Research On Intelligent Health Monitoring Based On Deep Learning And Human Physiological Dat

As China’s aging continues to intensify,the increasing number of elderly people makes more and more elderly people prefer to age at home,which can alleviate the occupation of medical resources to a certain extent and is an inevitable trend of future aging,and also makes the elderly themselves have a greater demand for chronic disease monitoring.This paper designs a prototype intelligent health monitoring system to analyze and warn the health status of the elderly from the perspective of their own health monitoring needs.In this paper,the following research is carried out from two perspectives: the physiological state and the mental state of the elderly.(1)Design a multi-headed self-attentive arrhythmia classification algorithm.Firstly,the db6 wavelet transform is used to pre-process the ECG signal,focusing on improving the data quality and reducing the noise of the ECG signal.Second,a linear projection layer for ECG signal semantic feature extraction is designed using the matching relationship between the ECG markers and the segmented ECG signal;Third,a spatio-temporal characterization method of ECG signal sequences based on location coding is designed to integrate the time series information into the matrix operation.Fourth,a multi-headed self-attentive mechanism capable of capturing global contextual information is proposed to extract relationships and semantic features between ECG fragments to achieve semantic association and information stitching of non-adjacent ECG signals.(2)A multi-channel convolutional attention mechanism-based mental state recognition algorithm is designed.Firstly,data preprocessing is performed on multiple physiological signals,and the sensor data with different sampling frequencies are resampled to ensure data length consistency.Then the convolution module is designed based on the structural characteristics of the input signals and the length of the signals,and four one-dimensional convolution kernels of different sizes are used to simultaneously extract features from the signals to enhance the feature extraction capability of the model,and the convolution results are stitched together,and finally the maximum pooling operation is performed on the stitched results to increase the perceptual field of the model.The long-distance feature expression between signals is achieved while extracting local feature signals.(3)A prototype system of intelligent health monitoring is designed.In order to continuously monitor the health of elderly people suffering from chronic diseases at home and to provide health warning for their collected data,a prototype system of intelligent health detection based on human physiological signals is designed.Firstly,we select the appropriate sensors for the measurement of human physiological signals according to the health condition of the elderly and their physical characteristics.Secondly,the database design was designed according to the signal transmission method and its data characteristics.Finally,an intelligent health data management system is developed to visualize the collected data and analyze the ECG signal and mental stress of the measured person,and to provide early warning for the health condition.