Multiple HONG network fusion by fuzzy integral

Author

Atukorale, Ajantha S. ; Suganthan, P.N.

Author_Institution

Dept. of Comput. Sci. & Electr. Eng., Queensland Univ., Qld., Australia

Volume

2

fYear

1999

fDate

1999

Firstpage

718

Abstract

Describes the Hierarchical Overlapped Neural Gas (HONG) architecture, which is built on the basis of the neural gas algorithm. The proposed architecture generated multiple classifications for every sample data presented, and these are registered as the confidence values of each network. A method-of-evidence fusion technique based on the fuzzy integral was used to combine the individual classifications produced by the proposed architecture. Finally, three classifiers based on three different feature sets were trained, and then combined using the fuzzy integral. An excellent recognition rate of 99.48% was consequently obtained for the NIST SD3 database

Keywords

fuzzy set theory; integration; learning (artificial intelligence); neural net architecture; pattern classification; sensor fusion; NIST SD3 database; classifier training; confidence values; evidence method; feature sets; fuzzy integral; hierarchical overlapped neural gas architecture; multiple HONG network fusion; multiple classification generation; neural gas algorithm; recognition rate; Feature extraction; Lattices; NIST; Pattern recognition; Programmable control;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Information Processing, 1999. Proceedings. ICONIP '99. 6th International Conference on

Conference_Location

Perth, WA

Print_ISBN

0-7803-5871-6

Type

conf

DOI

10.1109/ICONIP.1999.845684

Filename

845684