Research On ECG Signal Feature Classification Based On FCM Algorithm

With the development of social economy,people’s life style has changed,and the prevalence of cardiovascular diseases has become an obvious trend,which has gradually become the main disease burden of the population in China.Cardiac arrhythmia can be observed effectively by electrocardiogram,which is of great significance for the prevention and diagnosis of cardiovascular diseases.With the rapid development of artificial intelligence and big data,the research results on automatic classification of ECG signals by machine learning technology emerge one after another,but it has not been widely used in clinical diagnosis.Therefore,it is particularly important to study reliable and stable automatic analytical diagnostic methods.In this paper,the arrhythmia recognition technology at home and abroad is analyzed,aiming at the shortcomings of ECG signal processing,analysis and intelligent diagnosis algorithm,an improved ECG signal processing and recognition method is proposed,and verified by computer simulation experiment.The main work of this paper is as follows:(1)ECG signal preprocessing based on wavelet threshold.By analyzing the noise characteristics of ECG signals,a new threshold function between soft and hard thresholds is constructed by using wavelet transform to decompose the signals into multi-layer wavelet.Experiments show that this method can effectively improve the signal-to-noise ratio of the signal compared with the soft and hard threshold denoising.(2)Feature extraction of ECG signal based on fusion feature.Wavelet transform was used to detect the characteristic points of ECG signal,the ECG signal of a single period was taken as data sample,and the interphase and amplitude were extracted as time domain features.The Deep Sparse Auto-encoders is used to encode and decode ECG signals,and the network pretraining and fine-tuning can be completed with only a few labels,which reduces the dependence on sample labels.In the process,adaptive moment estimation is used for parameter optimization,and the implicit output of the highest level is extracted as the depth feature.Final,apply weighted fusion to fuse time domain features and depth features,and the fusion feature was proved to be effective in characterizing ECG signals by experiments.(3)Classification of ECG characteristics based on Fuzzy C-means algorithm.To solve the problem that FCM algorithm requires prior knowledge,the density peak algorithm based on k-nearest neighbor optimization is used to obtain the number of clustering prototypes adaptively.In addition,considering the extreme imbalance of ECG signal data set,the error rate of normal cardiac edge data was reduced by adding sample size information into the objective function of FCM algorithm.Experimental results show that the method has high accuracy in identifying arrhythmia samples,and has achieved 98.95% accuracy in identifying five kinds of arrhythmias.