Design of neuro-fuzzy based modular architecture for pattern recognition
S. Mercy Shalinie · 2003
Many pattern recognition methodologies and design techniques have been developed over the years and new approaches continue to emerge. For the solution of complex problems in pattern recognition and more generally machine intelligence, involving heterogeneous data sources of both numeric and symbolic information, the fundamental design philosophy is to employ hybrid methodologies rather than attempting to produce the solution using a single paradigm. The main objective of this paper is to integrate heterogeneous methodologies for intelligent solution of pattern recognition problems. A modular neural structure where a self-organizing Boolean neural network architecture is used as a front end processor to a feedforward neural architecture based on goal seeking principles is described. Fuzzy logic controller is included to provide decisions in uncertain conditions. The characteristics of the proposed methodology is illustrated by considering its application to a terrain image understanding system from multiple sensors and spatial databases.