Pattern recognition of a spiking neural network based on visual motion model

Abstract

With the rapid development of artificial intelligence, deep learning which has been broadly applied on the image processing, pattern recognition and data mining. However, it requires huge amounts of data and computing power. As we all know, the human brain is very complex but effective with much lower energy consumption. It is of great significance to process information with reference to the brain processing mechanism which not only helps us to understand how the brain works, but also can build smart chips with lower power consumption. In this paper, images preprocessed by the visual motion model and mapped into the V2 layer with different orientations, and then, we train the connection between V2 and output by supervised STDP rule. The results show that we can achieve the same recognition accuracy with fewer training samples, which contributed by the visual model preprocessing. The visual preprocess can amplify the spatiotemporal information and highlight the feature of images.