Implementation of YOLOv8-Pose Model for Identification and Estimation of the Length of Skipjack and Tuna Like Species on the Website

Abstract

Skipjack and tuna species are crucial to Indonesia’s capture fisheries sector as they are the largest export commodities. These groups are diverse, and the identification and measurement of their length can be timeconsuming due to the abundance of caught fish and their morphological similarities. The objective of this study is to utilize an artificial intelligence algorithm to detect the species and estimate the length of four types of fish: bullet tuna (Auxis rochei), black skipjack (Euthynnus lineatus), mackerel tuna (Euthynnus affinis), and skipjack tuna (Katsuwonus pelamis). This algorithm will be implemented through a website. The YOLOv8-Pose deep learning model is employed to identify each fish species and estimate their length by determining keypoints. The dataset used consists of 148 images with rulers and 185 images of the four types of fish. Computer Vision Annotation Tool (CVAT) is used to assist in the labelling of the dataset, enabling the detection of boxes and keypoints. The labelled dataset is trained using Google Collaborator, resulting in the production of two model weights. Both models achieve an accuracy rate of 100%, as well as precision, recall, and an F1-score of 1. The coefficient value between the actual fish length and the detected fish length is 0.8649 or 86.5%, indicating a relationship between the two variables. To facilitate the identification, measurement, and storage of data in CSV format, a website is created using the Streamlit framework. In summary, the models accurately identify the limited number of datasets for Auxis rochei, Euthynnus lineatus, Euthynnus affinis, and Katsuwonus pelamis, and can provide estimates of fish length through the website.