Implementation of YOLOv8 for Classifying Fertile and Infertile Eggs in the Chicken Hatching Process

Abstract

This study aims to develop an embryo detection system in chicken eggs using the YOLOv8 algorithm based on computer vision. This approach is proposed as a solution to the manual candling method which is often inaccurate and time consuming. The dataset used amounted to 4,396 chicken egg images, consisting of fertile and infertile categories. The model was trained using Google Collaboratory with GPU support, where the model was trained for 100 epochs to maximize accuracy. The evaluation results show that the YOLOv8 model is able to detect embryos with a high level of accuracy, indicated by a precision value of 93.2%, mean average precision (mAP) of 98.5%, and recall of 87.2%. The fertile category was successfully detected with a precision of 100% and a recall of 94.2%, while the infertile category had a precision percentage of 86.4% and a recall of 100%. These findings prove that the YOLOv8 algorithm can be effectively implemented to automate the selection process of fertile and infertile eggs, thereby improving efficiency and accuracy in the livestock production process.