A Study About Basing Machine Learning Method To Select Heart Beats From BCG Signal

In modern society,sleep disorders and heart diseases are plaguing many people.The important index to diagnose these two diseases is heart rate.The doctor can diagnose the patient's sleep status and quality through heart rate variability analysis,and can also monitor the occurrence of heart disease through heart rate testing.With the convenience and family of medical equipment development,the non-contact measurement of heart rate has received extensive attention and research because of its advantages of non-invasive,convenient and long-term monitoring.Ballistocardiograph(BCG)is an important method.Its principle is that when the heart beats,it will cause the pressure of the supporting object that is in contact with the human body to change.The pressure acquisition module converts the pressure signal into an electric signal for record keeping,is the ballistocardiograph.In the previous algorithms of BCG signal extraction of heart rate,people mainly use traditional signal processing methods,including morphological and unsupervised learning(clustering,etc.)to help calculate heart rate.However,because of the errors introduced by non-contact measurement,the poor robustness of traditional methods and individual difference,such methods can not guarantee the robustness of BCG signals.In order to improve this problem,machine learning method is introduced in this paper.Specifically,in this article:(1)the second chapter introduces the principle of the BCG signal acquisition system and the physiological significance of the BCG signal,and introduces the correspondence between the BCG signal and the ECG signal,so as to ensure the physiological feasibility of the supervised learning algorithm.Finally,the collection scheme and index of the system are introduced,which lays a foundation for signal processing.(2)the third chapter of this paper proposes an unsupervised learning algorithm to collect the heart rate.Here,the K-means clustering algorithm is used to deal with the signal,and two indexes are introduced to judge the existence of the periodic signal,thus enhancing the accuracy of the algorithm.(3)in the fourth chapter,we introduce the supervised learning method to calculate heart rate.By comparing other methods,it is found that the use of supervised learning can maintain a high quasi rate while increasing the robustness.The specific idea is to make certain filtering de-noising of the extracted BCG signal in the supervised learning,and then divide the BCG signal according to the RR interval of the ECG signal,and divide the sequence into two categories including heart rate interval and no heart rate interval,thus turning the problem into a problem.Turn it into a classification problem.Then the feature extraction was carried out in the equal length window,and the random forest and support vector machine were used to classify the windows respectively.After the post-processing,the real time heart rate was calculated.Then the heart rate calculated by the standard ECG signal was compared,and the comparative random forest model was the best in the existing data.