Enhanced Animal Detection Using YOLOv8 and Data Augmentation Techniques: A Deep Learning Perspective

Abstract

In various aspects of human life, animals are widely exploited as a labor, consumption, pet, and for research. While animal use contributes favorably to various sectors of life, the number of animal populations allows the farm to expand to a larger scale, increasing the scale of the operation. This has made it increasingly difficult for the monitoring of the farm to do so when animals roam the complex areas, courtyards, open fields, or farmland in search of food. As for the animals that are hard to monitor are cats, goats, chickens, and cows. Thus, the development of animal detection technologies using deep learning to make it easier to monitor animals as they prowl for food. In the study, four types of animal animals, chickens, cats, goats, and cows use deep learning, yolov8 (You Only Look Once version 8) that is known for accuracy in identifying objects. The stages taken on this study include data collection, data annotations, data division, augmentation, data training, and results evaluation. Excellent training results from 125 epoch on the 85% data share scheme training, 15% of validation data, and 5% of data testing with datasets at 11242 pictures achieve an accuracy rate of 99.5%, precision by 79.4%, recall by 79.9%, and f1-score by 76.6%.