A Research Of ECG Signal Recognition System Based On Neural Network

Heart disease is one of the major diseases that threaten human health.Wearable smart ECG monitoring equipment can complete the identification of abnormal ECG sig-nals in real time,and remind patients to go to hospital for treatment through early warning,thereby reducing the loss of life and property of patients.For such systems,the key re-quirement is to achieve low power consumption of system while ensuring high accuracy.However,the existing wearable smart ECG monitoring device has suffered from the prob-lem of high power consumption in both abnormal ECG recognition and ECG signal trans-mission.In this work,we propose a low-power intelligent ECG signal recognition system that includes an artificial intelligence ECG recognition algorithm with an event-driven neural networks and an adaptive ECG signal compression algorithm based on intelligent recognition.Compared with the existing technical schemes,the scheme proposed in this thesis greatly reduces the power consumption of abnormal ECG recognition processing and the transmission power of ECG signals while maintaining high accuracy.First,the thesis introduces the significance of smart ECG recognition and the current status of research at home and abroad.It briefly summarizes the existing work in this field and analyzes their advantages and disadvantages.Then it introduces the background knowledge related to ECG recognition,including the basic concepts of ECG signals,the database of ECG signals,the basic principles of neural networks and their applications in intelligent ECG recognition.Subsequently,the thesis introduces the ECG signal pre-processing method,including the basic principles of the pre-processing method,the ECG signal denois algorithm and the R-peak detection algorithm.Among them,in denoise algorithm,several common noise filtering methods and effects are introduced.For the R-peak detection algorithm,the widely used Pan-Tompkins algorithm is introduced.Next,the paper proposes the event-driven neural network algorithm and the biased training method applicable to the algorithm in detail.The event-driven neural network effectively reduces the processing power consumption of intelligent ECG recognitionand the impact on accuracy by biased training.At the same time,an adaptive ECG compres-sion algorithm based on intelligent recognition is proposed,which effectively reduces the power consumption of ECG signal transmission.In order to meet the real-time requirements,based on the proposed algorithm,a ded-icated neural network processing hardware for ECG recognition was also designed and implemented by FPGA.Its function and real-time performance were verified,and the hardware was also evaluated.The resource overhead lays the foundation for the next-generation low-power intelligent ECG special processing chip design.To summarize our experimental results,the proposed smart ECG recognition algo-rithm achieves a recognition accuracy rate of 98.4 % which saves 75.5 % of the cal-culation amount comparing with existing neural network-based smart ECG recognition algorithms,and comparing with the existing ECG compression algorithm,the compres-sion ratio is increased by up to 4.69 times.At the same time,FPGA-based neural network processing hardware for ECG recognition is implemented,which satisfies the require-ments of function and real-time performance.In the end,the full thesis is summarized,the shortcomings and shortcomings of the proposed scheme are analyzed,and future work is prospected.