"Is this blueberry ripe?": a blueberry ripeness detection algorithm for use on picking robots
Yan Liu1, Hongtao Zheng1, Yonghua Zhang1
1School of Information and Electrical Engineering, Hangzhou City University, Hangzhou, China.
Frontiers in Plant Science
|June 26, 2023
Summary
This study introduces a new YOLO-based algorithm for blueberry ripeness detection, improving robot picking accuracy and efficiency. The enhanced algorithm achieves higher detection rates and faster processing speeds for automated harvesting.
Area of Science:
- Agricultural Robotics
- Computer Vision
- Machine Learning
Background:
- Manual blueberry harvesting faces challenges due to labor scarcity and difficulty in accurately identifying fruit ripeness.
- Existing automated systems struggle with fruit shading, small size, environmental variations, and limited computational power.
- Developing efficient and accurate ripeness detection is crucial for robotic harvesting to meet market demands.
Purpose of the Study:
- To propose a novel, lightweight YOLO-based algorithm for enhanced blueberry ripeness detection in robotic harvesting.
- To improve the accuracy and efficiency of blueberry ripeness identification, addressing limitations of current automated systems.
- To develop an algorithm suitable for deployment on picking robots with limited computational resources.
Main Methods:
- A modified YOLOv5x architecture incorporating a lightweight attention module (Little-CBAM) within MobileNetv3.
- Implementation of a multi-method feature extractor (MSSENet) with a channel attention mechanism for improved feature representation.
- Utilized EIOU_Loss for faster training and k-means++ for optimized anchor frame clustering tailored to blueberry size.
Main Results:
- The proposed algorithm achieved a mean Average Precision (mAP) of 78.3%, a 9% improvement over YOLOv5x.
- Achieved a 2.1 times increase in Frames Per Second (FPS) compared to YOLOv5x on a PC terminal.
- Successfully translated to a picking robot, achieving real-time detection at 47 FPS, surpassing manual detection capabilities.
Conclusions:
- The developed YOLO-based algorithm significantly enhances blueberry ripeness detection accuracy and processing speed for robotic harvesting.
- The lightweight architecture and attention mechanisms effectively address challenges like fruit shading and small target detection.
- This advancement offers a practical solution for efficient and automated blueberry harvesting, improving market supply.
