Multiple Object Recognition System for Lake using The YOLOv8 Technique

Abstract

This research tackles the challenges of underwater photography in lakes, concentrating on developing and evaluating a multiple object detection system through the advanced You Only Look Once Version 8 (YOLOv8) architecture. The inherent limited visibility in underwater environments poses difficulties in accurately capturing object shapes and colors, crucial for applications like underwater robots engaged in search missions. Leveraging Python and Google Colaboratory, the project implements YOLOv8 for multiple object detection using a dataset of 1116 lake underwater images, processed with LabelImg for object recognition and dataset development. The publicly accessible dataset at http://tinyurl.com/32z25b serves as a valuable resource. YOLOv8 consistently demonstrates exceptional performance in lake environments, achieving an impressive mean Average Precision 50-95 (mAP 50-95) of 95.5% for single-object detection in both training and validation sets. Despite a gradual decrease to 73.8% for 5 objects in more complex scenes, the model maintains a robust overall average of 87.42% in the test set. These findings offer valuable insights for informed decisions when deploying YOLOv8 across diverse underwater settings, particularly in lakes.