Classification model in different forest strata in a floodplain environment using artificial neural networks

Abstract

The Amazon forest presents different forest strata, due to its heterogeneous structure. In which these strata can vary in upper, middle and lower. Knowledge about the different patterns of vertical structures found in the forest is extremely important for understanding the vegetation dynamics, influencing forest conservation strategies. In order to optimize the process of classifying the different types of strata, the objective of the present work was to use artificial neural networks (ANNs) to classify these strata. Two resilient propagation algorithms (Rprop + and Rprop-) were used, in four different configurations of input variables. The training and testing of the eight RNA models were performed using the R software. The models were evaluated using a confusion matrix. In which models with inputs: HT, DAP and QF; HT, DAP and only HT from the Rprop + algorithm obtained 100% correct answers in the classification of strata. Demonstrating a high rate of learning, reliability and generalization of data.