Research on dynamic monitoring of nitrogen-driven quality changes throughout the entire growth period of cucumbers based on deep learning.

Abstract

This study proposes a multivariable quality prediction framework for cucumbers based on hyperspectral imaging, addressing the limitations of single-indicator approaches through chemometric analysis. Experiments were conducted under varying nitrogen levels and growth stages, with principal component analysis identifying nitrate, soluble sugar, and soluble solids as core indicators significantly correlated with nitrogen content. Spectral data underwent preprocessing via SG smoothing, MSC, SNV, and their paired combinations. Feature wavelengths were selected using CARS, UVE, and SPA algorithms, followed by comparative modeling with PLSR, SVR, and CNN approaches. Results demonstrated optimal performance for the CNN model utilizing full-spectrum input, achieving calibration set R 2 values exceeding 0.913 for all three indicators. This model enabled visualization of spatial distribution patterns, revealing spatial heterogeneity in cucumber quality under different nitrogen treatments. The method offers systematic rigor and high accuracy, providing a technical foundation for precision nitrogen management and vegetable quality enhancement.

Plant phenotyping relevance

キュウリの品質形質をハイパースペクトル画像から推定・可視化する手法が研究の中心であり、前処理、波長選択、機械学習モデル比較まで技術的に評価している。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.