Identification of mildew degrees in honeysuckle using hyperspectral imaging combined with variable selection

Abstract

Mildew is one of the main reasons for the quality degradation of honeysuckle, which can lead to economic loss and threaten human safety. In order to detect different mildew degrees of honeysuckle, a method based on hyperspectral imaging technology was investigated. Different spectral pre-processing methods including Savizky–Golay filter (SG), standard normalized variable (SNV), multiple scatter correct (MSC), SG–MSC (a combination of SG and MSC) and SG–SNV were conducted for raw spectral data. A comparison was made among different pre-processing methods based on partial least squares regression models, of which the best method was SG–SNV. Then the uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were applied to extract effective variables (wavelengths) from the preprocessed data by SG–SNV. The extreme learning machine (ELM) models, which the ‘Sigmoidal’ was chosen as the incentive function and the number of neurons in the hidden layer was 40, were developed for identifying honeysuckle with different mildew degrees using full-spectrum data and the selected variables obtained by UVE, CARS, SPA, UVE–CARS, UVE–SPA, CARS–SPA and UVE–CARS–SPA. The classification results showed that the UVE–SPA–ELM model performed the highest accuracy of 100% for both training set and testing set and the proposed UVE–SPA method was optimal and powerful for the variable selection. The results of this study indicated that hyperspectral imaging technology could be a rapid and non-destructive analytical tool for identifying different mildew degrees of honeysuckle.