Use of artificial neural network and image analysis to predict physical properties of osmotically dehydrated pumpkin

The objectives of this research were to predict, using neural networks, the color intensity (Delta E), percentage of shrinkage as well as the Heywood shape factor, which is the representative of deformation, of osmotically dehydrated and air dried pumpkin pieces. Several osmotic solutions were used including 50% (w/w) sorbitol solution, 50% (w/w) glucose solution, and 50% (w/w) sucrose solution. Optimum artificial neural network (ANN) models were developed based on one to two hidden layers and 10-20 neurons per hidden layer. The ANN models were then tested against an independent data set. The measured values of the color intensity, percentage of shrinkage, and the Heywood shape factor were predicted with R-2 > 0.90 in all cases, except when all the drying methods were combined in one data set.