Restoration of Remote Sensing Imagery via Unified Denoising and Dehazing Approaches

Haze and noise severely degrade the quality of remote sensing images, leading to issues such as color distortion, reduced contrast, texture loss, and blurred edges, which compromise their utility in environmental monitoring and sustainable applications. This paper presents a novel method, the Cascaded Denoising with Dehazing Network (ConvDDNet), designed to enhance remote sensing imagery by simultaneously addressing denoising and dehazing in a unified deep learning framework. Unlike previous methods that treat denoising and dehazing as separate processes, this approach leverages superpixel segmentation to partition images into content-aware patches, enhancing efficiency and preserving detail. Atmospheric light and transmission characteristics are estimated at the superpixel level, enabling adaptive image restoration that accommodates varying ambient light conditions. Quantitative and qualitative evaluations on synthetic and real-world datasets demonstrate that ConvDDNet outperforms existing techniques in restoring image clarity and detail. This work supports sustainable development goals (SDGs), particularly SDG 13: Climate Action and SDG 15: Life on Land, by improving data quality for applications in environmental monitoring, resource management, and land-use assessment.