Automatic Fish Detection in Underwater Videos using Machine Learning

Fish have been around for about 450 million years, making them the oldest living organisms. There are about thirty different types of fish. Fish play a crucial role in the marine ecosystem as a source of nutrients. The economic well-being of humanity depends on fish. This paper aim is to find fish in underwater recordings and determine what kind of fish they are (based on species). In this study, 1200 photos of the 12 species represented in the LCF-15 dataset are considered. While the remaining 240 photos are used for testing, 960 are used for training. Different models of YOLOv5 (YOLOv5S, YOLOv5M, and YOLOv5L) are used to train and test our collected dataset. The proposed models are evaluated with F1 score. The YOLOv5S, YOLOv5M, YOLOv5L algorithms achieve a F1 Score of 92.5%, 94.9%, and 94.4% and mAP values of 94.9%, 95.6%, and 96.4% respectively. The findings of the best model show that YOLOv5M provides improved detection accuracy when compared to other methods.