Research On Heart Sound Classification And Recognition Based On BiLSTM And Design Of Auscultation Syste

The passing years has witnessed cardiovascular disease becoming one of the global killers with the most fatality,which poses a great risk to human health.Heart sound signal serves as the comprehensive response of the cardiovascular system,which includes the pathological and physiological information of heart.Given that,the collection and analysis of heart sound signal are very instrumental to the clinical diagnosis of cardiovascular disease.The thesis expounds the research and development of a heart sound auscultation system that is invented based on the characteristics of the heart sound signal,and as well brings forth a heart sound classification algorithm on the ground of adaptive noise reduction and deep learning.The main contents of work can be summarized as follows:1.Development of heart sound auscultation system: The development of heart sound auscultation system includes that of electronic stethoscope and auxiliary auscultation APP.Among others,the design of electronic stethoscope contains the structural design of auscultation head,the design of heart sound acquisition and conditioning circuit,the design of main control circuit,the design of Bluetooth circuit and the design of power supply circuit.The design of auxiliary auscultation APP contains that of Bluetooth communication transmission module,user information management module,heart sound playback module,heart sound waveform drawing module,heart sound playback module and file management module.2.Study and realization of noise reduction algorithms: According to the characteristics of heart sound signal and ambient noise,an adaptive noise reduction algorithm for heart sound signal is designed on the basis of wavelet threshold and improved genetic algorithms,which solves the problem that the noise type of heart sound signal is unknown and changeable,and it is difficult to reduce noise effectively.Through simulation,it is verified that the method can effectively reduce the noise of heart sound signals at different noise level,and the heart sounds collected in the experiment are tested.Referring to the subjective and objective evaluation indexes,it is proved that the noise reduction algorithm is practical and reliable.3.Research and realization of positive abnormal heart sound classification algorithm: A positive abnormal heart sound classification algorithm based on wavelet scattering network and BiLSTM neural network is proposed.It is found that the BiLSTM network has a high classification accuracy though the utilization of wavelet scattering transform instead of traditional wavelet transform to extract features from preprocessing(data enhancement and noise reduction)heart sound data,and using LSTM network and BiLSTM network that empirically set hyper parameter to train and test positive and abnormal heart sound feature.To further improve classification performance,Bayesian optimization of the super parametric combination of the WSTBiLSTM network model is used.A classification accuracy of 92.19% is obtained through the optimized classification model,demonstrating a better performance than that of other methods.