Root Nodule Estimation Using a Non-Destructive Machine Learning Approach

The root nodule formation in legumes such as faba bean is a key indicator of biological nitrogen fixation and sustainable agriculture. While image-based and deep learning methods have recently enabled automated nodule detection with high accuracy, these approaches rely on destructive root extraction, controlled imaging, and labour-intensive annotation, which hinder their scalability and practical field application. To address these limitations, this study proposes a non-destructive, and cost-effective machine learning framework for early prediction of root nodule counts in faba bean, using only easily measurable morphological traits. The developed dataaugmented model, trained exclusively on field-accessible, nondestructive features, achieved a test R2 of up to 0.95 and maintained low error rates (RMSE = 0.0384; MAE = 0.0282) even with limited data. Data augmentation further improved both prediction accuracy and model robustness. Overall, this approach offers a scalable solution for high-throughput, fieldready nodule phenotyping, overcoming significant barriers associated with image-based techniques and enabling practical automation in legume symbiosis research.