Comparative study of indirect and direct feature extraction algorithms in classifying tea varieties using near-infrared spectroscopy.

Abstract

Tea, a globally cherished beverage, has become an integral part of daily life, particularly in China. Given the extensive variety of teas, each distinguished by unique price points, flavors, and health benefits, effective classification within the tea industry is crucial to address the diverse preferences of consumers. This study utilized indirect and direct feature extraction algorithms to analyze the Near-Infrared (NIR) spectra of various tea varieties and compared their classification outcomes. Principal Component Analysis (PCA) was employed as a dimensionality reduction technique for indirect feature extraction algorithms. The study began with the collection of NIR spectra from different tea varieties, followed by the application of three spectral preprocessing algorithms. Indirect and direct feature extraction algorithms were then used to reduce the dimensionality of the preprocessed data. A K-Nearest Neighbors (KNN) classifier analyzed the dimensionality-reduced data to determine classification accuracy. The findings revealed that the classification accuracies of indirect feature extraction algorithms consistently exceeded those of direct feature extraction algorithms, with the former generally surpassing 90.0 %, while the latter remained lower. This indicates that indirect feature extraction algorithms are more adept at handling complex spectral data. A significant decline in classification accuracy was observed when data were processed with Savitzky-Golay (SG). An in-depth analysis led to the development of an optimization plan incorporating the Successive Projections Algorithm (SPA), which effectively enhanced all classification accuracies to above 90 %.