Rapid and non-destructive estimation of chemical properties of date syrup using visible-near infrared hyperspectral imaging combined with machine learning predictive models.

In this study, an innovative hybrid approach combining hyperspectral imaging and artificial intelligence was developed to replace costly and time-consuming laboratory methods of determining chemical properties of date syrup. Hyperspectral data related to acidity, Brix, fructose, HMF, pH, and viscosity were analyzed using three machine learning models: Artificial neural network, partial least squares regression, and support vector regression. These models were evaluated under two strategies, with and without feature selection step. Among the models, the Artificial Neural Network demonstrated superior performance compared to the other methods. The results revealed that the optimal performance was achieved with four neurons in the hidden layer and the use of the Levenberg-Marquardt activation function. The prediction accuracies for the mentioned properties were 100%, 99.99%, 100%, 99.99%, 100%, and 99.99%, respectively. In the strategy without feature selection, the prediction accuracy for all chemical parameters reached 100%. These findings highlight the capability of hyperspectral imaging and artificial neural networks in estimating chemical properties of date syrup.