Method for learning and recognizing image pulse data space-time information based on Spike cube SNN

Abstract

The invention discloses a method for carrying out joint learning on space-time information of an image pulse sequence and an image recognition method. The method comprises the following steps of establishing a pulse sequence unit Spike cube and LIF neuron model on the basis of the pulse neural network, and learning synaptic weights and excitation thresholds, which are connected with each other, ofneurons of the pulse neural network by adopting an STDP mechanism; and using the trained model to carry out image classification and identification. The invention provides a new technical scheme forstructural design and learning of a pulse neural network and image pulse sequence learning and recognition, and also provides a new processing method for pulse data output by a DVS and other bionic vision cameras.

Description

technical field [0001] The invention belongs to the technical fields of brain-inspired computing, spike neural network (SNN), STDP learning, and image recognition, and relates to a learning method of spike sequences, in particular to a spike cube-based spike neural network for performing time and space information contained in image pulse sequences. A joint learning approach. Background technique [0002] In recent years, with the rapid development of deep learning, large-scale artificial neural networks have been widely used in many fields. The learning and recognition of sequence information can also be realized through deep learning technology. For example, time series processing methods represented by numerical recurrent neural networks can learn and process time-related sequence information, and are widely used in natural language processing, machine translation, speech recognition, etc., which only contain one-dimensional time information. In two-dimensional scenes w...

AI Technical Summary

Problems solved by technology

[0004] Although most of the existing types of spiking neural networks do not accumulate the spiking sequence on the input source, most of the synaptic weights are obtained through weight transplantation of the numerical neural network, and the weight of the numerical neural network It is obtained by learning the pulse images accumulated by the pulse sequence. Therefore, the spike neural network built by weight transplantation does not have the ability to discover the timing information of the pulse sequence in essence, but only relies on the temporal accumulation of the pulse sequence. The potential of pulse sequences in the expression of spatiotemporal information has not been fully realized

There are also a few types of spiking nerves that use the STDP mechanism to directly learn the non-accumulated spike sequences. Although the object of learning is the spike sequence itself, the results learned by this method at present, that is, the synaptic weights between neurons and The neuron firing threshold is still a static value

Using static synapses without time characteristics to process pulse sequences with time characteristics naturally fails to effectively utilize the timing information contained in the pulse sequence

Images