A visual recognition model based on hierarchical feature extraction and multi-layer SNN

Abstract

In this paper, a visual pattern recognition model is proposed, which effectively combines hierarchical feature extraction model and coding method on multi-layer SNN. This paper takes HMAX model as feature extraction model and adopts independent component analysis (ICA) to improve it, so that the model can satisfy the sparsity of information extraction and the output result is more suitable for SNN processing. Multi-layer SNN is used as classifier and the firing of spikes is not limited in the learning process. We use valid phase coding to connect these two parts. Through the experiments on the MNIST and Caltech101 datasets, it can be found that the model has good classification performance.