A Recognition Method for Marigold Picking Points Based on the Lightweight SCS-YOLO-Seg Model.

Area of Science:

• Agricultural Robotics
• Computer Vision
• Machine Learning

Background:

• Automated marigold harvesting faces challenges in accurate picking point identification due to complex backgrounds and flower pose variations.
• Existing methods often struggle with efficiency and resource requirements for real-time applications.

Purpose of the Study:

• To develop a novel, lightweight segmentation model (SCS-YOLO-Seg) for accurate picking point identification in automated marigold harvesting.
• To enhance the YOLOv8n-seg architecture for improved model compression and segmentation performance.

Main Methods:

• The study proposes SCS-YOLO-Seg, a lightweight segmentation model enhancing YOLOv8n-seg by incorporating StarNet backbone and a C2f-Star module.
• A dual-path collaborative architecture (Seg-Marigold head) optimizes segmentation efficiency.
• Picking points are identified by intersecting elliptical mask fitting with the stem skeleton.

Main Results:

• SCS-YOLO-Seg achieves substantial model compression, reducing size, parameters, and computational complexity.
• The model demonstrates a picking point identification accuracy of 93.36% with an average inference time of 28.66 ms per image.
• It effectively balances model compression with high segmentation accuracy compared to YOLOv8n-seg.

Conclusions:

• SCS-YOLO-Seg offers a robust and efficient solution for vision systems in automated marigold harvesting.
• The lightweight design makes it suitable for resource-constrained robotic applications.
• This method significantly improves the feasibility of automated flower harvesting systems.