Neuromorphic visual target classification method based on improved spiking neural network

Abstract

The invention discloses a neuromorphic visual target classification method based on an improved spiking neural network. The method comprises the following steps: S1, acquiring a neuromorphic visual target classification data set; s2, pulse event stream serialization aggregation: aggregating the space-time pulse event stream data in the data set into new event frame sequence data according to a set time resolution dt; s3, constructing an improved spiking neural network model: improving a synaptic connection mode of a leakage-accumulation-emission (LIF) spiking neuron in a time dimension, and constructing an improved spiking neural network based on an improved LIF neuron layer; s4, for a data set after pulse event flow serialization aggregation, randomly extracting samples from a sequence as input, and training and testing the constructed improved pulse neural network; and S5, storing the trained improved spiking neural network structure and network parameters. According to the invention, the network classification accuracy in target identification and classification of neuromorphic vision can be effectively improved.

Description

technical field [0001] The invention belongs to the field of deep learning in machine learning, and in particular relates to a neuromorphic visual object classification method based on an improved pulse neural network. Background technique [0002] In recent years, deep learning represented by Artificial Neural Networks (ANNs) has achieved great success in the fields of image recognition, natural language processing, etc. The operating mechanism of the brain is fundamentally different, and it is difficult to achieve truly strong artificial intelligence. Spiking Neural Networks (SNNs) use a more biologically interpretable spiking neuron model as the basic unit. Compared with artificial neural networks based on pulse frequency coded information, they have the advantages of low delay and low energy consumption. Simulating various neural signals and arbitrary continuous functions is an effective tool for complex spatiotemporal information processing. [0003] The neuromorphic ...

AI Technical Summary

Problems solved by technology

Generally, a deeper network structure is a necessary condition for improving network performance. However, the gradient problem caused by non-differentiable pulses in training makes it difficult for deep spiking neural networks to converge during training.

Images