Dynamic detection of fresh jujube varieties based on least square support vector machines

The present research was focused on dynamic detection of fresh jujube varieties by visible and near-infrared spectroscopy. In the part of data gathering, fresh jujube was moving at the constant velocity of 0.1 m · s-1 on the conveyor belt, and the visible and near-infrared diffuse reflectance spectrum (350~2500 nm) was captured. In the part of data processing, moving average was employed to eliminate the spectra noise. Then the least square support vector machines (LS-SVM) models were established with the original spectra and the de-noised spectra (500~1500 nm) of 255 samples in the calibration set. At last, the models those were the Original/LS-SVM model and the Smooth/LS-SVM were used to predict the varieties of 90 samples in the prediction set. The results showed that the discrimination accuracy (95.56%) for the fresh jujube varieties by Original/LS-SVM model was slightly higher than the discrimination accuracy (91.11%) for the fresh jujube varieties by Smooth/LS-SVM model. And the fitting result of the former was better than the fitting result of the latter. So the Original /LS-SVM model can be taken as a dynamic detection method to detect fresh jujube varieties, but the precision and stability in the model is needed to be further improved.