Deep Neural Network model for Underwater Image Enhancement

Abstract

In recent years, there has been a growing interest in the field of underwater image enhancement, driven by its significance in underwater robotics and ocean engineering. Initially, research efforts focused on physics-based approaches, but with advancements in technology, the utilization of deep convolutional neural networks (CNNs) and generative adversarial networks (GANs) has become prevalent. These state-of-the-art algorithms have shown impressive results; however, their computational complexity and memory requirements pose challenges to their practical implementation on portable devices used for underwater exploration tasks. Furthermore, these models are often trained on either synthetic or limited real-world datasets, limiting their applicability in real-world scenarios. In this paper, we propose a novel deep neural network architecture that maintains high performance while reducing the number of parameters compared to existing state-of-the-art models. Our approach aims to address the computational and memory limitations associated with underwater image enhancement algorithms. By leveraging the strengths of our architecture, we demonstrate its generalization capability by evaluating its performance on a combination of synthetic and real-world datasets. This approach enhances the practicality and applicability of our model in real-world underwater scenarios.