USAGE OF THE NEURAL NETWORK TO PREDICT MEAT TENDER- NESS APPROACH

Resumo

Meat tenderness is one of the main qualitative attributes sought after by consumers when purchasing
beef. Among the various properties of meat, tenderness is one of the most appreciated by the public that buys this
type of food. The tenderness of the meat is influenced by several factors in the constitution, from genetic, food,
and environmental factors is the tenderness evaluated in the post-mortem of the animal is a direct and objective
measure to be quantified. This property is obtained through mechanical tests already known in the literature, by
obtaining the shear force necessary to break the set of muscle fibers of the tissue examined. In this way, this paper
estimates the tenderness of the meat in a non-destructive way through the use of computational techniques using
machine learning, such as the use of artificial neural networks, to quantify the dependence of variables that can be
obtained without the destruction of the sample, but that obtain a satisfactory approximation in obtaining the shear
force of the analyzed beef tenderloin samples. Thus, to evaluate the tenderness parameters, measurements made
from tests were used for the values of PH, sample color, hot carcass weight, loin eye area, breed, sex, infrared and
ultraviolet images, and shear force of fillet samples. In this way, the objective of the neural network was to find the
dependence of the variables on the shear force necessary to break the fibers of the sample. For this, a cross-data
model known as Random Forest was used for training neural network was performed based on the present data, and
an average error of 20% was obtained compared to the value obtained for the shear force through the mechanical
test. It was observed that the shear force prediction values are directly influenced by the number of variables to be
introduced in machine learning, as well as the number of observed samples.