Instance-level phenotype-based growth stage classification of basil in multi-plant environments.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Global challenges like climate change and labor shortages necessitate advanced indoor farming solutions.
• Current crop monitoring relies on time-based methods, lacking physiological accuracy and reproducibility.
• AI-driven phenotyping offers a path toward precise and automated crop management.

Purpose of the Study:

• To develop and validate a phenotyping-based pipeline for accurate basil growth stage classification.
• To establish a physiologically grounded method for real-time crop monitoring in indoor farming.
• To leverage low-cost imaging and AI for enhanced precision agriculture.

Main Methods:

• Utilized top-view images from fixed cameras under varied lighting conditions in growth chambers.
• Employed YOLO for automated multi-plant detection and K-means clustering for individual plant image segmentation.
• Developed a convolutional neural network regression model to predict leaf pair counts for growth stage determination.

Main Results:

• YOLO achieved high detection accuracy (mAP@0.5 = 0.995).
• The regression model accurately predicted leaf pair counts (MAE = 0.13, R² = 0.96).
• Final growth stage classification accuracy surpassed 98% with consistent cross-validation performance.

Conclusions:

• The proposed pipeline enables automated, precise growth monitoring in multi-plant indoor environments.
• This low-cost solution supports precision environmental control, labor reduction, and sustainable smart agriculture.
• The leaf pair count serves as a robust, non-destructive indicator for physiological growth stages.