FPGA Implementation Of ECG Identification Based On CNN

With the development of the times and the progress of society,people have higher and higher security requirements for personal identification technology.Compared with other identification technologies,biometrics is more secure and convenient,and is widely used in the field of identification.In the fields of financial security and military,there are some shortcomings of traditional biometric identification technology,such as face fingerprint and iris,which are easy to simulate DNA verification through camouflage and cost is high,so a new biometric identification technology with higher security is needed as an effective supplement.ECG(electrocardiogram)is common in the human body signals and current collection technology more mature,it is based on living signal acquisition,has a natural advantage in security risk,is the current hot research topic in the field of identification.ECG identification technology relative to face the fingerprint identification technology such as low recognition rate problems,in order to improve the accuracy of ECG identification,researchers often deep learning algorithm is adopted to improve the identification,the convolutional neural network algorithm to deal with the problem of object classification recognition has a unique advantage,convolution neural network algorithm computation intensive,however,of the poor real-time performance.To solve this problem,GPU(Graphics Processing Unit)or ASIC(Application Specific Integrated Circuit)and other parallel computing platforms are usually used to accelerate the algorithm.However,GPU has the disadvantage of large volume and high power consumption,and ASIC,as a special Integrated chip,has a high development cycle and cost.As a hardware platform with parallel computing capability,FPGA(Field Programmable Gate Array)has less energy consumption and more flexible configuration compared with GPU and ASIC.Xilinx's zynq-7000 series fully programmable on-chip system,a new generation of scalable processing platform,USES the latest architecture and integrates dual-core ARM Cortex A9 and FPGA on the chip.This chip can more easily and efficiently help developers achieve hardware development of system-level algorithms.In order to achieve the high real-time performance and the ECG identification system with offline recognition,this paper uses the PYNQ-Z2 embedded development board which built-in Zynq-7020 chips for ECG identification system development,the system adopts convolutional neural network identification,the convolutional neural network model training shall be carried out by the server,and the inference stage in the development board,does not depend on the server,and can work offline,so as the pressure of the server is reduced and the power consumption is lower and the realtime performance is higher.The main contents of this paper are as follows:1.Design of Intellectual Property core of denoising module and R crest detection module.In this paper,the median filter is selected to remove the power frequency noise of ECG signal,and the fast median filter algorithm is used to design the IP core of the median filter.According to the characteristics of R wave with high amplitude and sharp peak,an algorithm for real-time detection of R wave peak value point is designed,and the IP core design of R wave peak value point detection module is completed.2.Design of IP core of convolutional neural network.This paper makes full use of the advantages of parallel computation of FPGA.By analyzing the structure of convolution layer pooling layer and full connection layer,the IP core of convolutional neural network with parallel computing ability is designed to improve the computing speed of identity identification system.3.FPGA implementation of ECG identification system.According to the designed denoising module IP core,R crest detection module IP core and convolutional neural network IP core,the ECG identification hardware system was built on the PYNQ-Z2 development board,and the hardware system driver was developed to realize real-time ECG identification.