Vision model predicts root traits from underground images accurately
RootQuantV2: Adapting a Vision Foundation Model for Root-Trait Regression from Minirhizotron Imagery
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
Summary
Measuring root traits in crops grown in fields is hard because traditional methods are slow and require manual work. The authors improved a computer vision model to estimate root length and surface area directly from pictures taken underground by special cameras called minirhizotrons. They used a powerful image analysis method but only adjusted a small part of it, making the model efficient. Their updated model predicts root traits more accurately than previous attempts, enabling faster and automated assessments of roots without detailed manual drawings.
What this means in practice
- •For agricultural data teams: Automate estimation of root length and surface area from minirhizotron images to speed up below-ground crop trait analysis at scale.
- •For environmental monitoring teams: Use archived numeric root trait data combined with vision models to improve understanding of soil-root interactions without manual image annotation.
Authors
Kinjalk Parth, Sebastian Varela, Andrew D. B. Leakey
Abstract
A lack of high-throughput phenotyping solutions for root traits in field-grown crops has severely constrained understanding and improvement of below-ground traits and processes. Minirhizotrons are the standard non-destructive root-phenotyping method in field environments. Computer vision solutions are needed to allow automated trait estimation at scale, but training data is scarce and human annotations are often inaccessible because they reside in proprietary software that only exports per-image scalar totals of root length and surface area. Nevertheless, large numeric archives of these root traits already exist. RootQuant showed that the traits can be predicted directly from the whole image by regression, thus removing manually traced masks from the pipeline; RootQuantV2 takes that idea further by replacing RootQuant's CNN backbone with a self-supervised ViT. We adapt a frozen DINOv3 ViT-L/16 with a hybrid parameter-efficient scheme. Training only 11.9M parameters (3.78% of the model), RootQuantV2 achieves length and area $R^2$ of 0.950 and 0.930, respectively, while lowering length/area RMSE by 24.3%/20.7% over RootQuant. RootQuantV2 thus repurposes legacy numeric archives for high-throughput, automated root trait estimation.