Publication

Abstract : Earth observation applications rely on satellite data, but opaque or semi-transparent clouds affect optical observations. Deep learning techniques are evolving to generate cloud-free data from single impeded optical observation. This paper introduces a Generative Adversarial Network (GAN) based cloud-free reconstruction architecture combining the advantages of spatial-attention mechanisms. The proposed Multi Attention Enhanced Thin Cloud-Removal Generative Adversarial Network (MAE-CG) combines proven spatial attention frameworks of Convolutional Block Attention Module (CBAM) and Global-Local Attention Module (GLAM) to perform extraction of superior spatial context extraction. The proposed MAE-CG demonstrates superior performance in terms of Peak Signal Noise Ratio (PSNR), Structural Similarity Index (SSIM), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) compared to existing methods while tested against RICE-1 dataset. The model effectively reconstructed cloud-free images by focusing on critical features and spatial details, making it robust for thin cloud removal. The model successfully reconstructed cloud-free images by concentrating on essential features and spatial details, demonstrating its resilience to thin clouds.