New vision technology for multidimensional quality monitoring of food processes

Spectroscopy and spectral imaging in combination with multivariate data analysis and machine learning techniques have proven to be an outstanding tool for rapid analysis of different products. This may be utilized in various industries, but especially rapid assessment of food products in food research and industry is of importance in this thesis. The non-invasive spectroscopic imaging techniques are able to measure individual food components simultaneously in situ in the food matrix while pattern recognition techniques effectively are able to extract the quantitative information from the vast data amounts collected. Underlying qualitative features (latent structures) are extracted from multivariate spectral data in order to quantify desired quality parameters properly. Specically multispectral imaging which has been explored to a lesser extent than ordinary spectroscopy, having the possibility to exploit the inherent heterogeneity that exists in foodstuffs have been investigated here. An extra feature obtained by combining spectroscopy, imaging and chemometrics is exploratory analysis. This is central in food research, since novel hypotheses about the food systems under observation may be generated using this inductive analytical approach. For the food industry it is an additional advantage that the fast, non-invasive, remote sensing nature of the spectroscopic imaging methods allows on-line measurements. In this way spectroscopic imaging in combination with advanced data analysis meets the high throughput needs for quality control, process control and monitoring. In this Ph.D. project the possibilities provided by spectroscopic imaging and chemometrics have been utilized to improve the analysis and understanding of different food products. The work is presented in seven papers and two additional technical reports which make up the core of the thesis. Furthermore an introduction together with a linking of the contributions is presented in this thesis. The papers puts an emphasis on the use of multispectral imaging in the baking industry where especially the non-enzymatic browning appearance and features related to this are highlighted. These are features such as colour, water content and internal structure of bread. A paper presenting enzymatic browning in pre stir fried and thawn vegetables is also presented showing that imaging techniques such as the one investigated in this thesis is able to detect even subtle colour changes. The possibility for quantifying early as well as late spoilage in raw pork meat is investigated where use of the heterogenetic structure is utilized to obtain good results on predicting sensory evaluations as well on laboratory analysis. Colour in other settings such as in the shery industry is equally important, and a paper describing detection of cartenoid pigment in trouts using spectral images shows promising results. Finally, two technical papers present possible ways of mapping multispectral images to a visible colour space, as well as how an alternative multispectral imaging system, making use of lters, may be used to design new more broad ranged filters. Fewer filters will increase the speed of such systems. Methods for solving such problems is to the knowledge of the authors rarely covered in the literature.