Chili Ripeness Level Detection with YOLOv8 and Fuzzy Logic for Harvest Decision Making

Conventional chili harvesting relies on subjective human judgment, resulting in inconsistencies that necessitate a computer vision-based automation system. This study develops a decision support system integrating YOLOv8 for object detection and Mamdani fuzzy logic to assess chili ripeness levels. The YOLOv8 model was trained on 5,598 annotated chili images divided into three ripeness categories: ripe, unripe, and defective (rotten, diseased, or physically damaged), using an 80:20 training-testing split. YOLOv8 classification results serve as inputs to a fuzzy inference system that outputs three linguistic harvest decisions: delay, partial, or full harvest. Experimental evaluation indicates that YOLOv8 achieved 91.2% accuracy, 89.6% precision, and 87.3% recall on the test set. The fuzzy logic system obtained 88% accuracy in harvest decision-making on unseen data, demonstrating output consistency across repeated inferences. Overlapping triangular membership functions enable the fuzzy system to manage intra-class variations and image noise, thereby improving adaptability. These results confirm the feasibility of integrating YOLOv8 and fuzzy logic to support reliable and adaptive automated harvest decisions in chili farming, with potential application in precision agriculture.