Quality Detection and Grading of Rose Tea Based on a Lightweight Model

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rose tea grading is crucial for market competitiveness.
• Manual grading is time-consuming and inefficient.
• Automated systems are needed for objective quality assessment.

Purpose of the Study:

• To develop a lightweight, automated model for rose tea grading.
• To improve the efficiency and accuracy of rose tea quality classification.
• To provide technical support for intelligent rose tea grading systems.

Main Methods:

• Implemented a lightweight deep learning model for object detection.
• Integrated the Convolutional Block Attention Module (CBAM) for enhanced feature extraction.
• Utilized the C2fGhost module to reduce network size while maintaining performance.
• Employed SIoU loss for improved bounding box regression accuracy.

Main Results:

• The proposed model achieved a mean Average Precision (mAP) of 86.16%, Precision (P) of 89.77%, and Recall (R) of 83.01%.
• The model demonstrated reduced computational cost (GFLOPs and Params) compared to the original model.
• Achieved high Frames Per Second (FPS) of 166.58, indicating real-time processing capability.
• Outperformed other advanced detection models in rose tea grading tasks.

Conclusions:

• The developed lightweight model effectively automates rose tea grading.
• The model offers a balance of high accuracy, efficiency, and reduced computational resources.
• This research provides a foundation for intelligent agricultural product quality assessment.