ChronoRoot 2.0: An Open AI-Powered Platform for 2D Temporal Plant Phenotyping

Abstract

Plant developmental plasticity, particularly in root system architecture, is fundamental to understanding adaptability and agricultural sustainability. ChronoRoot 2.0 builds upon established low-cost hardware while significantly enhancing software capabilities and usability. The system employs nnUNet architecture for multi-class segmentation, demonstrating significant accuracy improvements while simultaneously tracking six distinct plant structures encompassing root, shoot, and seed components: main root, lateral roots, seed, hypocotyl, leaves, and petiole. This architecture enables easy retraining and incorporation of additional training data without requiring machine learning expertise. The platform introduces dual specialized graphical interfaces: a Standard Interface for detailed architectural analysis with novel gravitropic response parameters, and a Screening Interface enabling high-throughput analysis of multiple plants through automated tracking. Functional Principal Component Analysis integration enables discovery of novel phenotypic parameters through temporal pattern comparison. We demonstrate multi-species analysis, with Arabidopsis thaliana and Solanum lycopersicum, both morphologically distinct plant species. Three use cases in Arabidopsis thaliana and validation with tomato seedlings demonstrate enhanced capabilities: circadian growth pattern characterization, gravitropic response analysis in transgenic plants, and high-throughput etiolation screening across multiple genotypes.ChronoRoot 2.0 maintains the low-cost, modular hardware advantages of its predecessor while dramatically improving accessibility through intuitive graphical interfaces and expanded analytical capabilities. The open-source platform makes sophisticated temporal plant phenotyping more accessible to researchers without computational expertise.

Plant phenotyping relevance

植物の時系列画像から根・地上部・種子などの形態形質を抽出・追跡するオープンなAI基盤を開発し、精度向上、再学習、GUI、高スループット解析、検証まで扱っており、フェノタイピング手法が研究の中心です。

Code and data availability

The paper explicitly releases its full analysis source code (GitHub), the annotated infrared image dataset with multiclass segmentation masks (HuggingFace), demo phenotype video datasets, and a Docker image — all paper-specific, public, and actionable.