A Spatial Attention GAN (SPA-GAN) Model for Robust Cloud Removal in Multispectral Imagery

Abstract

Remote sensing imaging is widely employed in a variety of sectors including environmental science, national military security and its excellent resolution and stable geometric elements make it ideal for weather monitoring. When a remote monitoring sensor on a robotic satellite collects terrestrial data it is affected by the climate notably clouds. Cloud cover influences the precision of optically remote sensing images. Existing cloud removal techniques for Sentinel-2 data usually rely on basic image processing approaches which are vulnerable to diverse cloud patterns and struggle with accurate reconstruction. Cloud removal from high-resolution remote sensing satellite images is an important pre-processing step before analysis. Addressing the issue of cloud contamination in Sentinel-2 imagery. Sentinel-2 data is becoming more useful in a variety of disciplines, including environmental monitoring, resource management and disaster response. In the proposed framework, the deep learning model was used for removing clouds from satellite imagery. Using the SPA-GAN model, Sentinel-2 multispectral images were produced without the presence of clouds. The proposed model was implemented for image-to-image translation challenges. Moreover, the SPA-GAN model produced realistic and high-quality images by successfully preserving spatial characteristics. The experimental results showed that the proposed model produced cloud-free imagery and enabled precise observation of Earth. The proposed model helps the researcher by identifying the cloud area to generate high-quality cloudless imagery which enhances the visual data's dependability and clarity.