A Novel Rapid Technique for Measuring Wheat Protein Content Using Near-Infrared Hyperspectral Imaging

Abstract

This study systematically evaluated the capability of near-infrared hyperspectral imaging (HSI) for rapid and non-destructive detection of wheat grain quality across 14 varieties from multiple ecological zones in Ningxia. By integrating machine learning approaches, predictions were made on the protein content of these wheat varieties. Results demonstrated that the Convolutional Neural Network (CNN) model achieved the highest coefficient of determination (R²) on both training and test datasets, indicating superior fitting performance and predictive accuracy. Among the feature wavelength extraction methods, the iterative Variable Importance in Projection on Latent Structures (iVISSA) technique stood out. After applying this method, the CNN model attained a test set R² of 0.9058 and a Root Mean Square Error (RMSE) of 0.4283, significantly enhancing model performance. These findings suggest that iVISSA effectively identifies feature wavelengths highly correlated with protein content, thereby improving the model's precision and generalization capability. In conclusion, near-infrared hyperspectral imaging combined with machine learning models offers a powerful tool for accurately predicting wheat grain protein content, providing valuable technical support for wheat quality assessment in Ningxia.

Plant phenotyping relevance

小麦穀粒のタンパク質含量という植物形質を、近赤外ハイパースペクトル画像と機械学習で推定する手法が研究の中心であり、特徴波長選択とCNN性能評価も行っている。

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.