Morphological feature extraction of corn roots using deep learning

Abstract

The study of corn root morphology is critical for understanding root architecture, which directly influences water and nutrient uptake, plant stability, and overall yield performance. It also plays a crucial role in advancing crop breeding programs. Traditional methods of analyzing root morphology are often labor-intensive, time-consuming, and subject to variability. This research introduces a deep learning (DL)-based approach for the automated and precise extraction of morphological features from monochrome images of corn roots. While DL methods have been widely applied to various agricultural problems such as yield estimation, cultivar classification, and disease detection, its application to plant features, particularly root traits, remains limited. In this study, three DL architectures- EfficientNet_B0, DenseNet_121, and ResNet_50- were used to extract and predict 12 morphological features from both raw and background-subtracted side-view images of corn roots. The results showed that all three architectures performed similarly, with DenseNet_121 slightly outperforming the others in terms of coefficient of determination and normalized root mean square error (NRMSE) metrics for background-subtracted images (mean R² 0.9199 and mean NRMSE 0.0444), while EfficientNet_B0 showed superior performance with raw images (mean R² 0.9057 and mean NRMSE 0.0480). Importantly, no significant difference in architecture performance was observed between raw and background-subtracted images. The study shows the potential of end-to-end learning by providing a robust, automated tool for plant morphological feature extraction.

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.