ECG Identification Based On Integrated Learning Algorithm

With the rapid development of information and technology,the demand of people for information security is getting higher and higher,so traditional identification methods have been difficult to meet the current high demand.As a new type of identification technology,biometric recognition has become more and more popular.As a kind of biological feature,ECG has its unique advantages.It is the internal characteristics and is difficult to be stolen.Compared with other iological features,ECG is one-dimension signal,which is easy to process,and the acquisition technology has been quite mature.Many breakthroughs about feature researcg of ECG identification have been made at present,but there are still some problem.Such as low accuracy of identification,and poor timeliness are both difficulty in meeting the practicability,especially the change of individual heart rate will seriously affect ECG identification accuracy.In order to solve these problems,this paper studies feature processing,heart rate correction and classification model construction.1.The feature processing method of heart rate correction for S-T resampling is proposed.Heart rate of normal people will change because of many factors such as environment.The changes show different R-R intervals in ECG signals.Heart rate changes have different effects on each band of ECG signal,among which S-T band is the most affected.And the changes seriously affect ECG signal characteristics extraction accuracy.The original ECG signals are denoised by wavelet soft threshold method,then the peak point of R wave is detected by second-order difference threshold method.The T wave is determined based on the proportional relationship between each band and R-R interval.According to the proportional relationship,the heart beats are divided into several segments.The S-T bands are re-sampled to achieve heart rate correction,at last combining the segments into complete single heartbeat.2.ECG identity recognition system based on PCA-Adaboost algorithm is proposed in the paper.The feature of whole single heartbeat is extracted through heart rate correction,existing the timeliness problem influenced by high-dimension signals,so PCA is used for dimension reduction.PCA can remove the correlation between ECG signals,and transform high-dimension waveform features into low-dimensional,which reduces the redundant information of ECG signals.Adaboost algorithm adopts the re-weighting strategy to increase the weight of wrong classification samples and reduce the weight of right when training samples.The strong classifiers is constructed by reducing the weight of low accuracy weak classifiers,and improving the weight of high accuracy.The better classification effect of strong classifier is established through weighted combination.The ECG data in ECG-ID database are used to verify and analyze the proposed method.For ECG features after heart rate correction,the recognition accuracy is obviously improved.PCA dimension reduction can greatly improve the timeliness of system and the accuracy.Compared with support vector machine(SVM),BP neural network and random forest,the algorithm accuracy achieves 97.56%,and the effect is better than the other three algorithms.3.ECG identity recognition system based on XGBoost algorithm is proposed.In order to make ECG identity recognition more practical and require higher accuracy and lower time-consuming,XGBoost algorithm is used to construct classification model.The objective function of XGBoost algorithm introduces regularization term,and adds the second derivative term of Taylor series expansion,which can better fit the data.It can call computer's multi-CPU to perform multi-threaded data operations,and greatly reduce the time consumption for large data sets.The experiments of 41 individuals database,utilize XGBoost algorithm to construct ECG identification model,the accuracy of identification has achieved 100% accuracy and the timeliness has also been greatly improved.This algorithm tends to be more practical.The purpose of this paper is to provide a more practical ECG identification model,which is able to achieve better recognition accuracy and timeliness,and provide technical support for the practicality of ECG identification.