Data-driven optimization of nitrogen fertilization and quality sensing across tea bud varieties using near-infrared spectroscopy and deep learning

Abstract

Rapidly evaluating tea bud quality and diagnosing nitrogen status is crucial for optimizing nitrogen fertilization and enhancing tea quality. This study analyzed how key quality components (free amino acids (AA), tea polyphenols (TP), and the ratio of tea polyphenols to amino acids (RTA)) in tea buds from six varieties responded to nitrogen fertilizer. We also examined relationships between pigment levels (chlorophyll A (CA), chlorophyll B (CB) and carotenoids (TC)) and quality components across varieties. For quality estimation, our custom convolutional neural network (CNN) model, TeabudNet, offered superior prediction of TP, AA, and RTA (Rp values 0.924, 0.936, and 0.962 respectively) compared to traditional machine learning approaches. For nitrogen status diagnosis, we assessed RTA as an indicator of quality and nitrogen status, determining optimal nitrogen rates and thresholds delineating deficiency, sufficiency and excess for each variety. A ResNet-18 model reliably classified nitrogen status in tea buds and powder with 92–96% accuracy. This study provides robust technical support for optimizing nitrogen management and controlling quality during tea production.

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.