Comparative analysis of adaptive and general labeling methods for soybean leaf detection.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Soybean cultivation is vital for nutrition, economy, and industry.
• Precision agriculture relies on accurate soybean growth analysis.
• Effective leaf detection is key for monitoring soybean development.

Purpose of the Study:

• To evaluate the impact of different labeling methods on AI-based soybean leaf detection.
• To compare traditional general labeling with a novel context-aware labeling technique.
• To determine optimal labeling strategies for diverse soybean varieties.

Main Methods:

• Trained a YOLOv5L deep learning model using high-resolution soybean imagery.
• Implemented and compared a general labeling method.
• Implemented and compared a context-aware labeling method incorporating leaf length and bottom extremities.

Main Results:

• General labeling was more effective for soybean varieties with wider internodes and separated leaves.
• Context-aware labeling outperformed general labeling for medium soybean varieties with narrower, overlapping leaves.
• Labeling strategy significantly impacts AI detection performance.

Conclusions:

• Optimizing labeling strategies enhances the accuracy and efficiency of AI-based soybean growth analysis.
• Context-aware labeling offers advantages for specific soybean growth patterns.
• Improved AI detection supports better crop monitoring and yield prediction in high-throughput phenotyping.