Selective attention adaptive resonance theory (SAART) neural network for neuro-engineering of robust ATR systems

Abstract

This paper presents a novel real-time artificial neural network called selective attention adaptive resonance theory (SAART). SAART is a self-organising neural model that is based on a real-time neural theory of sensory information processing, high level biological vision, visual perception, object recognition and self-organised learning in complex sensory environments. SAART embeds new neural mechanisms (selective presynaptic facilitation and selective presynaptic inhibition) into a dynamic neural network that is capable of selective attention and visual perception. These new features enable the network to learn effectively in noisy inputs and to recognize familiar 2-D patterns of neural activity (representations of object's boundary) in complex, cluttered and noisy background. SAART also provides fundamental neural design principles and neuro-engineering foundations for the design of robust automatic target recognition and neuro-computational vision systems.