Research On ECG Identity Recognition Based On Non-linear Feature

With the advancement of high-end information technology in modern society,personal identification technology has played a crucial part in multiple regions,for instance,finance,security,confidentiality,and etc.The traditional identification technology has gradually exposed various defects.Therefore,due to its safety and convenience,the identification technology established on human physiological features has grown into a latest popular technology.The identification technology based on electrocardiogram signal(ECG)is getting more and more attention and has broad application prospects.After years of research,ECG signal has become a reliable data in the medical field.For the field of identification,ECG signal has mature acquisition technology and simple one-dimensional data processing compared with other physiological signals.At the same time,the data is extremely difficult to forge due to the fact that its internal characteristics are not exposed.At present,many progresses have been made in the research of ECG-based identification technology,but the existing identification technology still has many problems.At present,a large number of researches are carried out on small databases with only a few dozen or one or two hundred people.Therefore,although the recognition accuracy is close to 100%,there is a big gap between the scale of data and the actual application.When the same algorithm is applied The accuracy and efficiency of recognition will be significantly reduced when it comes to a larger data set.The traditional method has a large amount of calculation and the problem of model generalization is exposed.It is necessary to improve the timeliness and model generalization of identification technology.In response to these problems,this paper conducts the following research from the aspects of heartbeat segmentation,feature processing and classification recognition:1.Propose an ECG identification method based on t-distributed random neighbor embedding(t-SNE)combined with random forest classifier(Random Forest,RF).First locate the R point of the input signal,and perform heartbeat segmentation and normalization.Then use the t-SNE algorithm combined with the PCA feature vector for feature extraction,and finally use the random forest algorithm to construct and train the classifier according to the reduced dimensionality.Among them,the t-SNE algorithm can greatly reduce the dimensionality of ECG signal data processing,accelerate the training speed of the classifier,and effectively use the information and category labels of the ECG signal data to enhance the accuracy of ECG identification.RF classifier is hard to overfitting,so the application of RF classifier can enhance the model's generalization ability.On the purpose of figuring out the performance of this model,some experiments were carried out on Physio Net/CinC 2017 database.The outcomes demonstrate that the ECG identification method proposed by this paper has distinguished recognition ability and worked fast.When the sample number reaches1000,the recognition accuracy retains over 91%.At this time,the accuracy of traditional methods based on PCA+SVM,PCA+adaboost and other solutions has been reduced to below 85%.2.An ECG identification scheme based on TriMAP dimensionality reduction algorithm is proposed.Aiming at the shortcomings of the t-SNE algorithm that iterative time is too long and the timeliness is not strong,a better manifold learning algorithm TriMAP is used for dimensionality reduction.The TriMAP algorithm effectively overcomes the time-consuming defect of the t-SNE algorithm in terms of timeliness.Experiments show that when the sample size is 1000,the cost of this scheme is reduced by more than 95% compared with the t-SNE algorithm,while still maintaining 90% accuracy.In summary,the ECG identity recognition method proposed in this paper has excellent recognition accuracy and timeliness performance,and has better generalization performance than traditional methods on large data sets.It is a practical exploration of ECG-based identification technology.