Non-Invasive Sensing of Nitrogen in Plant Using Digital Images and Machine Learning for Brassica Campestris ssp. Chinensis L.

Abstract

Monitoring plant nitrogen (N) in a timely way and accurately is critical for precision fertilization. The imaging technology based on visible light is relatively inexpensive and ubiquitous, and open-source analysis tools have proliferated. In this study, texture- and geometry-related phenotyping combined with color properties were investigated for their potential use in evaluating N in pakchoi (Brassica campestris ssp. chinensis L.). Potted pakchoi treated with four levels of N were cultivated in a greenhouse. Their top-view images were acquired using a camera at six growth stages. The corresponding plant N concentration was determined destructively. The quantitative relationships between the nitrogen nutrition index (NNI) and the image-based phenotyping features were established using the following algorithms: random forest (RF), support vector regression (SVR), and neural network (NN). The results showed the full model based on the color, texture, and geometry-related features outperforms the model based on only the color-related feature in predicting the NNI. The RF full model exhibited the most robust performance in both the seedling and harvest stages, reaching prediction accuracies of 0.823 and 0.943, respectively. The high prediction accuracy of the model allows for a low-cost, non-destructive monitoring of N in the field of precision crop management.

Plant phenotyping relevance

画像から植物の窒素栄養状態を推定するフェノタイピング手法の開発・評価が研究の中心であり、画像特徴量と機械学習モデルを比較している。

Code and data availability

The paper describes 382 top-view pakchoi images, extracted phenotypic features, and RF/SVR/NN models for NNI prediction, but provides no public deposit of the image dataset, extracted features, trained models, or analysis code. The only paper-specific public resource is the MDPI supplementary material, which the text描述

No evidence-backed public reproduction asset is currently recorded.