Application of Electronic Nose with Multivariate Analysis and Sensor Selection for Botanical Origin Identification and Quality Determination of Honey

Characterization of the botanical origin and quality of honeys is of great importance and interest in agriculture. In this study, an electronic nose (e-nose) was applied for identifying the botanical origin of honey as well as determining their main quality components such as glucose, fructose, hydroxymethylfurfural (HMF), amylase activity (AA), and acidity. Principal component analysis (PCA) and discriminant factor analysis (DFA) were employed to generate scatter plots of honey samples from 14 botanical origins. Origin discrimination models with 100 % overall accuracy were established by least squares support vector machines (LS-SVM). LS-SVM outperformed the linear regression method of partial least squares regression (PLSR) for quality prediction, showing that the non-linear correlations between e-nose responses were important for the analysis of honey. Moreover, three sensor selection algorithms, namely, uninformation variable elimination (UVE), successive projections algorithm (SPA), and competitive adaptive reweighted sampling (CARS) were applied for the first time to analyze e-nose fingerprints of honey. After the calculation of the above three algorithms and the comparison of their results, from a total of 18 sensors, the important ones were selected for glucose (three), fructose (five), HMF (three), AA (five), and acidity (four) prediction, respectively. The results of sensor selection show the advantages of reducing redundancy of e-nose data, optimizing the sensor array of an e-nose, and improving the performance of models in terms of robustness. The overall results show that the laborious, time-consuming, and destructive analytical methods like high-performance liquid chromatography (HPLC), acid-base titration, and spectrophotometry could be replaced by e-nose to provide a rapid and non-invasive determination of the botanical origin and quality of honey.