•  
  •  
 

Abstract

Underwater environments pose significant challenges in assessing image quality and organism detection due to light scattering and absorption, which limits visibility and color fidelity. Existing solutions often fail to effectively address these challenges, resulting in suboptimal image quality and hindering organism identification. This paper examines the problem through a dual-focused approach: enhancing underwater images and detecting organisms using the Underwater Image Enhancement Benchmark (UIEB) dataset. The first step introduces a state-of-the-art method for image enhancement by applying reinforcement learning (RL) principles. By formulating the enhancement process as a Markov Decision Process (MDP)—where states are represented by image features and actions are guided by a Deep Q-Network(DQN) the system iteratively selects optimal actions to improve image quality. This reinforcement learning framework integrates exploration and exploitation strategies, which are expected to surpass conventional methods by revealing intricate details of the underwater world that were previously unseen. For organism detection, this work employs VGG19, a state-of-the-art object detection and classification model. The integration of image enhancement and object detection, guided by reinforcement learning, provides a comprehensive solution for capturing the underwater environment in greater detail and enabling more accurate identification of marine organisms. Experimental results show that the proposed method achieved a Structural Similarity Index Measure (SSIM) of 0.85, a Peak Signal-to-Noise Ratio (PSNR) of 26.70 dB, and an overall organism detection accuracy of 98.43%, outperforming traditional enhancement and detection approaches.

Download
COinS