Tensorizing Neural Networks Based Face Recognition Algorithm Design And FPGA Verification

In recent years,with the development of science and technology such as big data and high-performance computing,the artificial intelligence has also made considerable progress.Face recognition has gained widespread attention from the society,as the most widely used in the current artificial intelligence.The accuracy of face recognition algorithms and the implementation speed of field programmable gate arrays are the key indicators affecting the performance of face recognition systems.Therefore,it is of great practical significance to realize an accurate and rapid face recognition system.A face recognition algorithm is designed in this thesis includes face image preprocessing and tensorizing neural networks.Face recognition preprocessing system is designed by used contrast limited adaptive histogram equalization algorithm,high frequency emphasis filter algorithm and data normalization.Noise of the AR database images is effectively reduced and the details of the image are enhanced,which reduces the influence of light unevenness.Tensorizing neural networks algorithm model is designed by used Matconvnet toolbox of Matlab.The network recognition rate is improved to 95%,through parameters analysis and adjustment.The last layer of parameters is reduced to 89%.At the hardware implementation level,an FPGA based face recognition system was designed using RTL code and C code.The system is mainly composed of four parts: ARM-side logic control module,convolution accelerator module,communication module and tensorizing calculation module.Among them,by reducing the fixed-point number conversion of floatingpoint numbers to reduce the memory consumption of the hardware,the computational efficiency is improved.Finally,using the Vivado suit to verify the advantages of FPGA-based tensorizing neural networks face recognition system.The algorithm design of this thesis is based on Matconvnet development environment.The recognition rate of designed face recognition system reaches 95%.FPGA verification used Xilinx's Zynq-7000 series chips,with a recognition rate of 94.58% and a system identification time of about 32 ms for a single image.Compared with the current face recognition system,this thesis has obvious advantages for the recognition rate and recognition speed of AR database.Fast and accurate image recognition in AR database is realized by the tensorizing neural networks algorithm and FPGA system designed in this paper.