Application of wavelet transform coupled with artificial neural network for predicting physicochemical properties of osmotically dehydrated pumpkin Original Research Article

Since physical properties including deformation and color changes are important attributes of a food product and are generally used by customers to decide whether to buy or consume the product, the ability to predict the physical properties of a food product subjected to different drying methods and conditions, hence the ability to optimize the drying processes, are of great interest. The objectives of this research were to predict the moisture content, product deformation and color changes (ΔE) of osmotically dehydrated pumpkin undergoing hot air drying via the use of combined wavelet transform-artificial neural network. A multiple layer feed-forward neural network was established to predict the physical properties based on inputs of wavelet coefficients and drying time. Several pre-osmosed solutions were used including sorbitol, glucose and sucrose solutions. Optimized ANN models were developed for sorbitol, glucose and sucrose solutions based on 1–2 hidden layers and 10–41 neurons per hidden layer. ANN models were then tested against an independent dataset. Measured values of moisture content, deformation, in terms of the so-called Heywood shape factor, and ΔE were predicted with R2 > 0.9957 in all cases, except when pumpkin was osmotically dehydrated with glucose. The WT-ANN models were found to estimate the Heywood shape factor and ΔE with%MRE smaller than the ANN models alone in most cases.