Redes neurais artificiais para estimar a densidade básica de madeiras do cerrado

Abstract

The basic density of wood is an important property because it is related to the final product in the various uses of wood. However, its determination demands time and costs, which justifies the use of more refined techniques for its estimation, such as artificial neural networks (ANN). Our objective was to evaluate the use of artificial neural networks to estimate the basic density of species of cerrado stricto sensu with the use of Pilodyn and dendrometric variables. To compare the results obtained by ANN, regression models were adjusted. The best performing neural network was the one that used as input variables the depth of penetration (Pilodyn), species and diameter at 1.3 m above ground level, presenting R² = 0.72 and with root mean square error in percentage (RMSE%) = 5.69. The regression model presented R² = 0.72 and RMSE% = 9.19. The artificial neural networks can estimate the basic wood density of the studied species of cerrado stricto sensu with satisfactory results.