Answers are grounded in this paper and cite their sources.
Roots play a vital role in material cycling and energy flow within ecosystems, and minirhizotrons have become a widely used non-destructive tool for long-term observation of root growth. However, minirhizotron root images are often affected by complex soil background interference and inconsistent root colours at different growth stages, which renders conventional segmentation algorithms unstable. These algorithms struggle to resist colour variations or effectively filter soil noise, resulting in segmentation bias. A novel Convolutional Neural Network (CNN), called CFRA-Net, is proposed for segmenting plant roots in such images, based on a classification fusion (CF) strategy and a root attention (RA) mechanism. CFRA-Net enables the network to learn the colour characteristics of roots at different growth stages and assigns greater weight to roots than to the background, thereby comprehensively improving root segmentation performance. Leave-one-out cross-validation results showed that the accuracy, recall, precision, F1-score, and IoU of CFRA-Net were 95.1%, 83.2%, 80.4%, 82.2%, and 69.8%, respectively. Compared with commercially available methods, CFRA-Net enables robust segmentation of roots exhibiting varying colours and morphologies in dynamic soil environments. Subsequently, a 12-month quantitative analysis of Populus tomentosa roots at different soil depths under three irrigation treatments was conducted. Results showed that no irrigation produced the greatest 12-month root length and surface area density across all soil layers, while furrow irrigation promoted the accumulation of fine roots in the 0–340 mm topsoil layer, exhibiting the most consistent 12-month root growth. This segmentation method provides technical support for image analysis aimed at exploring root-soil ecosystem interactions.
No takes yet. Share an insight, caveat, or question.
Fu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: