Research On ECG Identification Based On Deep Learning

With the development of technology,people are paying more and more attention to their privacy protection.The emergence of biometric identification technology effectively protects people's privacy and security,and the uniqueness of individuals makes biometric identification as a more secure means of identification that can be used to establish secure and unique identity information in the crowd.At the same time,the development of biometric forgery technologies,such as fingerprint and facial data forgery,will perhaps become the biggest threat to people's identity security information.In contrast,electrocardiogram(ECG)identification is a good complement to existing biometric identification methods.ECG signal waveforms are the result of multiple sympathetic nerve interactions in the human body,and it is very difficult to control or try to imitate someone else's ECG signal,whose one-dimensional signal characteristics are easier to process and compute compared to other biometric features.In the classification,through the depth learning layer by layer of feature processing,makes the extracted features more representational.Furthermore,In order to further improve the accuracy of the identification system,reduce the dimension and redundancy in the feature processing,improve the stability of the classifier,and improve the performance of the identification system in the non-reference point feature extraction,the following studies were conducted.1.We built a PCA-gcForest-based ECG identity model.The Deep Forest is also called multi-grained cascade forest,with a layer-by-layer machining,internal characteristic transform,and adequate model complexity three deep learning models,enhance identity system characterization.At the same time,it is possible to improve the stability and accuracy of the system.In this paper,an ECG identity recognition model based on PCA-gc Forest is constructed to reduce feature redundancy and improve the accuracy of identity recognition.Firstly,most of the noise on ECG signal is removed by the wavelet denoising with weight threshold contraction,and then the R point is used as the reference point to segment the single beat signal.Primary component Analysis(PCA)was used to generate linear independent and low redundant low-dimensional features.Finally,gc Forest was used for feature processing and classification recognition.The experimental results show that thegc Forest algorithm can significantly improve the accuracy of the system.Compared with KNN(K-Nearest Neighbor),BP(Back Propagation)neural network and random forest,The accuracy of the proposed method in ECG-ID database and MIT-BIH-AHA database is97.36% and 99.14%,respectively,which are better than the above traditional classifiers.Therefore,the scheme effectively solves the redundancy of ECG signals,improves the accuracy and practicability of classification,has a higher classification and recognition results in multiple databases,and has a better robustness.2.We construct an ECG identification model based on non-base combined features.ECG is a weak electric signal in the human body,and the heart rate is easily influenced by various environmental factors such as disease,exercise,mood,causing the reference point positioning inaccurate,thereby affecting the identification accuracy.Therefore,this paper has been designed from the ECG identification model of the non-basis feature,no reference point positioning,and the non-reference point feature extraction leads to problems that the ECG identification system accuracy is not high,and combined with the characteristics of the ECG signal,one ECG identification method based on combined window features.Firstly,the original ECG signal is divided into window signal segments by non-reference point segmentation,and these window signal segments are used as the basic recognition units in ECG identification system instead of heart beat.Then,a Convolutional Neural network(CNN)is used to extract features from the window signal segment.As a popular deep learning model,the Convolutional Neural network can effectively extract features with identity differences from the window signal segment.Then,the time-domain features representing the identity information in the selected window signal segments are combined to enrich the identification criteria of window signal segments.Finally,gc Forest was used to complete the classification.By comparing several groups of experiments on MIT-BIH-AHA database and MIT-BIH-ST data,it is proved that the proposed algorithm can improve the performance of the system identified by ECG identification under the conditions of non-reference point feature extraction.