Open Journal Systems

IDENTIFICAÇÃO DE PONTOS DE APOIO PRÉ-SINALIZADOS COM O USO DE REDES NEURAIS ARTIFICIAIS E CORRELAÇÃO

ROMUALDO WANDRESEN, EDSON APARECIDO MITISHITA, JOSÉ BITTENCOURT DE ANDRADE

Abstract



Este trabalho apresenta uma introdução às redes neurais artificiais como uma
alternativa para a identificação de pontos de apoio pré-sinalizados. Inicialmente é
feita uma rápida revisão bibliográfica sobre redes neurais artificiais, onde são
apresentados os modelos de um neurônio, o perceptron, as redes feedforward e
retropropagação, bem como um resumo de seu algoritmo. Em seguida são realizadas
as etapas que compõem as redes neurais artificiais: coleta de dados, treinamento e
simulação dos dados. São analisados dois tipos de pontos de apoio pré-sinalizados:
em forma de cruz e em forma circular. São apresentados, examinados e comparados
os resultados das identificações dos pontos pré-sinalizados, nas duas formas em cruz
e circulares usando códigos fonte em MATLAB e em Visual C++, tanto para a fase
de treinamento dos dados, quanto para a fase de testes dos mesmos, usando em
conjunto a técnica de correlação estatística, com o objetivo de evitar ambigüidades.

Identifying Pre-signalized Points by the Use of Artificial Networks and Correlation

Abstract


This work presents an introduction to artificial neural networks as an alternative for
the signalized point recognition. At first a fast bibliographical revision on artificial neural networks is done, where the models of a neuron, perceptron, the networks are
presented feedforward and backpropagation, as well as a summary of its algorithm.
After that the stages are carried through that compose artificial the neural networks:
it collects of data, training and simulation of the data. Two types of signalized point
are analyzed: in cross form and circular form. They are presented, examined and
compared the results of the identifications of the signalized point are compared, in
the two form using codes source in MATLAB and Visual C++ , for the phase of
training and for the phase of simulation of the data, using as well the technique as of
correlation statistic for avoiding ambiguities.