Tomato Maturity Detection and Counting Model Based on MHSA-YOLOv8.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Manual tomato grading and counting are labor-intensive, time-consuming, and prone to human error.
• Digital supervision and precision agriculture require automated solutions for fruit growth monitoring.
• Integrating artificial intelligence and machine vision offers a promising approach to overcome manual limitations.

Purpose of the Study:

• To develop an automated system for tomato fruit maturity grading and counting.
• To enhance the capabilities of object detection models for agricultural applications.
• To improve the efficiency and accuracy of tomato harvesting and grading processes.

Main Methods:

• Collected a tomato fruit image dataset from real production environments, accounting for occlusion and lighting variations.
• Proposed MHSA-YOLOv8, an object detection model incorporating the MHSA attention mechanism to improve feature extraction.
• Developed and evaluated models for both tomato maturity grading and fruit counting.

Main Results:

• The MHSA-YOLOv8 model achieved high performance in tomato maturity grading (Precision: 0.806, Recall: 0.807, F1-score: 0.806, mAP50: 0.864).
• The counting model demonstrated excellent accuracy (Precision: 0.990, Recall: 0.960, F1-score: 0.975, mAP50: 0.916).
• The developed models are effective for both online and offline detection scenarios.

Conclusions:

• The MHSA-YOLOv8 based system significantly enhances automated tomato maturity grading and counting.
• The proposed method offers a robust solution for complex agricultural environments.
• This technology can substantially boost harvesting and grading efficiency for tomato growers.