A rapid identification based on FT-NIR spectroscopies and machine learning for drying temperatures of Amomum tsao-ko

Abstract

Amomum tsao-ko (A. tsao-ko), is a medicine and food homology plant with high economic value. Different drying temperature affects flavor and quality of A. tsao-ko, which is closely related to its economic value and consumer acceptance. This research investigated the feasibility of combining Fourier-transform near infrared spectroscopy (FT-NIR) with chemometrics and machine learning to the identification of drying temperatures of A. tsao-ko. A total of 626 fresh samples from Honghe Prefecture, Yunnan, were dried to constant weight at 40 ℃ (62 h), 50 ℃ (54 h), 60 ℃ (43 h), 70 ℃ (35 h) and 80 ℃ (25 h) using an electric thermostatic drying oven. Principal component analysis (PCA) was used to observe the clustering of samples treated at different drying temperatures. Partial least squares-discriminant analysis (PLS-DA), support vector machines (SVM), and Residual Neural Network (ResNet) are used to build discriminant models. The accuracy of GS-SVM, GA-SVM, and ResNet models was more than 99% in the test set for identifying the drying temperature of A. tsao-ko. In contrast, the ResNet model based on synchronous two-dimensional correlation spectroscopy (2DCOS) images obtained the most satisfactory discrimination results. The use of appropriate data pre-processing (SD) can optimize the performance of the model to a certain extent and improve the accuracy of discrimination. This study demonstrates the feasibility of FT-NIR in the identification of the drying temperature of A. tsao-ko, providing a rapid and nondestructive method for the assessment and control of A. tsao-ko quality.