Hybrid Vision Approaches for Sustainable Agri-Food Automation: Comparing YOLOv8 and Classical Techniques in Produce Sorting

Abstract

Due to recent advancements in AI and ML technologies, AI has proven its superiority in vast variety of fields. Recently, computer vision emerged as one of these technologies that can be employed in agricultural sector by means of robots used for crops collections and sorting. This paves the road for a more reliable, efficient and robust approach of crops harvesting while reducing labor dependence and enhancing yield quality. This paper aims to investigate the role and benefits of implementing YOLOv8 AI vision framework as a technique in crops, such as fruits and vegetable, detection in comparison with present traditional techniques. The research will investigate several aspects including and not limited to contour detection, HSV-based color segmentation, and geometric analysis. The study demonstrates the lead, supported by validations, for YOLOv8 in both earlier detectors and traditional baselines in precision, recall, and mAP, even under challenging real-world conditions. The framework achieved near-perfect performance across three fruit categories: 98% accuracy for bananas, 100% for apples, and 99% for peaches, with minimal background confusion. Validation metrics further showed precision above 0.98, recall above 0.96, and mAP50 exceeding 0.99, with mAP50–95 reaching ~0.94. Inference efficiency was also competitive, with preprocessing at 3.4 ms and inference at 188 ms per image.The key contribution of this research is the creation of a benchmark that compares traditional approaches with YOLOv8 framework backed by experimental validation. Rather than stopping at a direct performance comparison, the study introduces a hybrid vision framework that combines the adaptability of deep learning with the efficiency of classical filters for grading by size, shape, and color.