Research And Hardware Implementation Of Character Classifier Based On Convolutional Neural Network

With the gradual popularization of the concept of artificial intelligence,the development of artificial intelligence has attracted more and more attention.Even,artificial intelligence has been included in the national key development industries.At present,AI is still in the era of weak AI.In the future,it will have a chance to break through,which can better help human solve problems,change people's life style,and enter the era of intelligence.There will be intelligent robots,driverless technology and so on.AI will play a greater role in all fields of society.Deep learning is to use neural networks to express the learning process.The learning method of neural networks vividly simulates the human brain's way of thinking,showing better results,among which convolutional neural networks are the most typical.The main purpose of this study is to run the convolutional neural network on the hardware platform of FPGA.It not only explores the methods to improve the performance and delay optimization of the convolutional neural network,but also studies the development process of HLS.The application of character classification is taken as an example for practical research.This paper first describes the history of the development of neural networks,learns the forward and backward operations of convolutional neural networks and representative back propagation algorithms,then introduces the structure of general convolutional neural networks,and illustrates the typical convolutional neural networks Alex-Net and VGG-Net.The convolution neural network for character classification is designed.Then,the convolution neural network is trained and the model file is generated by setting up the network training environment.The structure of the network is analyzed,and an optimization scheme is proposed to improve the performance of the network.By comparing the simulation results,the best scheme is found.Then,briefly describe the design flow of HLS software,the source files are transformed by HLS synthesis and co-simulation,and the correctness of convolution neural network function is further verified.Finally,design the hardware implementation system of convolution neural network.FPGA has great advantages in running forward prediction of convolution neural network.It guarantees the system's requirement for timing,inserts ILA core to monitor the output signal and state signal,observe the real-time operation of convolution neural network in the field of FPGA,analyze the results and obtain a conclusion.In the end,the convolution neural network is successfully run in the FPGA.