Palm fruit ripeness detection and classification using various YOLOv8 models.

Abstract

The significance of palm oil, which contributes 30 % of the world's total vegetable oil production, cannot be overstated. Its numerous applications, ranging from soap to cosmetics, have increased demand, thereby increasing the importance of yield management. Human graders have traditionally been responsible for determining the ripeness of oil palm fresh fruit bunches (FFBs), a task upon which the oil extraction rate (OER) relies heavily. This rate has significant economic implications: a 0.13 % drop in OER due to unripe fruits can result in a staggering RM 340 million loss. Precision is stressed, prompting automated detection research. Computer vision and Artificial Intelligence are becoming more effective at assessing oil palm fruit ripeness. However, many methods require complex operations, controlled settings, or manual calibrations. Although innovative, microwave sensors and inductive techniques have drawbacks like sample preparation and equipment dependence. This study investigates the potential of the YOLOv8 framework, particularly its YOLOv8m variant, for ripeness classification and detection. This model's mAP50-95 of 0.927 balances computational efficiency and accuracy, indicating its potential to revolutionize the palm oil industry's fruit assessment procedures. The findings here shed light on the model's efficacy and highlight its potential as an industry-standard solution, bridging gaps in ripeness detection methodologies.