You Only Look Once (YOLOv8) for Fish Species Detection

As a maritime nation characterised by vast expanses of seas, Indonesia committed to preserving and managing its abundant marine resources. Therefore, developing the fisheries industry with the ability of computers and its technology, especially in the process of fish species detection precisely. In this research, the author proposed a new model with the state-of-the-art deep learning YOLOv8s (You Only Look Once) for fish species detection. The model is trained with Fish-gres dataset to provide ground truth labels for training and evaluation purposes. Four species observed in this research: Chanos Chanos , Upeneus Moluccensis , Rastrelliger Faughni , and Oreochromis Niloticus. The research succeeds in detecting all four species in a total dataset of 2,000 images with 70% training data, 20% validation data, and 10% testing data. From the best-trained model testing on the test dataset, the mAP@0,5 performance for all classes is 98.2%.