ECG Signal Recognition Based On Deep Learning

ECG signal recognition is the key to the intelligent diagnosis of arrhythmia.Electrocardiogram is widely used in the diagnosis of arrhythmia.It is a medical monitoring technology that reflects the physiological activity of the heart.With the development of 24-hour ambulatory electrocardiogram monitoring,the number of cardiac beats needed to be diagnosed for the same patient is too large to be handled by doctors alone.In order to solve the above problems,this paper makes use of the big data features of dynamic monitoring ECG signals and combines with depth learning network to conduct in-depth research and analysis on the feature extraction and recognition of ECG signals.The main contents include:The paper analyzes the generation mechanism and waveform characteristics of ECG signals in detail,studies the ECG signal feature extraction methods,systematically expounds the current research status of feature extraction and classification in ECG signal recognition at home and abroad,and discusses the deep learning and recognition Related algorithms are analyzed in detail,including single hidden layer feedforward neural network,back propagation algorithm,autoencoder,sparse autoencoder,and extreme learning machine.According to the characteristics of the three main noise baseline drifts,electromyographic interference,and power frequency interference in the ECG signal,and the differences between the main frequency band and the interference frequency band of the ECG signal,a zero-phase shift digital filter,Butterworth filter,and FIR filter removes noise from ECG signals.Construct an adaptive double-threshold function to locate the peak value of the denoised ECG signal.Experiments show that the method can accurately locate the R peak and avoid missed detection and false detection.Finally,based on the detected R waves,take points forward and backward,respectively,to capture the heartbeat.The core of constructing a deep stacked network is to use the layer-by-layer learning of stacking multiple sparse autoencoders to effectively complete the extraction of ECG signal features from low-dimensional to high-dimensional,and combine with the Softmax classifier to build a deep stacked network to complete the ECG.Automatic identification of signals.The structure and important parameters of the deep stacked network are analyzed.The experimental results show that compared with the multilayer perceptron,stacked autoencoder and principal component analysis algorithm,the proposed method can more effectively extract the deep features in the ECG signal.The total accuracy rate is as high as 99.69%,with higher accuracy,which verifies the reliability of feature extraction.Build a deep converged network.The stacked sparse autoencoder fuses the extreme learning machine to complete the construction of the deep fusion network.After the stack sparse autoencoder completes the feature extraction,the obtained feature matrix is sent to the extreme learning machine.Since the hidden layer parameters of the network are randomly initialized,no reverse optimization strategy is needed,saving algorithm running time.And experimentally analyze the number of nodes in the hidden layer of the extreme learning machine to obtain the optimal ratio parameter.Finally,experimental comparison with multilayer perceptron,stack autoencoder,BP neural network and deep stacked network proves that the deep converged network built in this paper has high recognition accuracy,fast convergence speed,The practical application has higher reference value.