Automatic Recognition Method Of Electrocardiograph

Electrocardiograph(ECG)is a kind of time series which can reflect the health state of human heart.Therefore,the automatic recognition technology of ECG has a substantial economic effect and social benefit.Because of the weakness of ECG signal,noise is easily brought during the measurement of ECG.It increases the difficulty of abnormal ECG detection.This thesis employs several time series data mining and machine learning algorithms,and solves the feature representation,similarity measure and automatic classification of ECG to improve the accuracy and efficiency of ECG recognition.For the problem that ECG data contain much noise and the obscure of features,this thesis presents a time series feature representation algorithm which can provide a concise representation of time series based on its changing trend.This algorithm can not only reduce the high-dimension of time series but also extract the trend feature of data effectively.Compared with other time series representation model,experimental results show that the series trend feature representation algorithm has better feature extraction effect.Considering that the current time series measure algorithms are inefficient,and they can't meet the real-time requirement,a fast time series similarity measure algorithm is proposed.This algorithm matches time series according to the changing trend of data,and can compute the distance result in linear time which breaks the square level limitation of time complexity among other methods.Experiments show that the fast similarity measure algorithm not only has a great advantage in time efficiency but also has higher overall recognition accuracy than other methods.In view of the fact that the current automatic classification method of ECG can't effectively utilize the conventional features of ECG data,this thesis proposed a multi feature classification framework of ECG abnormal recognition.The framework integrates the advantages of similarity measure in pattern recognition and expertise of medical expert in ECG diagnosis.With the help of neural network,the framework integrates the time series similarity measure results and the conventional features of ECG to classify ECG abnormal signals.By comparing the classification results with the one nearest neighbor-dynamic time wrapping framework,it proved that the multi feature classification framework has better recognition performance.