Desenvolvimento e validação de métodos para análise direta de alimentos usando ferramentas quimiométricas, espectroscopia no infravermelho e imagens digitais

Abstract

The main objective of this thesis was the development of methods using chemometric tools, infrared spectroscopy (near and mid) and digital image analysis obtained using low cost equipment for foodstuff quality control. In all applications, multivariate analytical validation was emphasized, a subject with growing importance in chemometrics. All the proposed methods were validated searching for an harmonization um national and international validation guides. Chapter 3 presents the development of methods for the determination of qualityparameters in cheese, in a real legal laboratory condition, in an official laboratory. PLS models were built using NIR spectra for the determination of moisture and fat in mozzarella cheese and moisture in minas cheese. The stability of the proposed models was evaluated over a year. The fat determination model for mozzarella presented erros between -8.7% and 8.1%, and the moisture determination model for the same cheese type presented smaller errors (-3.5% 2.9%). The moisture determination model for minas chesse presented errosranging from -6.5% to 9.8%. Chapter 4 present the development of a screening method for the simultaneous detection of five adulterants in raw milk using MIRS. Water, starch, sucrose, formaldehyde and sodium citrate were chosen as adulterants due to its presence in adulterated milk samples recently discovered in the last Federal Police operations. PLS-DA models were built for the individual identification of the five adulterants, even in samples that contained two or moreadulterants. All the the models presented a false negative rate below 5%, been the starch model the only exception. Chapter 5 describes the deveploment of a method for the determination of the artificial dye sunset yellow in beverages using digitais images obtained from a commercial flatbed scanner. The model presented prediction errors varying from -6.2 to 9.0%. This method was also validated, proving that RGB histograms obtained from digital images are able to be used as analytical data, with a series of advantages, such as quickness and simplicity. In chapter 6, a similar application to the one presented in chapter 5 was developed, but using a cellphone to obtain images of hard candies, and thisimages were used to quantify the concentration of the allura red dye. This model presented higher errors than the ones seen in the other models in this thesis (from -24.8% to 39.5%), but the great majority of the errors (80%) were between -20% and 10%.