Predicting Loquat Quality Using Visible, Near Infrared Spectroscopy and Artificial Neural Network

The main objective of this study is using Visible and Near-Infrared Spectroscopy (Vis/NIRS) as a simple, fast and non-destructive method, combined with Artificial Neural Network (ANN) in order to build three ANN models for predicting fructose, glucose and pH levels in loquat. A huge amount of spectral data characterizing each sample was extracted. In addition, by applying destructive procedures, soluble sugar content (glucose and fructose) and pH were determined. Feature selection was employed to select the most suitable wavelengths from the raw spectra. Afterwards, ANN was applied to quantify the three physico-chemical properties of loquat. Different feature selection algorithms were tested, namely Least Absolute Shrinkage and Selection Operator (LASSO), Interval Random Frog (iRF), Variable Iterative Space Shrinkage Approach (VISSA) and Interval VISSA (iVISSA). As a result, ANN models based on feature selection (VISSA) gave the best results compared to those obtained using full spectra. The performance of the ANN models in quantifying the three loquat properties were: a prediction determination coefficient (R2p = 0.97), (R2p = 0.98), (R2p = 0.98) and a standard error of prediction (SEPpH = 0.08), (SEPglucose = 0.30°Bx) and (SEPfructose = 0.34°Bx) for pH, glucose and fructose prediction respectively. The results indicate the power and efficiency of combining Vis/NIRS and ANNs to predict loquat quality. Furthermore, this research can be extended to other food products.