Espectrofotometria FTIR (Fourier Transform Infrared) e técnicas de aprendizado de máquina para a detecção de fraude por adição de soro de queijo ao leite cru.
Ano de defesa: | 2021 |
---|---|
Autor(a) principal: | |
Orientador(a): | |
Banca de defesa: | |
Tipo de documento: | Tese |
Tipo de acesso: | Acesso aberto |
Idioma: | por |
Instituição de defesa: |
Universidade Federal de Minas Gerais
Brasil VET - DEPARTAMENTO DE TECNOLOGIA E INSPEÇÃO DE PRODUTOS DE ORIGEM ANIMAL Programa de Pós-Graduação em Ciência Animal UFMG |
Programa de Pós-Graduação: |
Não Informado pela instituição
|
Departamento: |
Não Informado pela instituição
|
País: |
Não Informado pela instituição
|
Palavras-chave em Português: | |
Link de acesso: | http://hdl.handle.net/1843/38868 |
Resumo: | Fraud in milk causes economic damage and potential health risks to consumers. Cheese whey addition to milk is one of the major frauds, done to increase the volume provided at a reduced cost. Methods for fraud detection usually are laborious, expensive and time-consuming. The Infrared Spectroscopy (FTIR) technique presents itself as a promising alternative for the identification of this type of fraud, mainly because a large number of analytical milk data can be obtained and evaluated in the Laboratories of the Brazilian Network of Milk Quality Analysis Laboratories (RBQL), accredited by the Ministry of Agriculture, such as the Milk Quality Analysis Laboratory of the Veterinary School in the “Universidade Federal de Minas Gerais” (LabUFMG). Among the methods used for machine learning are the Artificial Neural Networks (ANN) and the Decision Trees, which have great potential for use as tools for evaluating possible adulterations of milk with cheese whey. The objective of this work was to use computational methodologies capable of identifying adulteration in milk by adding whey cheese to samples analyzed by FTIR spectroscopy. 585 samples of raw milk diluted with whey obtained from the manufacture of Fresh Minas cheese were used in concentrations: 0%, 1%, 2%, 5%, 10%, 15%, 20%, 25% and 30%. The samples were stored at 7°C, 20°C and 30°C for 0, 24, 48, 72 and 168 hours and analyzed in an FTIR device from LabUFMG. Compositional results of 585 samples of authentic bulk tank raw milk were added to the experimental samples and utilized for the “machine learning” models training, test, and validation. Different classifiers were used to recognize the compositional and or spectral changes caused by the addition of cheese whey to milk. The composition records and the spectra resulting from the analysis of the FTIR equipment were organized in order to serve as an entry for the classification task in the ANN models and Decision Trees. The classification was performed using two types of data, for three types of models: the numerical components were analyzed by ANN and Decision Tree and the total infrared spectrum was analyzed by Convolutional Neural Network (CNN). The proposed methodology was able to generate average accuracy of 97,8%, when using composition data in the ANN, and 96,2% using Decision Tree. Using the spectrum data (CNN), the total accuracy rate of the model was 93,84%. The use of FTIR associated with Artificial Neural Networks was efficient to differentiate samples with milk and added with cheese whey. |