Marine Object Detection using YOLOv4 Adapted Convolutional Neural Network

Abstract

This research presents an innovative application of the YOLOv4 object detection model for the identification and classification of marine objects within a dataset encompassing seven distinct classes. The study focuses on enhancing the robustness and accuracy of object detection in challenging marine environments, leveraging the unique capabilities of YOLOv4. Pre-processing steps involve resizing raw images, applying data augmentations, and normalizing pixel values to ensure optimal model training. Specifically tailored for underwater scenarios, additional color space transformations address variations in lighting conditions. The model is trained to detect marine objects such as fish, corals, and underwater structures, contributing to advancements in underwater exploration, environmental monitoring, and marine resource management. Experimental results demonstrate the effectiveness of the proposed approach, showcasing YOLOv4's ability to accurately identify and classify marine objects across the specified seven classes. This research not only expands the applicability of YOLOv4 in the marine domain but also provides valuable insights for the development of intelligent systems for underwater object detection.