net architecture for image segmentation to predict cloud - shadow masks. - Denoise by directly predicting clean images or predicting cloud - shadow masks (in the form of transmittance or residuals). 2. **Conditional Generative Adversarial Network (C - GAN)**: - Combine U - Net and C - GAN and use adversarial training to improve the generalization ability and robustness of the model. ### Experimental Evaluation The author conducted experiments on two different imaging modes (Ca - II and H - α), using real images and newly created synthetic cloud datasets. Evaluation metrics include Root Mean Square Error (RMSE), Peak Signal - to - Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Fréchet Inception Distance (FID). The results show that the proposed method outperforms traditional cloud - removal techniques and sparse - coding baseline methods under various cloud types and texture conditions. ### Summary This research provides an effective solution for cloud - shadow removal in ground - based solar images, especially in dealing with complex cloud textures. Future work can further explore the processing of streaky clouds and optimize the recovery effect of dark structures such as filaments.

Removing cloud shadows from ground-based solar imagery

Removing cloud shadows from ground-based solar imagery

Deep neural networks for removing clouds and nebulae from satellite images

Cloud Removal for Remote Sensing Imagery via Spatial Attention Generative Adversarial Network

Cloud Removal in Remote Sensing Images Using Generative Adversarial Networks and SAR-to-Optical Image Translation

Cloud removal from satellite imagery using multispectral edge-filtered conditional generative adversarial networks

Filmy Cloud Removal on Satellite Imagery with Multispectral Conditional Generative Adversarial Nets

Gated Convolutional Networks for Cloud Removal From Bi-Temporal Remote Sensing Images

Cloud Removal of Full-Disk Solar H Images Based on RPix2PixHD

Attention Mechanism-Based Generative Adversarial Networks for Cloud Removal in Landsat Images

A Flexible Spatiotemporal Thick Cloud Removal Method with Low Requirements for Reference Images

Detection and Removal of Clouds and Associated Shadows in Satellite Imagery Based on Simulated Radiance Fields

Thick Cloud Removal Under Land Cover Changes Using Multisource Satellite Imagery and a Spatiotemporal Attention Network

Shadow removal method of soil surface image based on GAN used for estimation of farmland soil moisture content

Automated Detection of Cloud and Cloud Shadow in Single-Date Landsat Imagery Using Neural Networks and Spatial Post-Processing

Research on Cloud Removal Based on Fusing Multi-temporal Remote Sensing Images

Thick Clouds Removal From Multitemporal ZY-3 Satellite Images Using Deep Learning

DeCloud GAN: An Advanced Generative Adversarial Network for Removing Cloud Cover in Optical Remote Sensing Imagery

Thin cloud removal in optical remote sensing images based on generative adversarial networks and physical model of cloud distortion