Research On Ecg Signal Classification Algorithm Based On Deep Belief Network

Cardiovascular disease is the number one killer of human health,and its prevention and treatment is a major medical topic in the world.As the most commonly used diagnostic tool for medical staff,ECG fully reflects the physiological status of the heart and is widely used in the diagnosis of arrhythmia.ECG is a weak physiological signal,which is easy to be disturbed by noise.It is easy to be misdiagnosed by manual map recognition.Therefore,the automatic classification algorithm of ECG has important clinical value.However,the traditional classification algorithm needs to extract the signal features artificially,which can not make full use of the effective information of the signal,and will reduce the classification performance.Based on this,this paper uses the deep learning model to study the automatic classification of ECG,the main contents include:(1)Use adaptive filter to preprocess the ECG signal.According to the noise characteristics of ECG signal,LMS filter and RLS filter based on the least mean square criterion are designed to effectively remove the power frequency interference.On this basis,a low-pass filter is designed to remove the baseline drift noise in the signal.(2)ECG classification method based on Deep Belief Network(DBN).Deep Belief Network including three-layer RBM and one-layer Softmax classifier is constructed to complete the classification and recognition of ECG signals.The RBM of each layer and Softmax classifiers are pre trained.RBM acts as a feature extractor in the network.It uses its unsupervised automatic learning ability to learn the high-dimensional features of ECG data,and uses Softmax classifier to classify the features.The trained weights are used to initialize the DBN network,and the Error Back Propagation(BP)algorithm is used to fine tune the parameters of the whole network model to further improve the classification performance of the network.(3)ECG classification method based on Discriminative Deep Belief Network(DDBN).In DBN,RBM only acts as feature extractor and does not have classification function.Softmax classifier needs to be added to realize classification.DRBM can learn features and classify them at the same time.Compared with the traditional classifier,it is not sensitive to parameters and can improve the performance of network classification.The third layer RBM of DBN network is changed to DRBM to form DDBN.In order to improve the performance of network classification,Softmax regression layer is added at the end of the network,and BP algorithm is used to fine tune the network weight parameters.The final classification is achieved by initializing DDBN with the fine-tuning weights.In this paper,DBN add Softmax and DDBN were constructed to realize the classification and recognition of ECG signals.Experimental results has discussed the difference of network structure,the difference of classification accuracy before and after fine tuning and has compared results of DBN plus Softmax and DDBN classification.It shows that the ECG classification algorithm based on Deep Belief Network can fully mine the deep features of ECG signals and realize the automatic classification and recognition of ECG signals.And DRBM can be used as classifier to improve the performance of the whole network.