Abstract: Hyperspectral imaging technology was used to detect the contents of polysaccharides and total sugar as two important quality indicators for rapid evaluation of Chinese wolfberry. For this purpose, the optimal spectral waveband was explored. Firstly, the original spectra were preprocessed using three commonly used methods, namely multiplicative scatter correction, Savitzky-Golay smoothing and standard normal variate, and comparison of the results obtained showed that multiplicative scatter correction was selected to eliminate the scattering effect. Then the average spectral reflectance value was extracted for use as characteristic parameters from hyperspectral images in the effective wavebands, the visible wavebands, the near-infrared wavebands and the full wavebands based on the correlation coefficients and the spectral characteristics in different waveband ranges. Finally, BP neural network models were established based on different characteristic parameters to predict the contents of polysaccharides and total sugar in Chinese wolfberry. The results showed that the prediction model based on the full bands was the best one. The correct prediction rate of the model for polysaccharide content was 97.59% with a correlation coefficient of 0.997 4 and a root mean square error of 0.077 7. The correct prediction rate of the model for total sugar content was 100% with a correlation coefficient of 0.996 8 and a root mean square error of 0.250 6. Therefore, it is feasible to detect the contents of polysaccharide and total sugar in Chinese wolfberry by hyperspectral imaging technology.

Key words: hyperspectral imaging, BP neural network, Chinese wolfberry, polysaccharide, sugar