Research On ECG Classification Algorithm And Recognition System Based On Deep Learning

As the continuous development of society,people's life and work pressure is increasing,leading to the increasing risk of cardiovascular disease,cardiovascular disease is one of the main diseases threatening human life.Electrocardiogram(ECG)can accurately record the physiological state of the heart system,which is the most important data support for cardiovascular disease screening because of its high noninvasive and reliable,and is commonly used in clinical diagnosis of cardiovascular disease.At present,the existing ECG recognition system is low in accuracy and effectiveness.Therefore,it is urgent to develop a ECG acquisition,classification and recognition system.The main work completed in this thesis is as follows:Firstly,an ECG classification model based on deep learning is proposed.The model uses residual join,batch standardization and focus loss function.Residual connection can suppress network model degradation and improve classification performance.Batch standardization can prevent gradient explosion or gradient disappearance due to network model being too complex.The focus loss function mainly solves the problem of data class imbalance.Then a combined heartbeat segmentation method is designed.Finally,the MIT-BIH rhythm database was used to verify the proposed method.Compared with traditional heartbeat segmentation method,the experimental results showed that the proposed classification model,the traditional segmentation methods and put forward the combination of heartbeat heartbeat segmentation method divided heart respectively96.23%,98.93% accuracy,that combination heartbeat segmentation method is used to segment the heart beating in the classification of the proposed model is more likely to be classified.Secondly,the image classification model VGG16 is first used to migrate as an ECG image classification model,and an ECG image classification model based on the combination of residual network and attention mechanism is proposed.Then the two models use the grayscale image of ECG and the logarithm-mel(Log-mel)spectrum of ECG,Mel-Frequency Ceptral Coefficients(MFCC),Log-mel spectrum and MFCC,respectively.Combine feature images as input.Finally,the MIT-BIH heart rhythm database is used to verify the proposed method.The experimental results show that,in the same classification model,the combined heartbeat segmentation method is superior to the traditional heartbeat segmentation method,and the combined spectral features are better than the single spectral features.Thirdly,an ECG signal recognition system is designed.Firstly,the STM32 main control module is used to drive the ADS1298 ECG acquisition module to collect,process,send and save ECG signals.Then Py Qt,thread,serial port technology is used to design the recognition system,which loads the weight of the deep learning model.Finally,the collected ECG data were used for verification.The experimental results show that the ECG signal recognition system can quickly and accurately identify the collected ECG...