FPGA-based Convolutional Neural Network Application

Convolution neural network is a kind of feedforward neural network with depth structure including convolution computation.It is one of the representative algorithms of deep learning.At present,convolution neural network is mostly implemented in Python,Matlab,C language and so on,which has the disadvantages of slow speed and poor real-time performance.As a kind of programmable logic device,(Field Programmable Gate Array,FPGA(Field Programmable Gate Array)has the advantages of flexibility,convenience,high speed and small size.Matching with the characteristics of parallel operation of neural network,it can shorten the time of network training,realize real-time processing,and can be used in small embedded systems.This paper studies the method of using FPGA to realize convolution neural network,and implements a convolution neural network with five layers in Verilog language,which fully excavates the advantages of hardware implementation under the condition of high recognition rate.By studying the working principle and structure of the network,a complete circuit model is constructed,and the design scheme of FPGA is given.The correctness and integrity of the network are verified by handwritten digit recognition.In this design,60000 handwritten digital sample images in MNIST database are trained by back propagation,the weights and biases with the highest accuracy are extracted,and then the forward propagation of the network is used to complete digital recognition.The design process is realized by Quartus II and ModelSim simulation tools.The simulation results show that all the samples take 50 Ms to train under the 100 MHz clock.Compared with the software implementation,the speed is obviously improved,the real-time performance of hardware design is satisfied,and the accuracy is higher.This research provides a method and strategy for image recognition of embedded devices,and has practical application value.