Underwater Object Detection Using Deep Learning Techniques

Abstract

Underwater image detection plays a vital role in various marine applications such as oceanography, underwater robotics, and environmental monitoring. The challenges of underwater imaging, such as low visibility, distortion, and noise due to water quality and lighting conditions, make effective detection techniques crucial. This paper proposes an underwater image detection approach using the YOLOv8 (You Only Look Once version 8) model, a state-of-the-art deep learning architecture known for its efficiency in real-time object detection tasks. YOLOv8’s ability to balance high accuracy and computational efficiency is leveraged to detect various objects in underwater environments, addressing challenges like water turbidity and poor lighting. A dataset of underwater images with annotated objects is used to train the model, evaluating its accuracy, speed, and robustness to distortions. Experimental results demonstrate that YOLOv8 outperforms traditional methods in detecting underwater objects with mAP50 of 80% and faster inference times, making it a promising solution for real-time underwater image analysis in diverse applications.