Dehazing of Multispectral Images Using Contrastive Learning In CycleGAN

Picture dehazing is a necessary aspect in computer vision, especially when employed with surveillance and satellite data. In this study, we provide a unique descriptive learning method for multispectral image augmentation based on CycleGAN. The suggested technique enhances the efficiency of the dehazing model by utilizing the benefits of descriptive learning and CycleGAN. Contrastive learning is used to separate hazy and clear images, while CycleGAN is used to transform hazy images into their dehazed counterparts. The proposed system is trained and evaluated on the RESIDE dataset, a benchmark dataset for image dehazing. The experiments demonstrate that the proposed system outperforms several existing methods, including CNN, DCP, HOT, DehazeNet, AODNET, and GDN. Our proposed system has practical implications for applications such as remote sensing and surveillance, where accurate image dehazing is essential.