Redes neurais artificiais na estimação de volume de mogno africano (Khaya ivorensis)

Abstract

African mahogany has noble wood of high economic potential and extensive use in the furniture industry, civil construction and production of panels and laminates presenting good workability. In order to estimate the volume of individual trees in an African mahogany (Khaya invorensis) stand in the municipality of Pirapora, Minas Gerais, the objective of this work was to evaluate the use of Artificial Neural Networks in Brazil. Through the NeuroForest 4.0 software, training of 540 nets was performed for selection of algorithm with data separation of 60% for training and 40% for validation, configured with: input variables total height (Ht), diameter at 1.3 m (DAP) and diameters at different heights (0.0, 0.3, 1.3, 2.0), the output was volume, in the architecture was used a hidden layer, the activation function used was the sigmoidal and the stopping criterion was 3000 cycles. The algorithm with better correlation and square root mean error for training and validation was used to train 1500 RNAs with the same configurations mentioned above, varying the number of cycles. The networks that presented the best results were those trained with input variable diameter 1.3 m from soil (DBH) and total height (Ht), recommended to larger stands, training of Artificial Neural Networks for volume estimation in Khaya ivorensis, from the input variable diameter at different heights can be used in smaller stands by dispensing the height measurement of the trees.