Principal Component Analysis and pattern recognition combined with visible spectroscopy in the classification of food quality

The online measurement of the colour of food internally and externally has already been shown to be an invaluable parameter in the process control of large industrial ovens. The system, described in this article is based on optical fibre technology is intended for accurate measurement of food colour. It employs artificial intelligence through the use of Neural Networks to make decisions regarding the cooking stage of the product. This paper examines the application of Principal Component Analysis, using Karhunen Loeve Decomposition, to the spectral data before applying the pattern recognition technique. With Karhunen Loeve decomposition it is possible to reduce the dimensions of this solution to a smaller subspace by only including significant data and thus eliminating redundant or highly correlated information. This method was tested on the following food products: Steamed Skinless Chicken Fillets, Roast Whole Chickens, Sausages, Pastry, Bread Crumb Coating and Char-grilled Chicken Fillets.