Application of image processing and artificial neural networks as a non-destructive approach to prediction of fat content and classification of camel meat

Camel meat can be a suitable alternative for other red meat types in human nutrition, due to its low cholesterol and low-fat content and the appropriate protein content. This research aims to investigate and evaluate the fat content and freshness of camel meat using machine vision technology as a non-destructive method. Therefore, using image processing as a non-destructive method and Soxhlet device as a destructive method, the amount of fat content was predicted, and also the freshness was classified for camel meat. In the image processing section, 108 textual features and 39 color features were extracted in the RGB, HSV, HIS, and CIElab color spaces. Moreover, to predict the freshness and quality of meat, feed-forward back propagation artificial neural networks with one and two hidden layers, a various number of neurons, and threshold functions were used. Also, according to the regression diagram of fat content obtained from the destructive method (fat content obtained from Soxhlet device) with fat content obtained from non-destructive method (machine vision), the coefficient of determination and accuracy between them achieved 0.841. The results of the evaluation of the neural networks showed that the best desirable network for classification based on freshness is a one-hidden layer network with topology 147-3-1, tangent-sigmoid transfer function at hidden layer and purelin transfer function at output layer (R2= 0.996), and also to prete of fat content the best network is two-hidden layer network with linear, log-sigmoid, log-sigmoid transfer function at first hidden layer, second hidden layer and output layer respectively (R2= 0.99). Therefore, the results of this study show that the proposed system with the help of machine vision technology can predict the freshness and fat content of camel meat with acceptable accuracy.