Camarine: A Fish species recognition system through You Only Look One

Abstract

Data collection for marine sciences has always been arduous, mainly because of cost. The higher the cost is, the slower the growth of knowledge. To ease that cost, Camarine was built. An application for fish species recognition, Camarine used the algorithm You Only Look Once (YOLO) to seep through convolutional layers to detect and identify fish species. Twelve species of fish were categorized according to likeness and lack thereof. Over 4800 images were augmented to sport better results for the trained model. For testing, around 600 images were collected in various locations, including experiments done in a controlled environment. Results in detection showed an average of 88.63%, while the results in identification showed an average of 88.10%. For fishes of different appearances but the same species, the recorded accuracy was 92.66%. And for fishes of similar appearance but different species, the recorded accuracy was 86.60%. And finally, for general identification, 90.83% was the recorded accuracy. This all cumulates to the said 88.10% identification accuracy. Indeed, YOLO works well with identification, but this remains untested against turbid underwater images.