Metodologia inovadora para determinação da densidade básica da madeira utilizando imagens multiespectrais e câmera mapir RGNIR

Abstract

Basic wood density directly influences yield, operational costs, and final product quality, making it an essential parameter in the forestry sector. However, traditional methods for its determination are time-consuming, destructive, and costly, which highlights the need for faster and more economical alternatives. This study used multispectral images obtained with a Mapir Survey 3 (RGNIR) camera to estimate the basic wood density of Eucalyptus grandis and Pinus oocarpa. A total of 142 samples were analyzed, with images captured at four drying stages: saturated, naturally drying, equilibrium moisture content, and oven-dry. The images were processed using ENVI Classic software to extract reflectance; Minitab 16 was used for correlation analysis, and Weka 3.8.6 software was used to fit machine learning models: Linear Regression, Multilayer Perceptron (MLP), Random Forest (RF), and Support Vector Machine for Regression (SMOreg). Reflectance in the cross-sectional plane, especially in the red spectral band, showed the strongest correlation with basic density. The Random Forest model achieved the best results, with correlation coefficients above 83% and 81%, and RMSE below 11% and 12% for the training and cross-validation datasets, respectively. The MLP, SMOreg, and Linear Regression models also performed satisfactorily, particularly with saturated wood. This research demonstrates the feasibility of estimating basic wood density through multispectral imagery, offering a rapid, non-destructive, low-cost, and easy-to-apply methodology for the forestry sector.