Título: Rede Morfológica Não Supervisionada-RMNS
Autores: Silva, Francisco de Assis Tavares Ferreira da; Banon, Gerald Jean Francis
Resumo: In this work we propose a new paradigm of Artificial Neural Net (RNA): the Unsupervised Morphologic Net (RMNS portuguese-speaking), characterized by competitive learning. This paradigm belongs to a class of translation invariant nets and it is based on gray level Mathematical Morphology (MM), Carpenter and Grossberg’s ART, and Kohonen net models. In the activation phase, the template matching operator is implemented using translation invariant MM elementary operators. These operators enable a robust pattern detection with respect to addictive or subtractive noise, and/or small deformations of the patterns to be recognized. In the training phase, the RMNS uses a Kohonen learning rule variation and a reset system inspired by Carpenter and Grossberger’s ART model.
Artigo em pdf: 4cbrn_095.pdf
Arquivo BibTex: 4cbrn_095.bib