Non-destructive detection of Flos Lonicerae treated by sulfur fumigation based on hyperspectral imaging

Abstract

Flos Lonicerae is often sulfur-fumigated during post-harvest handling, but excessive sulfur fumigation can reduce the quality of Flos Lonicerae products and then is harmful to human health. In order to achieve rapid and non-destructive identification for sulfur fumigated Flos Lonicerae, hyperspectral imaging was explored to establish detection models for sulfur fumigated Flos Lonicerae. A total of 450 Flos Lonicerae samples with different sulfur fumigation degrees were collected by a hyperspectral imaging system (371–1024 nm). Principle component analysis was applied to explore the separability of different sulfur fumigated Flos Lonicerae samples, and preliminary results demonstrated that sulfur fumigated Flos Lonicerae samples showed a trend of classification. Then, a method for data pretreatment, Savitzky–Golay filter (SG) combined with the standard normalized variable (SNV), was applied for de-noising and smoothing of the original data. The competitive adaptive reweighted sampling (CARS) algorithm was used to extract the optimal wavelengths ranged from 421 to 917 nm, and simultaneously the effect of CARS algorithm was compared with that of the successive projection algorithm and regression coefficient algorithm to determine the best method for the selection of optimal wavelengths. Least squares support vector machine (LS-SVM) and partial least squares regression (PLSR) were applied to build the classification models based on full spectra preprocessed by SG–SNV method. In order to simplify the calibration model, the PLSR, LS-SVM and BP neural network algorithms were used to build models based on optimal wavelengths. The overall results revealed that all models could achieve fast and non-destructive identification of sulfur fumigated Flos Lonicerae. The model based on CARS obtained optimal performance, and the CARS–LS-SVM model had the best classification effect for the prediction set with the correlation coefficient $$R_{P}^{2}$$ of 0.9109 and the root mean square error of 0.3353. Therefore, hyperspectral imaging technology combined with chemometrics can be utilized to achieve fast and non-destructive inspection for Flos Lonicerae fumigated with different sulfur concentrations.