Lightweight ESRGAN-Based Underwater Image Quality Enhancement

Abstract Underwater imaging is a critical tool for industries and research sectors that operate in marine environments. Image quality is often degraded due to light scattering, absorption, and turbidity in underwater. These challenges impact several industrial and environmental sectors that rely on underwater imagery. This paper presents an underwater image enhancement method using the lightweight deep learning technique ESRGAN (Enhanced Super Resolution Generative Adversarial Network), designed to improve visual clarity and detail in submerged environments. Clearer, more accurate underwater images can aid in the preservation of marine biodiversity, improve the efficiency of renewable energy installations, and promote sustainable fishing practices. Through extensive experiments on the openly available U45 dataset, the lightweight ESRGAN model is demonstrated to provide superior enhancement in regard to performance metrics PSNR, SSIM, MSE, and UIQM by 3.83%, 5.33%, 1.86%, and 9.16%, respectively, when compared with conventional methods.