ECG And Pulse Feature Recognition And Application Based On Deep Learning

Electrocardiogram(ECG)and pulse signals contain a variety of physiological and pathological information of the human body.The important roles of ECG and pulse signals are reflected in the detection of cardiac function and cardiovascular health status.Diabetes is a chronic metabolic disease characterized by hyperglycemia.Diabetes can cause cardiovascular disorders and a variety of complications.ECG and pulse signals are used as effective indicators to evaluate the physiological status of patients with diabetes.In order to study the effect of diabetes on cardiovascular function,ECG and pulse signals are synchronously collected from healthy people and diabetes.In this thesis,the characteristics of two physiological signals are fused to construct a prediction classification model for the diabetic population.In order to explore the differences of physiological signals in different populations,the models are compared and analyzed.The specific research content is as follows:Firstly,the physiological signals acquired synchronously are denoised.This thesis introduces the effect of ensemble imperial mode decomposition in signal denoising,and analyzes the shortcomings of this method.On this basis,an improved EEMD denoising algorithm based on Null Space Pursuit is proposed.This method uses EEMD algorithm to remove the baseline drift of the signal,and NSP removes the high-frequency burr of the signal to obtain high-quality signals.Results analysis and index comparison show that the improved EEMD has better denoising effect.Secondly,a feature classification model for two physiological signals is constructed.The Convolutional Neural Networks(CNN)is used to construct the classification model,and the CNN-BiLSTM classification model based on model fusion is proposed in combination with the pre-and-post relationships of signal extraction using bi-directional Long Short-Term Memory(BiLSTM).A feature classification model suitable for ECG and pulse signals is designed in order to lay a good foundation for the diabetes classification model based on multimodal fusion.Finally,a multi-modal fusion diabetes classification model based on attention mechanism is proposed.Based on the single modal classification model of ECG and pulse,the high and low layer features of ECG and pulse are fused and classified.Aiming at the problem of too many parameters in the process of feature fusion,combined with attention mechanism module,the accuracy and robustness of multimodal classification model are improved.In this thesis,CNN-BiLSTM fusion classification model is used to explore the influence of diabetes on ECG and pulse signals.In the classification results,the accuracy of classification model based on ECG and pulse reached 94.59% and 93.82%,respectively.The results show that ECG and pulse signals can distinguish physiological differences among different populations.Based on the above research results,we further use the information fusion of multiple physiological signals to predict classification,and the accuracy rate reaches 96.46%.At the same time,after introducing attention mechanism,it is higher than the single signal classification model,and the accuracy rate reaches 98.07%.