AITP-YOLO: improved tomato ripeness detection model based on multiple strategies.
Wenyuan Huang1,2, Yiran Liao3, Peiling Wang1
1College of Information Engineering, Sichuan Agricultural University, Yaan, China.
Frontiers in Plant Science
|June 10, 2025
Summary
A new AITP-YOLO model improves tomato ripeness detection by enhancing small target identification and feature fusion. This advanced model achieves higher accuracy and precision while reducing model size for practical agricultural applications.
Area of Science:
- Computer Vision
- Machine Learning
- Agricultural Technology
Background:
- Accurate tomato ripeness detection is crucial for automated harvesting.
- Existing models face challenges with small target identification and feature fusion in complex environments.
Purpose of the Study:
- To develop an enhanced YOLOv10s model, termed AITP-YOLO, for improved tomato ripeness detection.
- To address limitations in identifying small tomatoes and fusing multi-scale features.
Main Methods:
- Implemented a four-head detector with a small target detection layer.
- Utilized multi-scale feature fusion with cross-level features for enhanced generalization.
- Modified the bounding box loss function to Shape-IoU for improved regression precision.
- Applied Network Slimming pruning for model compression.
Main Results:
- The AITP-YOLO model achieved 92.6% average precision, 89.7% accuracy, and 87.4% recall.
- Compared to YOLOv10s, model weights were compressed by 7.64%.
- Average precision, accuracy, and recall improved by 4.6%, 5.8%, and 7.3%, respectively.
Conclusions:
- The enhanced AITP-YOLO model offers a reduced size with superior detection capabilities for tomato ripeness.
- It enables efficient and precise recognition of tomato stages in complex backgrounds.
- Provides a valuable technical reference for automated tomato harvesting systems.
