| The aging population in China has led to a significant increase in the proportion of people suffering from cardiovascular and cerebrovascular diseases,as well as an increase in tension and lack of medical resources.Therefore,there is a great need to develop wearable portable health monitoring and early warning devices for families.For cardiovascular diseases,there are cardiovascular diseases and cerebrovascular diseases.For cerebrovascular diseases,the resulting motor dysfunction seriously affects the ability of patients with cerebrovascular diseases to take care of themselves;in serious cases,they completely lose the ability to take care of themselves,and there is an urgent need for intelligent assistive devices to help the elderly and the disabled.To address the above two problems,this paper proposes to develop a portable ECG health monitoring device and study the ECG abnormality recognition algorithm;in addition,the key technology of motion imagination EEG control smart home based on motion imagination EEG signal classification is studied.For the above two aspects of research,this paper mainly does the following research work:First,for the problem of information redundancy or loss in the traditional fixed-length EEG sample selection method,the 3R EEG sample selection method is proposed.Then,the time domain features,subband spectral features and harmonic ratio features of the denoised MIT-BIH 3R ECG samples are extracted and combined to obtain seven feature combinations.The seven combined features are then combined with KNN,RF and SVM and LSTM classification algorithms,and feature selection is performed according to the recognition rate to obtain the ECG classification algorithms based on the selected features,which are TH-KNN,TS-RF,TS-SVM and TSH-LSTM algorithms,respectively.Second,ECG data were collected based on the self-developed wearable portable ECG Holter and made into ECG-H,and the four previously obtained ECG classification algorithms were applied to ECG-H by transfer learning.four ECG classification algorithms were first applied to the dataset obtained from manually labeled samples for model training,and the trained models were evaluated for classification ability and generalization ability.The trained models were evaluated for their classification and generalization abilities.The optimal ECG classification model,the TSH-LSTM model,was then selected based on the evaluation results and applied to the unlabeled 3R ECG samples of ECG-H to calculate the abnormal ECG rates.Finally,the health warning assessment model obtained from the weighting of abnormal ECG rate and heart rate variability coefficient is used to predict the human health status recorded by 16 sets of ECG data,and the comparison of the health warning results with the actual situation,the two situations are basically consistent,and the human health warning based on ECG signal is realized.Third,to address the problem that CSP features are vulnerable to interference rhythmic waves and the EEG signal will be group delayed when using filters,an improved algorithm WPT-CSP is proposed.then the two feature algorithms are applied on BCI IV 2a dataset and self-collected EEG data,and motion imagery classification experiments are performed using LSTM networks.The significant differences in recognition rates and kappa coefficients between the two algorithms on the two datasets indicate that the WPT-CSP algorithm is superior to the CSP algorithm.And the average overall recognition rates of the WPT-CSP-LTSM algorithm on the two EEG data sets were 91.99% and 88.16%,respectively,indicating the excellent motor imagery classification ability of the WPT-CSP-LSTM algorithm.Figure [48] Table [21] Reference [79]... |
PDF Full Text Request