Underwater Image Enhancement Using Improved Shallow-UWnet

Currently, with the development of industrialization, ocean exploration is actively carried out to investigate energy resources and undersea ecosystems. Remotely Operated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs), which are fully automated and not operated by humans, are used for these surveys. However, it is difficult to obtain clear images underwater due to low contrast and blur caused by the unique optical characteristics of the underwater environment. In recent years, many methods have been proposed for underwater image enhancement, with the most common methods using deep learning, such as generative adversarial networks (GANs) and convolutional neural networks (CNNs). Most of these methods are computationally and memory intensive, making real-time underwater image correction difficult. The Shallow-UWnet method is developed to solve this problem, and it enables a significant reduction in computational complexity compared to conventional methods. In this study, we improve Shallow-UWnet using Deformable Convolution, propose a new method with higher accuracy, compare its accuracy with that of the previous method, and verify its usefulness.