Tomato Ripeness Detection and Localization Based on the Intelligent Inspection Robot Platform

Area of Science:

• Agricultural Robotics
• Computer Vision
• Precision Agriculture

Background:

• Tomato field inspection and ripeness detection in China rely heavily on manual labor.
• Existing robotic solutions face challenges with functionality, environmental adaptability, cost, and positioning accuracy.

Purpose of the Study:

• To develop an intelligent tomato inspection robot integrating real-time ripeness recognition and precise spatial localization.
• To address limitations of manual labor and current robotic systems in greenhouse agriculture.

Main Methods:

• A Raspberry Pi 5 core controller and a lightweight, modular architecture were used for robot construction.
• A YOLOv8n-based detection model was trained on a multi-dimensional tomato ripeness image dataset.
• An Ultra-Wideband (UWB) positioning system was integrated for spatial localization.

Main Results:

• The YOLOv8n model achieved high performance with mAP@0.5 of 87.8% and mAP@0.5:0.95 of 72.7%.
• Field inspection achieved high detection precisions: 82.22% (immature), 92.66% (half-ripened), and 100% (fully ripe).
• The UWB system achieved a Root Mean Square Error of 0.231 m, with positioning errors within 0.24 m.

Conclusions:

• The intelligent robot offers a practical, low-cost, and flexible solution for tomato inspection in greenhouse environments.
• The integrated system addresses challenges of labor-intensive inspection and delayed harvesting, improving plantation management.
• The robot demonstrates efficient operation, inspecting a 50-m row in 10 min with a 2-hour battery life.