T10SLRE: A novel ensemble learning approach for rapid and non-destructive prediction of bread loaf volume in wheat using NIR spectroscopy

Bread loaf volume is a critical indicator of wheat processing quality, but conventional bread-making tests are laborious and time-consuming. This study evaluated near-infrared spectroscopy combined with machine learning for rapid prediction of loaf volume. A dataset of 5003 wheat samples was divided...

Descripción completa

Detalles Bibliográficos
Autores principales: Li, Lei, Tian, Wenfei, Zhao, Zihui, Xiao, Yonggui, Zhang, Yong, Liu, Jindong, Velu, Govindan, Zhonghu, He, Ibba, Maria Itria
Formato: Journal Article
Lenguaje:Inglés
Publicado: Elsevier 2025
Materias:
Acceso en línea:https://hdl.handle.net/10568/179156
Descripción
Sumario:Bread loaf volume is a critical indicator of wheat processing quality, but conventional bread-making tests are laborious and time-consuming. This study evaluated near-infrared spectroscopy combined with machine learning for rapid prediction of loaf volume. A dataset of 5003 wheat samples was divided into 4002 for model training and 1001 for independent testing. Conventional regression models and a convolutional neural network achieved testing set coefficient of determination (R2) of 0.70 to 0.73 with root mean square error (RMSE) of 23.29 to 25.08 mL/100 g. We propose a novel top 10 % sampling linear regression ensemble (T10SLRE) approach, which achieved a R2 of 0.74 and an RMSE of 22.92 mL/100 g, without spectral preprocessing or wavelength selections. It also classified wheat lines below the bread-making threshold of 850 mL/100 g with 97.90 % accuracy. This approach provides a practical tool that reduces cost and effort in wheat quality evaluation.