YOLO-SAM AgriScan: A Unified Framework for Ripe Strawberry Detection and Segmentation with Few-Shot and Zero-Shot...

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Traditional image segmentation methods for agriculture are slow and require extensive manual annotation.
• This labor-intensive process limits the scalability of automated agricultural monitoring systems.

Purpose of the Study:

• To develop an efficient and scalable framework for on-plant ripe strawberry segmentation.
• To overcome the limitations of manual annotation in agricultural computer vision tasks.

Main Methods:

• A hybrid framework, YOLO-SAM AgriScan, integrating YOLOv11 for few-shot object detection and Segment Anything Model 2 (SAM2) for zero-shot segmentation.
• YOLOv11 was fine-tuned with minimal annotated samples, while SAM2 generated masks without further supervision.
• The system was evaluated on custom and public datasets in both full-data and data-constrained scenarios.

Main Results:

• The framework achieved high performance, with a mean Dice score of 0.95 and IoU of 0.93 on a custom dataset.
• Competitive results were maintained on public data (Dice: 0.95, IoU: 0.92).
• Demonstrated robustness, generalizability, and practical relevance in diverse agricultural environments.

Conclusions:

• Combining few-shot detection and zero-shot segmentation offers an effective annotation-light approach for agricultural phenotyping.
• YOLO-SAM AgriScan enables scalable and efficient strawberry segmentation, accelerating intelligent farming systems.