文献类型： 外文期刊

作者： Huang Hua ¹ ; Liu Ya ² ; Kuerbanguli, Dulikuni ¹ ; Zeng Fan-lin ¹ ; Mayiran, Maimaiti ¹ ; Awaguli, Maimaiti ¹ ; Maidinuerhan, Aizezi ¹ ; Guo Jun-xian ³ ;

作者机构： 1.Xinjiang Agr Univ, Coll Math & Phys, Urumqi 830052, Peoples R China

2.Xinjiang Acad Agr Sci, Comprehens Testing Ground, Urumqi 830013, Peoples R China

3.Xinjiang Agr Univ, Mech & Traff Coll, Urumqi 830052, Peoples R China

关键词： Visible/near-infrared spectrum; Fractional differential; Permutation importance-random forest; K-nearest neighbors (KNN) regression; Soluble solids content

期刊名称：SPECTROSCOPY AND SPECTRAL ANALYSIS （ 影响因子：0.7； 五年影响因子：0.6 ）

ISSN： 1000-0593

年卷期： 2023 年 43 卷 10 期

页码：

收录情况： SCI

摘要： Soluble solids content (SSC) is an important physiological indicator of apple quality and maturation, and can be used for predicting the quality and maturity of apples. In this paper, 552 samples were collected at equal intervals of 3 d from the fruit swelling and setting stage to the complete mature stage, and the SSC was determined by collecting visible/near-infrared spectra from 380 to 1100 nm, and fused with fractional differential (FD) technique and replacement importance-random forest (Permutation Importance-Random Forest, PIMP-RF) algorithm to construct an ensemble learning model for SSC prediction in apple during maturing period. The results showed that the fractional differential orders of the PLS model were 0, 0. 4, 1.1, and 1. 6, and the results of feature importance and interpretability analysis by the PIMP-RF algorithm showed that the key wavelengths for predicting the SSC of maturity apples using visible/near-infrared spectroscopy were mainly in the visible band, which provided a theoretical basis for the future development of a rapid nondestructive detection device for Xinjiang Red Fuji apples. The ensemble learning model of apple ripening SSC constructed based on fractional differential technique and PIMP-RF algorithm has good prediction ability, with the correlation coefficient r equal to 0. 989 2, mean absolute error MAE equal to 0. 241 2, root mean square error RMSE equal to 0. 309 1 and mean absolute percentage error equal to 0. 018 3 in the training set. The correlation coefficient r of the test set is equal to 0. 903 8, the mean absolute error MAE is equal to 0. 549 9, the root mean square error RMSE is equal to 0. 740 8, and the mean absolute percentage error is equal to 0. 043 4, compared to the FD0-PIMP-RF, FD0. 4-PIMP-RF, FD1. 1-PIMP-RF, and FD1. 6-PIMP-RF models, the ensemble learning model is optimal. Therefore, the integrated fractional order differentiation technique and PIMP-RF algorithm, combined with visible/near-infrared spectroscopy, can successfully and effectively predict the soluble solids content of apples during maturing period.