Using of Artificial Intelligence and Image Texture to Estimate Desiccation Rate of Quince Fruit

Image processing can quantitatively define the visual properties of foods. Textural properties provide lots of information from a captured image which can be used for further system design. The combination of artificial neural networks (ANN) and machine vision can provide simple solutions to problems associated with agricultural products processing. In this research, the effect of drying on the parameters related to the image texture of sliced quinces was investigated dried at constant temperature level of 40 in an oven dryer. Images were captured from samples at several intervals during drying period. Co-occurrence matrix was computed separately for L*, a*, b* color components as well as grey-level images. Five textural features were extracted from each matrix. Dehydration causes contraction on quince slices which in turn produce special texture look on the dried sample. A feed-forward back-propagation artificial neural network (ANN) with three hidden layers and different transfer functions was developed and trained for instantaneous prediction of moisture content of fruit being dried. The ANN with tangent sigmoid transfer function and Levenberg-Marquardt learning algorithm was the most efficient networks for prediction of drying behaviors of quince slices with an R2 of 0.994 and uncertainty of 0.12%.