Related Experiment Video
Updated: Jul 13, 2025
11:38
Author Spotlight: Enhancing PSC-to-Functional Cell Differentiation Using ML Models Based on Live-Cell Bright-Field Imaging
Published on: October 4, 2024
614
A tea bud segmentation, detection and picking point localization based on the MDY7-3PTB model.
Fenyun Zhang1, Hongwei Sun1, Shuang Xie1
1School of Automation, Hangzhou Dianzi University, Hangzhou, China.
Frontiers in Plant Science
|October 16, 2023
Summary
This study introduces the MDY7-3PTB model for accurate tea bud identification and localization, crucial for automated tea picking. The model enhances precision and speed in complex environments, overcoming challenges in distinguishing tea buds from leaves.
Area of Science:
- Computer Vision
- Robotics
- Agricultural Technology
Background:
- Accurate identification and localization of tea picking points are essential for automating the harvesting of famous teas.
- Distinguishing tea buds from surrounding leaves is challenging due to similar visual characteristics, hindering manual and automated picking efforts.
Purpose of the Study:
- To develop an advanced model for precise segmentation, detection, and localization of tea picking points in mechanical tea harvesting.
- To improve the efficiency and accuracy of automated tea picking systems by addressing the limitations of existing computer vision models.
Main Methods:
- Proposed the MDY7-3PTB model, integrating DeepLabv3+ segmentation with YOLOv7 detection capabilities for a sequential picking point identification process.
- Utilized MobileNetV2 as a lightweight backbone for enhanced computational speed and incorporated Convolutional Block Attention Modules (CBAM) for performance optimization.
- Employed the Focal Loss function to mitigate class imbalance issues within the dataset, thereby improving segmentation and detection accuracy.
Main Results:
- Achieved high performance in tea bud segmentation with a mean Intersection over Union (mIoU) of 86.61% and mean pixel accuracy (mPA) of 93.01%.
- Demonstrated superior tea bud picking point recognition and positioning, yielding a mean Average Precision (mAP) of 93.52% and a positioning precision of 96.41%.
- Outperformed existing mainstream segmentation and detection models, showcasing strong versatility, robustness, and minimal missed detections in complex scenarios.
Conclusions:
- The MDY7-3PTB model offers a robust and accurate solution for identifying and localizing tea picking points, significantly advancing automated tea harvesting.
- The model's ability to provide precise 2D coordinates for tea buds lays a strong theoretical foundation for future developments in robotic tea picking.
