Aplicação de técnicas espectroscópicas vibracionais e imagens hiperespectrais na detecção de fraudes em carnes bovinas in natura

• Ano de defesa: 2019
• Autor(a) principal: Karen Monique Nunes
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ICX - DEPARTAMENTO DE QUÍMICA; Programa de Pós-Graduação em Química; UFMG
• Palavras-chave em Português: Análise forense; Adulteração de carnes; Análise discriminante por mínimos quadrados parciais PLS-DA; Espectroscopia de absorção no infravermelho médio FTIR; FT-Raman; Imagens hiperespectrais; Reflectância total atenuada ATR; Attenuated Total Reflectance ATR; Forensic analysis; Meat adulteration; Hyperspectral analysis; Partial Least Squares PLS-DA; Fourier Transform Infrared FTIR; Química analítica; Quimica legal; Espectroscopia; Espectroscopia de infravermelho; Raman, Espectroscopia de; Quimiometria
• Link de acesso: http://hdl.handle.net/1843/30978

Large and recent meat fraud scandals have raised concerns about food security in Brazil. The addition of salts solutions and others adulterants, such as carrageenan (a linear sulphated polysaccharide extracted from red edible seaweeds) and maltodextrin (rapidly absorbing polysaccharide) increase the meat's water-holding capacity, leading to economic fraud. The Thesis had as main objective the detection and identification of adulterations in samples of bovine meat in natura using mid-infrared absorption spectroscopy (FTIR) and Raman spectroscopy. Data were processed using supervised classification chemometric methods (partial least squares discriminant analysis, PLS-DA). In the first application (FT-Raman), the global model for prediction adulterated meat samples was considered a poor model for the systematic detection of fraud in study with reliability rate (RLR) equal 67%. However, the purge analysis was satisfactory (RLR = 80%). Confidence intervals were estimated for individual prediction values using the bootstrap algorithm. In the second application (FTIR), the best results were obtained for the global fraud detection model with RLR above 91%. The model optimization, producing a soft version PLS-DA, was performed by outliers detection with RLR equal to 93% and correct prediction of 100% of the fraudulent samples with an unmodulated adulterant, maltodextrin. Outliers detection was performed by calculating confidence limits for y/class predicted values and by analyzing of the Hotelling’s T2 plot versus Q residues. Methods employing Raman image spectroscopy and curve resolution method were shown proved to be efficient for detecting fraud in meat in natura (third application). Individual prediction models, class prediction (salts and polysaccharides) and a global model were obtained with promising results.