Image recognition method of deep pulse neural network based on dynamic threshold neurons

Abstract

The invention discloses an image recognition method of a deep pulse neural network based on dynamic threshold neurons, and belongs to the field of brain-like calculation and deep learning. The problems that when deep SNN is achieved through an existing DNN-to-SNN conversion technology, the pulse transmission rate is low, the issuing rate is low, and conversion precision is lost are solved. The implementation method comprises the following steps: firstly, training a DNN to obtain a weight and storing the weight; then, on the basis of a DNN-to-SNN conversion method, an activation function ReLU in the DNN is converted into an IF neuron with a dynamic threshold value in the SNN, and the weight of the DNN is normalized and mapped to the SNN; and finally, simulating and running the SNN, and calculating the threshold value of each neuron in each time step. The method is suitable for the artificial intelligence fields of image classification and recognition, target recognition and tracking and the like, the pulse transmission rate can be increased, the pulse distribution rate can be increased, information transmission can be accelerated, and meanwhile conversion loss can be reduced.

Description

technical field [0001] The invention relates to an image recognition method of a deep impulse neural network based on a dynamic threshold neuron model, and belongs to the fields of brain-like computing and deep learning. Background technique [0002] Deep Neural Networks (DNN) based on a highly simplified brain dynamics model, as a powerful computing tool, has achieved worldwide attention in many artificial intelligence fields such as image recognition, target recognition and tracking, speech recognition, and machine translation. 's results. However, problems such as ultra-high computing requirements and high power consumption greatly limit the application scope of DNN. [0003] Spiking Neural Networks (SNN) are fundamentally different from DNNs. SNN works in an event-driven manner. Neurons transmit information through discrete pulses instead of continuous values. It is highly bionic and can process dynamic data in spatiotemporal patterns. It has great application scenario...

AI Technical Summary

Problems solved by technology

[0005] When the existing DNN-to-SNN conversion technology implements deep SNN, the pulse transmission rate is low, the pulse firing rate is low, and the conversion accuracy is lost. The purpose of the present invention is to provide a deep pulse neural network based on dynamic threshold neurons. The image recognition method can speed up the pulse transmission rate, increase the pulse emission rate, and reduce the conversion loss

Images