Snack quality evaluation method based on image features and neural network prediction

Quality maintenance of food products like snacks and chips is a challenging problem in the food processing industry. At present, it is typical for food companies to determine “quality’’ standards using a sensory panel of trained experts. Process operators sample and observe the product to ensure it is similar to target flavor and quality. This process is subjective and prone to differences. A neural network approach via image texture and shape features is investigated in this study for the evaluation of the quality of typical snack products. Although quality of food, especially snacks and chips, is very difficult to quantify, some external attributes of the product are indicators of the snack eating quality. External texture features (which can reflect internal structure), together with the size and shape features of snacks are used to describe the quality from a texture (mouthfeel) standpoint. A backpropagation neural network was trained with a large number of samples having texture and morphological features as input and sensory attributes as output. The network is shown to predict the sensory attributes of the snack quality with a reasonable degree of accuracy. The analysis is validated through a comparison of the predicted sensory attributes obtained by the network to those from a taste panel on untrained samples. The developed methodology can be used in the food industry to evaluate the snack quality in a nondestructive sense.