Using three levels of classification to enhance the impact of standard neural networks

Abstract

The human visual system contains a hierarchy of modules that participate in visual perception at hyperboloid, primary and sub-classification levels. During previous decades, a different computational model has been proposed to the hierarchical treatment of the anterior cortical feeding of the visual cortex, but many important characteristics of the visual system have been ignored, such as Actual Neuronal Learning and Processing Mechanisms. We propose a mathematical model to recognize objects at different classification levels, Where an escalating neural network equipped with the reinforcement learning base is used as a modular unit at each classification level. Each unit solves the object recognition problem at each classification level, based on the first class-specific neuron spike in the last layer, without using any external classifier. According to the information required at each classification level, the filter band-pass images are used. The performance of our proposed model is evaluated through various evaluation criteria with three standard data sets and a significant improvement in the accuracy of our proposed model recognition was achieved across all trials.