Work

Maturity detection based autonomous robotic system for tomato harvesting in green houses

Year: 2026

Type: article

Abstract: • Maturity detection based autonomous harvesting system for tomatoes. • YOLOv8m achieved 99.9% accuracy in real-time tomato maturity detection. • YOLOv8m integrated robotic harvester attained 83.3% harvesting accuracy. • Raspberry Pi 4 ensures low-cost, real-time harvesting within 19.0 seconds. • Scalable AI-robotic solution ideal for precision agriculture in greenhouses. Efficient and precise harvesting, alongside real-time maturity classification, is vital for improving agricultural productivity and reducing post-harvest losses. This study presents a fully integrated, autonomous robotic tomato harvesting in greenhouses, using the You Only Look Once version 8 medium (YOLOv8m) deep learning model based on real-time fruit maturity. Tomatoes are categorized into unripe, semi-ripe, and ripe stages using a dataset of 1,137 annotated images. Trained over 200 epochs using the AdamW optimizer (learning rate: 0.001429), the model achieved a classification accuracy of 99.9%, with measures to prevent overfitting. The trained YOLOv8m model is deployed on Raspberry Pi 4 and processes real-time video feeds from USB Webcam. Upon detecting a ripe tomato, the system initiates a harvesting sequence that includes visual and auditory alerts, robot alignment and fruit detachment using robotic arm and a scissor-action cutting mechanism. The robotic platform, powered by Johnson motors, ensures row-wise navigation, while harvested tomatoes are gently collected into an onboard basket to minimize damage. Field evaluations demonstrated an average harvesting accuracy of 83.3%, a low fruit damage rate of 4.0%, and a processing time of 19.1 seconds/fruit (6.7 seconds for detection and 12.4 seconds for harvesting). The system achieved a throughput of 22 kg/h and an operational capacity of 18.98 m²/h. This research underscores the transformative potential of integrating advanced deep learning with robotic automation, providing a scalable, accurate, and efficient solution for autonomous tomato harvesting in precision agriculture. The system significantly reduces labour dependency, enhances harvesting consistency and contributes to sustainable farming practices.

Source: Results in Engineering

DOI: 10.1016/j.rineng.2026.110044