PS-Plant

PS-Plant applies photometric stereo (PS) imaging and machine learning to generate high-resolution 3D phenotypes of Arabidopsis thaliana for tracking and predicting growth responses across environmental conditions.

Key Features:

• Photometric stereo 3D imaging: Captures high-resolution 3D surface data using photometric stereo (PS) imaging across temporal series to enable analysis of diel leaf hyponastic movements.
• Automated segmentation and trait extraction: Employs bespoke computer vision algorithms and deep neural network architectures to segment rosettes and individual leaves and extract basic and advanced growth traits from 3D data.
• Temporal tracking and prediction: Tracks plant growth dynamics over time and supports prediction of growth performance under varying environmental conditions.
• Training dataset: Provides a curated dataset of 221 manually annotated Arabidopsis rosettes across 1,768 images for machine learning development and benchmarking.
• Environmental response analysis: Quantifies growth architecture changes under diverse environmental conditions to inform models of environmental impact on phenotype.

Scientific Applications:

• Diel phenotyping of Arabidopsis thaliana: Monitoring Arabidopsis thaliana across its diel cycle to quantify temporal growth patterns and hyponastic movements.
• Genotype–phenotype studies: Extraction of quantitative growth traits to support links between phenotype and genotype.
• Machine learning development and benchmarking: Training and evaluation of computer vision and deep learning models for automated plant phenotyping using the annotated dataset.
• Predictive environmental modeling: Informing prediction of plant growth responses to varying environmental conditions relevant to climate-change studies.

Methodology:

Uses photometric stereo (PS) imaging to obtain high-resolution 3D data, applies bespoke computer vision algorithms and deep neural network architectures for automated segmentation of rosettes and individual leaves, and extracts quantitative growth traits; training and evaluation use a dataset of 221 manually annotated rosettes across 1,768 images.