UWE-Net: A Deep Learning Framework for Underwater Image Enhancement Integrating CBAM and Charbonnier Loss
S Jayasurya, S. Geetha, A. Sheik Abdullah, Utkarsh Mishra
Abstract
The enigmas surrounding aquatic ecosystems are solved mainly by underwater imaging, which provides valuable insights for scientific exploration, environmental monitoring, and conservation initiatives. However, the inherent challenges of underwater photography, such as reduced visibility, colour distortion, and limited contrast, hinder accurate species identification and habitat analysis. This paper addresses these challenges by proposing a novel approach named UnderWaterEnhanceNetwork (UWE-net). UWE-net combines VGG16, a pre-trained convolutional neural network, with Convolutional Neural Network (CNN) architecture and incorporates a Convolutional Block Attention Module (CBAM) within its decoder skip-connections to enhance underwater images effectively using both channels as well as spatial attention. We also introduce char bonnier loss as a loss function to train this architecture. The suggested procedure is accurate, as shown by the experimental findings., achieving a PSNR of 22.53 and SSIM of 0.85 for image enhancement. These advancements contribute to underwater imaging, paving the way for improved species identification and environmental monitoring in diverse marine ecosystems.