Artificial Neural Networks for Volume Estimation of Caatinga Tree


Caatinga in Bahia state comprises of transiction ranges which species show a greater size features, distinguishing from shrubby feature which is more common in arid regions. Meanwhile, few information are gathered about that phytophysiognomy, primarily when volume estimation is under dicussion. Therefore, main subject of this paper is to design artificial neural networks (ANN) and evaluate their volume estimation performance of caatinga biome species. Thus, in order to gather total and comercial volume 166 tress where mesured in a caatinga tree area and and many Multilayer Perceptron architectures (MLP) where trained with backpropagation algorithm and compared to volumetric models. Neural Networks presented greater statistics results when compared to linear and non-linear equations and error were grouped in lesser ranges. As neural network as equations shown strong bias to volume overestimation. As a way to improve neural network estimations we conclude more variables migth be included.

In Proceedings of 3rd Brazilian Meeting of Forest Measurement (MensuFlor)

Text in Portuguese, Original Title: Redes neurais artificiais para estimativa de volume na Caatinga arbórea.

Fabrizzio Soares
Fabrizzio Soares
Associate Professor and CS Chair

Fabrizzio Soares is a professor of Computer Science, Information Systems and Software Engineering at INF/UFG. His research interests include Computer Vision, Human Computer Interaction, Machine Learning and Programming topics. He is the leader of the Pixellab group, which develops solutions for accesibilty, Precision Agriculture, and Interactive Systems.