Atmospheric Image Correction and Removal of Cloud Cover for Satellite Images

Accurate atmospheric correction is crucial for surface analysis and environmental monitoring. Adverse weather conditions, such as clouds and haze, often degrade the contrast and fidelity of outdoor images, leading to uncertainties in surface observations. Cloud removal techniques in computer vision and image processing, including recent approaches like Generative Adversarial Networks (GANs), Image Inpainting, and Image Fusion, have been developed to address this challenge. However, these techniques face various challenges, such as limited availability of diverse training data, artifacts introduced in the resulting images, and computational complexity. This paper focuses on the removal of cloud cover and haze in multiband satellite images using the S2 Cloudless dataset and the Dark Channel Prior (DCP) algorithm. An approach is proposed to reconstruct clouded areas and generate cloud-free spatio-temporally continuous images for in-depth time-series analysis. The objective is to enhance the utility of the data by restoring the polluted areas in cloudy satellite imagery. The methods are evaluated using metrics such as Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Average Mean Pixel Value to assess the performance of both the input and output images. The results demonstrate the effectiveness of the methods in improving the quality and utility of cloud-covered satellite imagery, overcoming the challenges posed by limited training data, artifacts, and computational complexity.