Aplicação de imagens orbitais na espacialização do volume de Eucalyptus sp. via redes neurais artificiais
Abstract
The sustainable agriculture of forest plantations demands the permanent monitoring of the quantity of processed wood, thus becoming difficult to monitor large planted areas by using only manual procedures. Therefore, in this research artificial neural networks (RNA) of multilayer perceptrons, were modeled to estimate the spatial of wood volume in a Eucalyptus sp plantation located in the state of Mato Grosso del Sur in the Central-West region of Brazil. As input variables in the RNA spectral bands, the textures of the bands obtained with Gray Level Co-Occurrence Matrices and vegetation index were used, which were derived from digital satellite image Spot 6. The resulting RNA with the best performance presented
accuracy of 93.32% and coefficient of determination of 0.9761. However, this network presented a mean square error of 16.32% (RMSE de 7.85 m3ha-1), but with a unbiased distribution of the residuals. So, the model showed to be suitable to monitor the amount of wood in large areas without overestimating or underestimating the prediction. Compared with different machine learning methods using the same variables, the built network seems to have a higher precision and accuracy. Even in the neuronal models using only spectral bands and vegetation indexes, a better performance was evidenced, showing that the textures contribute in the improvement of predictions.