Research On Multidimensional Health Monitoring And Intelligent Diagnosis For Wearable Devices

Chronic diseases,represented by cardiovascular and respiratory diseases,have become one of the most common causes of human death worldwide.Chronic diseases usually have a variety of complications,and have a strong temporal correlation with physiological characteristics such as respiratory rate and heart rate.With the improvement of health awareness,people pay more and more attention to the monitoring of chronic diseases and its complications,so multidimensional characteristics extraction and abnormal analysis have become a very important research task.With the development of smart terminal devices,wearable devices are becoming an emerging medical monitoring method that can universally cover people's daily lives,and can be widely used in monitoring tasks of various human characteristics.However,due to the complex application scenarios of wearable devices,the photoplethysmography(PPG)signals are extremely susceptible to noise interference,which will reduce the accuracy of characteristic extraction.Therefore,monitoring and analysis of multidimensional characteristics based on PPG signal denoising are becoming the focus and difficulty of current research.In order to achieve more accurate monitoring of human health and intelligent diagnosis of abnormal conditions for wearable devices,this thesis first combines the advantages of deep learning in feature extraction to design a deep denoising model based on convolutional neural networks.Then proposed a multidimensional characteristics extraction and intelligent diagnosisscheme,combined with machine learning algorithms to correct the feature extraction results and evaluate human health level.Because PPG signals have periodic characteristics,convolutional neural networks are selected for signal denoising and reconstruction.In order to improve the learning ability of the model,a tuning algorithm based on prior regularization is also designed in this thesis.By adjusting the network parameters during model training,the complexity of the model can be significantly reduced and the denoising performance is improved.After the signal denoising task is completed,the spectral peak tracking algorithm based on spectral analysis is used to extract the respiratory rate and heart rate.Combined with the integration strategy based on regression model fusion to modify the characteristics extraction process,which improves the accuracy and robustness of the model in complex environments.Finally,based on the characteristics extraction and waveform recognition of PPG signals,this thesis combines machine learning algorithms to design intelligent diagnosis schemes,which further extended the health level abnormality detection method.The main work and contributions of this thesis are as follows:· A deep learning denoising model based on convolutional neural network is proposed.In the case that traditional denoising methods cannot effectively filter the noise with same frequency,the PPG signal denoising and reconstruction are performed.Experimental results show that the model can complete PPG signal denoising task under complex conditions through effective learning of waveform features.· A tuning algorithm based on prior regularization is proposed,which greatly improves the learning ability of the model.And through the unlabeled pre-training method,the initialization values of the hyperparameters are effectively selected,so that the model can quickly select the optimal configuration and further reduce the complexity.· A multidimensional characteristics extraction algorithm based onspectral peak tracking is proposed.It combines the integrated strategy based on regression model fusion to optimize the heart rate and respiration rate extraction algorithms,experimental results show that the characteristics extraction algorithm has high accuracy and robustness in complex environments.· This thesis also proposes an intelligent diagnosis scheme combined with machine learning algorithms.Aiming at the case where a small amount of data has labels in the data set,feature selection and intelligent diagnosis schemes are designed.The experimental results show the effectiveness of the method in abnormal sample detection task.In view of the complexity of wearable device application scenarios and requirements of long-term multidimensional health level monitoring,this thesis proposes a deep denoising model based on convolutional neural networks.When there is complex noise interference,the model also shows superior denoising performance.Combined with the proposed multidimensional characteristics extraction and intelligent diagnosis analysis methods,the robustness of multidimensional health level monitoring is improved,and it has contributed to the popularity of wearable device application scenarios.The deep learning-based signal denoising method also provide new idea for removing co-frequency interference in the signal.