Implementation And Application Of Neural Network Based On FPGA

Artificial neural network is an important part of artificial intelligence. With the advent of the era of artificial intelligence,more and more enterprises and research institutes have been put into the research and development of intelligent products and tools,to help businesses and individuals to provide more convenient and efficient intelligent tools.Neural network has been widely used for its advantages of high speed and high accuracy.For example,in the pattern recognition,information processing,robots and other aspects of the application to achieve continuous breakthroughs.At present, most of the neural networks adopt the traditional general-purpose computer, while the traditional general-purpose computer uses the serial computing method.The application of this approach to neural networks has been limited, therefore, a better way to achieve neural networks is particularly important.Because of the characteristics of FPGA’s hardware and the highly parallel computing, it can effectively make up the limitation of serial neural network, which can provide a new way for the realization of artificial neural network.The method of neural network is realized by FPGA hardware, which has great commercial value and industrial, significance in solving the fast automatic recognition of electronic shelf label and the character image of industrial instrument.In this paper, the different methods of software and hardware simulation to achieve the recognition of character images.From the theory of neural network and software and hardware realization of the two aspects of the paper, and the MATLAB software implementation, Libero hardware simulation to achieve the key issues on the analysis.The software implementation method is based on the neural network correlation function in the MATLAB function library.Hardware simulation method is used in the Libero integrated development environment, through the hardware description language to achieve the various functions of the neural network modules.Selection of the FPGA structure for the Fusion Actel series devices, it is the only analog function of the Flash architecture of FPGA devices, can meet the functional requirements of the artificial neural network.The main research contents of this paper include:First, in the process of character recognition,using MATLAB to select training samples and determine the structure of the neural network analysis and repeated experiments (neural network hidden layer node number).And through the noise free input samples to compare the recognition accuracy, further optimization of the network.Then, in the FPGA hardware implementation of neural network, the key part of the neural network to achieve the key part of the design and implementation:including the composition of a single neuron module, the data bits of the representation, the choice and implementation of the incentive function.The implementation process of the hidden layer is described in detail, and the calculation process of each stage of the network is analyzed and simulated.At last, through the simulation of each module and the verification of each stage, the whole hardware network system is simulated to identify the experiment, verify the results of network identification and character recognition.From the analysis of the results of the experiment, the construction of the system can achieve character recognition, with high stability and indeed, to achieve the purpose of the experiment.