Utilização de espectroscopia no infravermelho médio, fusão de dados e métodos quimiométricos de classificação na análise de fraudes em carnes bovinas in natura

• Ano de defesa: 2015
• Autor(a) principal: Karen Monique Nunes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Espectroscopia no infravermelho médio; Fusão de dados; Adulteração de carnes; PLS-DA; Análise forense; Química analítica; Alimentos Adulteração e inspeção; Carne bovina; Espectroscopia de infravermelho
• Link de acesso: http://hdl.handle.net/1843/BUBD-A2KHKZ

The concern with the food safety and authenticity of meats is increasing in the last years, due to the occurrence of great fraud scandals. The addition of some salts and other adulterants, such as carrageen (a linear sulphated polysaccharide extracted from red edible seaweeds), can increase the meat water holding capacity, providing an economic fraud by weight gain. This dissertation aims the detection and characterization of adulterations in bovine meat in natura by determining five physico-chemical variables, the contents of protein, ash, chloride, sodium and phosphate, and using attenuated total reflection mid infrared spectroscopy (MIRS). The generated data were treated with non-supervised (principal component analysis, PCA) and supervised (partial least squares discriminant analysis, PLS-DA) classification methods. In the first application, 43 adulterated meat samples of different types of cuts were analysed, which were obtained from a real police operation. The models were built with these samples plus 12 control samples, which were guaranteed to be non-adulterated. A semi-supervised PCA model using the physico-chemical data detected 74% of the adulterated samples. However, better results were obtained with PLS-DA. A model built with only the physico-chemical data had a better predictive ability than a model built with the MIR spectra. But, the best model were obtained with the data fusion of the physico-chemical data and MIR spectra, which correctly classified all the control samples and provided only 4 false-negatives for the prediction of adulterated samples. A second data fusion model were constructed by combining the most predictive physico-chemical variable, chloride, and 8 spectral regions selected from an informative vector (VIPscores). This model provided somewhat lower prediction ability, but has the advantage of utilizing variables which were measured simpler, faster and at a low cost. In the second application, meat samples were adulterated in a controlled manner by the injection of 4 types of adulterants (water, tripolyphosphate, and mixtures containing salts, collagene, carrageen and maltodextrin (a complex carbohydrate obtained from starch). The purges, the liquid exudated after the meat thawing procedure, of 51 meat samples were analysed by MIRS and the constructed PLS-DA models were able to correctly classify all the non-adulterated samples, allowing in addition to identify the spectral regions more selective for each type of adulteration.