Classificação de alfaces e barras de cereais a partir da espectroscopia NIR e análise discriminante linear

• Ano de defesa: 2014
• Autor(a) principal: Brito, Anna Luiza Bizerra de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraí­ba; BR; Química; Programa de Pós-Graduação em Química; UFPB
• Palavras-chave em Português: Barras de cereais; Alfaces; Espectroscopia no infravermelho próximo; Análise Discriminante Linear - LDA; Cereal bars; Lettuce; Near infrared spectroscopy; CIENCIAS EXATAS E DA TERRA::QUIMICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/tede/7125

The search for a better quality of life has led to increased consumption of foods with fewer calories, high in fiber and vitamins, and obtained from different forms of cultivation. Amid these foods, there are cereal bars and lettuce, foods that are easily accessible, widely consumed and have high nutritional values. Like any other food, require efficient methods that can ensure its quality. Thus, the need for rapid, accurate analytical methods and low cost, which can help to identify and classify these foods safely arises. Within this perspective, this paper makes use of Near Infrared Spectroscopy (NIR) combined with Linear Discriminant Analysis (LDA) to classify samples of cereal bars and lettuce. A total of 121 samples of cereal bars, three distinct types (conventional, diet and light) and 104 samples of three different types of lettuce cultivation (conventional, organic and hydroponic) was used. The acquisition of the spectra was made on equipment Spectrum 400 (Perkin Elmer) with accessory NIRA (Near Infrared Reflectance Acessory) in the range 10000 - 4000 cm-1. Classification models were constructed by combining the LDA algorithms and variable selection: Stepwise (SW), Successive Projections Algorithm (SPA) and Genetic Algorithm (GA). Strategies for pre - processing data were evaluated and the efficiency of the models was determined from the of correct classification rate (CCR) for the full set of samples and the test set. For both matrices the model that generated a better CCR was the GA-LDA valued 95% to matrix of the cereal bars and 97.1 % for array of lettuces, both based on the total set of samples (training, validation and testing). Regarding the set of test models presented results of CCR with performance of 90.3 % and 95.4 % for matrices of cereal bars and lettuce respectively.