Título: Redes Neurais Artificiais Invariantes à Rotação de 10º e a Translação Utilizada no Reconhecimento de Figuras Geométricas
Autores: Queiroz, Aderlon Marcelino; Flôres, Edna Lúcia
Resumo: This paper deals with the recognition of geometric figures throgh the use of Neural Networks Artificial (NNA’S). Before the geometric figures are trained and can be recognized by the NNA, in order to archive a satisfactory result, these figures are submitted to a translation and rotation of 10 o by an invariant network. This invariant network is constituted of a single neuron ADALINE, which has its synaptic weights translated and rotated. This article is composed of five main aspects, firstly is done a general revision on the works that use NNA ́s and shown pattern recognition techniques for this purpose. The second item, shows the method used to make the ADALINE neuron invariant to a translation and rotation of 10 o The third aspect, deals briefly with the NNA training and recognition algorithms. The fourth item, shows the results, and finally the fifth presents the conclusion on this work.
Código DOI: 10.21528/CBRN2003-063
Artigo em PDF: 6CBRN_063.PDF
Arquivo BibTex: 6CBRN_063.bib