Pheno-Deep Counter: a unified and versatile deep learning architecture for leaf counting.

Area of Science:

• Plant science
• Computer vision
• Machine learning

Background:

• High-throughput plant phenotyping relies on accurate trait measurement.
• Manual plant trait observation is time-consuming and limits scalability.
• Automated leaf counting is crucial for plant research but current methods lack broad applicability.

Purpose of the Study:

• To develop a versatile deep learning model for automated plant leaf counting.
• To enable leaf counting across multiple plant species and imaging modalities.
• To provide an open-source solution for accelerating plant phenotyping.

Main Methods:

• A single deep neural network architecture, Pheno-Deep Counter, was developed.
• The network processes two-dimensional (2D) plant images, including visible light, fluorescence, and near-infrared.
• The architecture is designed for flexibility, allowing adaptation to new data modalities without internal structural changes.

Main Results:

• Pheno-Deep Counter accurately predicts leaf count in various plant species with rosette-shaped growth.
• The model demonstrates robustness across different imaging modalities.
• Leaf counting is achieved in seconds after model training, significantly improving efficiency.

Conclusions:

• Pheno-Deep Counter offers a universal and adaptable solution for automated leaf counting.
• The open-source approach facilitates broader adoption of machine learning in plant phenotyping.
• This tool has the potential to overcome bottlenecks in high-throughput phenotyping research.