Related Experiment Videos
RTD-YOLO: a precise quality detection and grading model for rose tea.
Zezhong Ding1, Zhiwei Chen2, Bin Hu3,4,5
1College of Mechanical and Electronic Engineering, Shihezi University, Shihezi, China.
NPJ Science of Food
|June 5, 2026
Summary
This study introduces an improved YOLOv12n model for automated rose tea quality grading, enhancing market competitiveness. The new model achieves higher accuracy and efficiency, enabling real-time deployment on edge devices.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- The commercial value of roses necessitates quality control in rose tea production.
- Current manual grading of rose tea is inefficient and lacks objectivity.
- Deploying large-scale models for real-time quality assessment presents challenges.
Purpose of the Study:
- To develop an efficient and accurate automated system for rose tea quality detection and grading.
- To improve the market competitiveness of rose tea through intelligent classification.
- To address the limitations of manual grading and large model deployment.
Main Methods:
- An enhanced YOLOv12n model was proposed for rose tea grading.
- Key modifications included integrating the SEFF module, replacing Conv with ADown, and constructing a parameter-sharing detection head.
- The model was optimized using the Lamp pruning method.
Main Results:
- The improved model achieved 90.48% F1-score and 95.85% mAP.
- It demonstrated a significant increase in FPS (298.5) and real-time deployment capability (55.3 FPS on Jetson Orin NX).
- The model size was reduced to 2.0MB with decreased FLOPs and Params.
Conclusions:
- The developed model offers a viable solution for intelligent and automated rose tea grading.
- It provides technical support for enhancing the quality and marketability of rose tea.
- The model's efficiency and accuracy make it suitable for edge device deployment in production environments.
