Enhancing Agricultural Efficiency through YOLOv8 for Papaya Ripeness Detection

Abstract

The increasing production of papaya fruit in Indonesia, which will reach 1.24 million tons by 2023, poses a challenge in the manual identification of fruit ripeness. The ripeness of papaya fruit greatly affects its quality, flavor, and selling value, and is important in determining the use of the fruit, whether as fresh fruit, vegetable, meat tenderizer, or ingredient for salad. Manual identification, which relies on visual, textural, and aroma assessments, is prone to eyestrain, subjective perception, and varying degrees of accuracy, leading to inconsistent results. Therefore, a more effective and efficient solution is needed to overcome this problem. This research aims to develop a papaya fruit ripeness detection system using YOLOv8. The dataset used consists of images of papaya fruit on the tree and those that have been cut, which have gone through the process of bounding box annotation, preprocessing, and augmentation using Roboflow. The hyperparameter used is epoch 50 and learning rate 0.01. The results of the training model show an accuracy rate of mAP50 of 0.873 for all classes, with values ​​in the unripe of 0.883, semi-ripe 0.852, and ripe 0.884. With this model, it is hoped that the public can obtain more accurate and precise information about the level of ripeness of papaya fruit, reduce dependence on manual methods, and make an important contribution to the papaya fruit farming industry by increasing the accuracy of ripeness assessment.