RootEx: An automated method for barley root system extraction and evaluation

Abstract

Plant phenotyping plays a crucial role in agricultural research, especially in identifying resilient traits essential for global food security. Quantitative analysis of root growth has become increasingly vital in evaluating a plant’s resilience to abiotic stresses and its efficiency in nutrient and water absorption. However, extracting features from root images presents substantial challenges due to the complexity of root structures, variations in size, background noise, occlusions, clutter, and inconsistent lighting conditions.

In this study, we introduce “RootEx”, a comprehensive automated approach for extracting barley plant root systems from high-resolution images acquired from 2D root phenotyping systems set up in transparent growing mediums. Our method involves several stages, beginning with preprocessing to identify the Region of Interest (ROI). Subsequent stages utilize deep neural network-based segmentation, skeleton construction, and graph generation to produce detailed representations of root systems stored in RSML format. Notably, our dataset exclusively comprises primary roots without secondary roots or bifurcations, allowing for a focused examination of primary root characteristics and environmental adaptability.

Evaluation against established methods, RootNav 1.8 and 2.0, reveals significant improvements in root system reconstruction accuracy across various performance indicators. Although RootEx may exhibit slightly lower performance due to the absence of neural network-based tip detection, its advantages include minimal losses in missing root lengths and independence from dedicated training datasets. Our approach effectively mitigates detection errors, providing a reliable tool for precise barley root analysis in agricultural research.