Image recognition method based on FPGA customized pulse neural network

Abstract

The invention discloses an image recognition method based on an FPGA customized pulse neural network and performs image recognition by customizing a convolutional pulse neural network on an FPGA platform; the convolutional pulse neural network includes a convolutional layer, a downsampling layer, a full joint layer and a classification layer; the image recognition method includes processes of: pulse sequence generation, convolution operation, downsampling, full joint, and classification and recognition; the specific adopted development platform is Xilinx FPGA development board Virtex-7, the adopted development software is Vivado, and the programming language is Verilog. The method can recognize the pulse sequence information that the numerical neural network cannot recognize, and has the advantages of a faster recognition speed, a higher accuracy rate and lower power consumption in the high speed scene.

Description

technical field [0001] The invention belongs to the technical fields of pulse neural network, brain-inspired computing and FPGA, relates to the pulse realization of neural network, and in particular relates to an image recognition method based on FPGA customized pulse neural network. Background technique [0002] In recent years, artificial neural networks have developed rapidly, especially the research and application of neural networks represented by deep learning technology, which has set off an upsurge in China. This type of neural network is mostly based on numerical neural network. With the continuous increase of network scale, the power of computing platforms such as GPU and CPU is also increasing rapidly, and power consumption has become a problem that cannot be ignored. In addition, with the increase of network complexity, the numerical neural network greatly reduces the information processing speed, so the effect is not good in high-speed real-time scenarios. FPGA...

AI Technical Summary

Problems solved by technology

[0004] However, most of the existing FPGA-based neural network customization technologies are the realization of numerical neural networks, and it is still impossible to realize the customization of FPGA-based convolutional spiking neural networks, and there is no image recognition application based on FPGA-based customized convolutional spiking neural networks. , which is an unstudied content in the field of "brain-like computing"

Images