Aplicação de redes neurais artificiais na ciência e tecnologia de alimentos : estudo de casos

AUTOR(ES)

Isabella Peres Gualda

DATA DE PUBLICAÇÃO

2008

RESUMO

Artificial Neural Networks (ANNs) are a non algorithm computing method capable of solving complex problems, getting better results than mathematical methods. The artificial neural networks has been used in many areas of technology and food science, most of them in classification problems, prediction, pattern recognition and control. This study approach two different situations. The first one simulates the brining of prato Brazilian cheese and uses a radial basis function (RBF). These networks are considerate universal function approximation. The model that have the best result was develop with 7 input variables: Three dimensions of cheese (X, Y, and Z), time of brining, NaCl and KCl inicial concentrations in the brining and boundary conditions (stationary or with agitacion brine), 29 units in the hide layer and 13 K-Nearest neighbors. The simulation deviation was about 5,5% for NaCl and 4,4% for KCl.The second situation was an attempt to assort some horticultural based on the cultivation (conventional, hydroponic and organic). Multi Layer Perceptron (MLP) networks have been used to do that. The MLP has a great ability of generalization and is very used in classification problems. The topologies that acquired classification (100% in training and validation) was as follow: Strawberry, network with 12 input variables (Mg, Al, Fe, Mn, Co, Cu, fructose, sucrose, nitrate, lipid e carbohydrate) and 6 units in the hide layer. For crispleaf lettuce networks with 30 input variables (composition centesimal, sugars, all the minerals, sugars’s sum, minerals’s sum, nitrate, nitrite e sum of nitrate+nitrite) and with 10 and 13 units in the hide layer, was classificated when compared with conventional cultivation X organic cultivation and organic cultivation X hydroponic cultivation. For the comparation between conventional cultivation and hydroponic cultivation was used 13 input entries variables (Na, Mg, Al, Fe, Mn, Se, Hg, Pb, sugars’ sum, ash, lipids e energy) and 5 units in the hide layer. Using the cherry tomato the model that could be classified was the one that has used 15 input variables (Na, Mg, Al, Ca, Fe, Mn, Cd, Pb, nitrate, nitrite, sum of nitrate+nitrite, ash, lipids, proteins and energy) and 10 units when compared agaist organic and hydroponic cultivation. Yet, 11 input variables (Na, Mg, P, Ca, Fe, Mn, Zn, Cd, nitrate, sum of nitrate+nitrite e lipids) and 4 units in the hide layer when the conventional and hydroponic cultivation where compared each other. This study adds some evidence to the potential of the ANN application to modeling complex tasks in the control and simulation of food process and the capacity of data classification in food analysis.

ASSUNTO(S)

natural foods food - neural networks (computer science) tecnologia de alimentos - redes neurais (computação) alimentos - qualidade alimentos naturais alimentos - redes neurais (computação) food - quality

Documentos Relacionados

• Aplicação de redes neurais artificiais na previsão da produção de álcool
• Redes neurais artificiais: uma aplicação em petrofísica e estudo dos efeitos de estímulos persistentes
• Redes neurais artificiais e softness químico para classificação de grupos de moléculas : aplicação em aminoácidos.
• Redes neurais artificiais e softness químico para classificação de grupos de moléculas : aplicação em aminoácidos.
• Sirena : um simulador de redes neurais artificiais