Rapid quantification of royal jelly quality by mid-infrared spectroscopy coupled with backpropagation neural network

Abstract

Royal jelly is rich in nutrients but its quality is greatly affected by storage conditions. To determine the quality of royal jelly accurately and quickly, a qualitative discrimination model was established based on the fusion of conventional parameters and mid-infrared spectrum, using support vector machine. The prediction models for three representative quality parameters were developed by the backpropagation neural network with various algorithms. The results demonstrated that the recognition rate of the multi-source information fusion model was increased to 100% when compared with that of the spectral data preprocessed by Savitzky-golay smoothing (95.83%). The mean square errors of the constructed model for moisture, water-soluble protein, and total sugar were 0.0032, 0.0058, and 0.0069, respectively. The constructed model had an ensured accuracy for the calibration set, with the correlation coefficient of prediction maintained at 0.9353, 0.9533, and 0.9563, which could meet the requirement of non-destructive rapid detection of royal jelly quality.