Prediction of texture characteristics in extrusion food based on computer vision and artificial neural networks

The surface color values ( HIS & L*, a*, b*) and texture characters of extrusion products were measured by using computer vision system and texture profile analysis. The correlation between color values and texture characteristics was analyzed by using linear fitting model as well as predicted texture characteristics from color value based on back-propagation neural networks.Results obtained from linear fitting model between the texture and surface colors values showed that the hardness and gumminess scores were indirectly reflected through a*value and Intensity due to their high correlation coefficients of 0.9558, 0.9429, 0.9741 and 0.9619, respectively. The coefficients ( R2) of springiness with a*value and Intensity were over 0.8675 and 0.8320, respectively. A back-propagation artificial neural network ( BP-ANN) with 2 hidden layers, which derived from hardness, gumminess, a*and Intensity, was trained for simulating and predicting. The number of neural in each hidden layers were 20, 20 ( RMS = 4.25%) and 20, 40 ( RMS = 3.85%) in two experiments, respectively.A desirable and accurate prediction BP-ANN between hardness, a*and gumminess, Intensity were established with R2= 0.9671, 0.9770 and 0.9766, 0.9856 in two experiments.Results of this study indicated that texture characteristics of extrusion food could rapidly and accurately predicted through their surface color values by using computer vision system and artificial neural networks.