Rapid quantitative typing spectra model for distinguishing sweet and bitter apricot kernels

Abstract

Amygdalin content in apricot kernels is an essential factor in the rapid and nondestructive identification of sweet or bitter apricot kernels through spectroscopy. Now, amygdalin content has been determined by high-performance liquid chromatography and near-infrared spectral database to construct a model so that the sweet or bitter apricot kernels could be identified and classified. Principal component analysis-K-nearest neighbor classification algorithm combined with multivariate scattering correction pretreatment method could distinguish sweet and bitter apricot kernels in the wavelength range of 1650-1740 nm with 98.3% accuracy and apricot kernel species with 96.3% recognition rate in the full wavelength spectrum. Furthermore, prediction of amygdalin content in bitter and sweet apricot kernels by partial least squares model was superior to that by back-propagation neural network model. This study provides a theoretical basis for quality identification of apricot kernel quality, as well as a method for nondestructive and rapid detection of sweet and bitter apricot kernels.

Supplementary information: The online version contains supplementary material available at 10.1007/s10068-022-01095-y.

Keywords: Amygdalin; Apricot kernel; High-performance liquid chromatography; Near-infrared spectroscopy; Quantitative detection model.

Fig. 1

Measurement results of amygdalin. (A) Standard curve of amygdalin content. (B) Comparison of amygdalin content in different varieties of apricot kernels

Fig. 2

KNN clustering results for eight MSC-preconditioned apricot kernels

Fig. 3

Predictive modeling of amygdalin content in apricot kernels. (A) Original spectral PLS model and error. (B) BP neural network model and error after SNV preconditioning. (C) Characteristic wavelengths selected by CARS