Application of Cycle-GAN-Based Water Surface Highlight Removal Network in Inland Water Scenes

Inland water scenes are often intricate, and the impact of water highlights on the images captured in these settings is readily apparent. Image processing techniques are commonly employed to process pictures of inland waters. For example, when unmanned surface vehicle (USV) platforms are used to capture water images, the presence of highlight areas can detrimentally affect overall image quality. Therefore, it becomes imperative to eliminate highlights from the water surface in the images. The objective of this study is to investigate the effect of highlight removal in water images using Cycle-GAN (Generative Adversarial Network) architecture for training as a method for eliminating water surface highlights. Experiments were conducted using the USVInland dataset, and results indicate that the network effectively removes water surface highlights. The successful removal of image highlights demonstrates that properly trained networks have potential to yield favorable outcomes, thereby offering valuable insights for future research involving inland water imagery.