Data mining and pixel distribution approach for wood density prediction

Abstract

The wood material has strategic importance in economic development. Innovations are the basic premise of commercial success in the wood industry, as in all industries. The density of wood provides valuable information about the physical and mechanical properties of the wood, and it is also directly related to the productivity in the forest industry. Many non-destructive test studies have been conducted to evaluate the physical properties of wood structures. This study was conducted to predict the density of wood in the species of oak (Quercus robur) and beech (Fagus orientalis L.) using the number of pixels in a grayscale image and data mining. To this purpose, pixel density of data was processed with the data collected from the images of wood specimens. This data was used as descriptor variables in artificial neural networks and random forest algorithm. The designed artificial neural network model and random forest algorithm allowed the prediction of density with an accuracy of 95.19% and 96.36%, respectively for the testing phase. As a result, this study showed that pixel density and data mining have the potential to be used as an instrument for predicting the density of wood.