Avaliação dos impactos dos fatores de pré-abate no dripping test de frango utilizando redes neurais artificiais

• Ano de defesa: 2021
• Autor(a) principal: Silva, Thiago Flores
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Londrina; Brasil; Programa de Pós-Graduação em Tecnologia de Alimentos; UTFPR
• Palavras-chave em Português: Indústria avícola; Frango de corte; Água - Absorção e adsorção; Redes neurais (Computação); Poultry industry; Broilers (Chickens); Water - Absorption and adsorption; Neural networks (Computer science); CNPQ::CIENCIAS AGRARIAS::CIENCIA E TECNOLOGIA DE ALIMENTOS::TECNOLOGIA DE ALIMENTOS; Tecnologia de Alimentos
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/26552

The growing increase in the world consumption of chicken meat has, as a consequence, an increase in production and consumer demand for quality, taking into account sensory attributes. Meat quality is controlled by physical-chemical parameters of appearance, texture, pH and succulence and is directly linked to the bird's post mortem biochemical reactions. With the increase in demand, companies need to remain competitive. Within the poultry slaughter process, during cooling in water immersion, each carcass can gain up to 8% in water absorption and the control parameter is the loss by dripping test, not exceeding 6%, according to the legislation of the MAP. A recurrent problem in the industry is the non-compliance with the dripping test, exceeding 6%, generating a high economic loss. The present work aimed to evaluate the impacts and identify the pre-slaughter factors that influence the water absorption of chicken carcasses, using artificial neural networks. For the development, the Multilayer Perceptron algorithm that composes the program Weka version 3.8.3 was used to identify the pre-slaughter variables. With the study it was found that the Percentage Split method presented better results than the Cross Validation, since it uses 70% of the data for ANN training, unlike the Cross Validation which uses 100% for validation, not exercising the training of the network. The best network structure found was with two hidden layers with 20 and 10 neurons per layer with 500 epochs, resulting in correlation (R) of 0.6138 and RMSE 0.1064. The variables with the greatest impact were absorption coefficient, house temperature, outdoor temperature, house waiting time, transport distance, demonstrating that the thermal stress of birds has a direct influence on the water absorption of chicken carcasses.