ECG Identification Based On Nonlinear Manifold Learning

With the rapid development of science and technology,people's lifestyles have been widely and profoundly affected by informatization and digitization.Individuals and organizations have increasing requirements for information security.However,in recent years,personal information leakage has occurred frequently,and people are in urgent need of a safe and efficient way of identification to prevent data leakage.In this case,biometric identification technology arises at the historic moment.As a one-dimensional signal in the human body,ECG signals are not easy to be stolen,have high security,low storage space,and meet the four characteristics of biometric identification,namely universality,uniqueness,stability,and collection,etc.,which are paid more and more attention by scholars at home and abroad.Research on ECG identification has made a lot of progress,but there are still some shortcomings in feature redundancy,accuracy,and generalization ability of classifier.In view of this problem,this paper conducts the following research:1.In order to solve the problem of feature redundancy,the Uniform Manifold Approximation and Projection(UMAP)dimensionality reduction technique is firstly introduced into identity recognition.Firstly,the ECG signal is filtered by wavelet threshold method,and then a single beat is segmenced based on the R point,which is the legal bit of difference threshold.In view of the problem that the feature dimension of the initial heartbeat is relatively high and there are many redundant information interfering with the identity recognition,UMAP is adopted to reduce the feature dimension.UMAP has no calculation limitation on the embedded dimension.When processing large data sets,UMAP has obvious advantages,which can retain the features of the original data to the maximum extent and significantly reduce the feature dimension.After experimental verification,UMAP features can not only remove the redundant information in the initial heart beat features,but also strengthen the differences between different individuals,and enhance the correlation between different heart beats of the same individual.In terms of heartbeat recognition accuracy and identity recognition accuracy,both UMAP features are better than the traditional PCA feature dimension reduction method.Finally,extremely randomized trees(ET or Extra-Trees)algorithm is used for classification,and identification is completed based on statistical feature classification results of confusion matrix.Experimental results show that the proposed scheme can effectively improve the accuracy of identification.Compared with decision tree,KNN,random forest and Adaboost,the algorithm achieves 95.62% accuracy of subject identification among 160 individuals,which is better than the other four algorithms.2.Aiming at the problem of poor Generalization ability of the classifier,this paper further chooses Stacking algorithm(Stacked Generalization)as the classifier,which is a technique to combine the information of multiple learners to generate a new model.In this paper,the primary learner of Stacking is selected as the extremely randomized trees,and the secondary learner is Extreme Gradient Boosting(XGBoost).First,the extremely randomized trees is used to the original heartbeat training set for training,training method is adopted by five-fold cross-validation method.The output results of the training set and the average output results of the testing set are combined as secondary features,which are input to UMAP for dimensionality reduction,and then XGBoost is used for prediction and discrimination.Experimental results show that the scheme not only improves the accuracy of identification,but also has good generalization performance.The accuracy of the scheme is 96.88% in 160 individuals.The experimental results show that the UMAP dimensionality reduction algorithm can remove the redundant information in the initial cardiac beat very well,and the classification combined with extreme random tree can achieve a high accuracy.At the same time based on Stacking ECG identification method,not only in the identification accuracy has been improved,but also has a good generalization ability.