Utilização da modelagem matemática (redes neurais artificiais) na classificação de autotetraploides de bananeira (Musa acuminata Colla)

Abstract

The objective was to develop a methodology to enable the classification of banana crop subjected to induction of chromosome doubling using Neural Networks (NN). The data used in this study were taken from a thesis already presented, whose authors studied the correlation between fresh weight of leaf discs and DNA content. The NN was implemented by the ranking function. The learning rate and momentum term used were respectively equal to 0.01 and 0.2, the number of training epochs was 1000. These values were determined by trial and error. For training, 90% of the plants were employed, and for validation, 10% of the total of 114 autotetraploids artificially produced by exposure to antimitodic agent colchicine. The NN correctly classified 10 of the 11 samples used for validation. Kappa statistics was 63.33%, which indicates that the NN can be further improved. The artificial neural network-type Multi Layer Perceptron is effectively implemented in the pre-selection of polyploid desirable banana Tong Dok Mak.