Improved KS-GMM algorithm applied in classification and recognition of honey based on laser-induced fluorescence spectra.

Abstract

The laser-induced fluorescence (LIF) technique, which has been widely used for food testing, can be combined with various algorithms to classify and recognize different kinds of honey. This paper proposes the Kolmogorov-Smirnov test-Gaussian mixture model (KS-GMM) algorithm, which is coupled with the LIF technique to realize accurate classification and recognition of different types of pure honey. The experiments are designed and carried out to obtain a set of LIF spectrum data from various honey and syrup samples. The proposed KS-GMM algorithm is applied for classification and recognition, with GMM, k-nearest neighbor (kNN), and decision tree algorithms as cross-validation methods. By comparing recognition results of training sets containing different amounts of data, it is found that the KS-GMM algorithm exhibits a maximum recognition accuracy of 96.52%. The research results prove that the KS-GMM algorithm outperforms, to the best of our knowledge, the other three algorithms in classifying and recognizing the honey types.