Potential of hyperspectral imaging for nondestructive determination of chlorogenic acid content in Flos Lonicerae

Abstract

Chlorogenic acid (CGA), as a major active component, is an important index for evaluating the quality of FlosLonicerae. Hyperspectral imaging (HSI) technology was applied for nondestructive estimating CGA content in FlosLonicerae. In order to obtain the best performance of calibration models, nine different pretreatment methods were investigated and compared based on partial least squares regression (PLSR) models. The optimal method was determined as a standard normal variable (SNV) method with RP2 of 0.9766 and RMSEP of 2.711 for further analysis. To simplify calibration models, different variables selection methods, including the uninformative variable elimination (UVE), successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), UVE–CARS, UVE–SPA, CARS–SPA, and UVE–CARS–SPA, were used to extracted characteristic wavelengths from the full spectrum. And then PLSR and least squares support vector machine (LS-SVM) were established based on full spectrum and the selected characteristic wavelengths, respectively. The results showed that the nonlinear UVE–CARS–LS–SVM model (RP2 = 0.9785 and RMSEP = 2.496) was the optimal model for predicting CGA content in FlosLonicerae. Therefore, this study revealed that the combination of HSI with SNV preprocessing method, UVE–CARS variable selection method and LS-SVM modeling had great potential to nondestructively and rapidly determine CGA content in Flos Lonicerae during storage.