Related Experiment Video
Updated: Jun 3, 2025
04:36
Author Spotlight: Exploring Tea Aroma Using Solvent-Assisted Flavor Evaporation Technique
Published on: May 26, 2023
3.0K
Impurity detection of premium green tea based on improved lightweight deep learning model.
Zezhong Ding1, Mei Wang2, Bin Hu3
1Tea Research Institute of Shandong Academy of Agricultural Sciences, Jinan 250100, China; College of Mechanical and Electronic Engineering, Shihezi University, Shihezi 832000, China.
Food Research International (Ottawa, Ont.)
|January 8, 2025
Summary
This study developed a lightweight deep learning model to efficiently detect impurities in premium green tea, improving detection accuracy and speed while reducing computational resources for industrial application.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Agricultural Technology
Background:
- Manual sorting of impurities in premium green tea is inefficient and impacts quality.
- Existing deep learning models are often too resource-intensive for practical deployment in tea processing.
Purpose of the Study:
- To develop a lightweight deep learning model for effective impurity detection in premium green tea.
- To address the limitations of computational resources in actual production environments.
Main Methods:
- Utilized You Only Look Once version 8 (YOLOv8) models on a custom dataset.
- Implemented model lightweighting techniques including loss function replacement, lightweight convolution, model pruning, and knowledge distillation.
Main Results:
- The improved model achieved 4.2 Giga floating point operations (GFLOPs) and 791966 parameters, with precision (P) of 0.9379, recall (R) of 0.8959, and mean average precision (mAP) of 0.9484.
- Achieved a Frames Per Second (FPS) of 1362.7, representing a 368.0 FPS improvement over the original model.
- Reduced GFLOPs by 48.15% and parameters by 73.66% while enhancing detection performance.
Conclusions:
- The proposed lightweight model offers a viable solution for intelligent impurity sorting in premium green tea production.
- Demonstrated significant improvements in detection performance and efficiency with reduced computational load.
- Provides essential technical support for the automation and upgrading of the green tea industry.
