Artificial Intelligence Vision Methods for Robotic Harvesting of Edible Flowers

Area of Science:

• Agricultural Engineering
• Computer Vision
• Robotics

Background:

• Edible flower production faces labor shortages due to intensive hand-picking.
• Automated harvesting is crucial for meeting increasing market demand and grower profitability.

Purpose of the Study:

• To develop and evaluate an AI vision system for robotic harvesting of diverse edible flower species.
• To enable automated detection, 3D localization, pose estimation, and plucking point identification.

Main Methods:

• Utilized YOLOv5 for 2D flower detection.
• Employed the Segmentation Anything Model for 3D point cloud analysis and feature extraction.
• Developed a pose estimation algorithm and a linear regression model for plucking point determination.

Main Results:

• Achieved effective 2D detection of edible flowers.
• Demonstrated promising 3D localization and pose estimation using zero-shot and standard machine learning.
• Successfully estimated plucking points, correlating flower diameter with plucking height.

Conclusions:

• AI vision technology shows significant potential for automating edible flower harvesting.
• The developed framework is adaptable across various edible flower species and varieties.
• This innovation can enhance efficiency and reduce labor costs in edible flower cultivation.