Face Recognition System Based FPGA

Convolutional neural networks are widely used in the field of image recognition.This paper proposes an face recognition system based on FPGA,which uses convolutional neural network algorithm to realize face recognition.The system uses hardware design techniques including image cropping,convolution,activation functions,pooling,pipeline processing,and parallel processing of multiple convolution kernels.Using the characteristics of parallel computing of hardware circuits,the processing speed of the face detection system is accelerated.In the design of the hardware structure,the scalability of the neural network is fully considered,and the hyperparameters of the neural network can be changed to some extent.According to the neural network structure of different training processes,the necessary Verilog files can be created by the scripting language.The face detection architecture proposed in this paper was designed using Verilog HDL and implemented in the Altera DE2 FPGA development board.The convolutional neural network is divided into training process and recognition process.This study realizes the simulation verification of the convolutional neural network algorithm through the detection of the Public Figures Face Database face database by MATLAB,and discusses such as cross entropy cost function,normalization,parameters.Optimization methods such as initialization.Furthermore,an FPGA-based face detection system was built.The training process of face detection is completed by MATLAB training,and the recognition process is implemented on the FPGA platform.The system uses a neural network structure with an input layer,a convolution layer,a pooling layer and an output layer.The system has the characteristics of high speed and high recognition rate.The circuit structure design corresponding to convolution and pooling in the hardware platform of the FPGA can effectively reduce the delay and maintain a high throughput rate.Through MATLAB simulation of the algorithm,the recognition rate of the face database Public Figures Face Database reached 64.32%.The training process of the face data set is completed on MATLAB,and then the training weight values,offsets and other parameters are passed to the Verilog file through the scripting language to build the FPGA hardware platform.The hardware platform for constructing the corresponding convolutional neural network structure is equipped with a camera and an image acquisition module to realize the acquisition of the face in reality and preprocess it,and then the neural network can identify it.The simulation results show that the system can be reliable.Identify faces.