Study On Time-Frequency Feature Fusion And Classification Method For Heart Sound Signals

The heart sound signal is an important physiological signal for hospital clinical equipment monitoring,contains information such as atrial and ventricular,and is often used in early diagnosis of cardiovascular disease.Heart auscultation is one of the most commonly used cardiovascular diagnostic methods.Doctors diagnose heart disease by listening to whether there is noise or distortion in the heart sound,such as arrhythmia,heart valve disease.Due to the influence of related factors such as experience and environment,different doctors have not been identical to the judgment on the heart sound signals.In recent years,with the rapid update of computer technology,computeraided diagnostic technology has become a research hotspot in the field of medical sector.Therefore,the automatic classification technology of heart sound signals has important research significance and application value for the auxiliary diagnosis of cardiovascular disease.Based on the research of heart sound signals characteristics and depth learning theory,this paper proposes an automatic heart sound classification method based on time-frequency characteristic fusion.Through two methods of serial fusion and parallel fusion of heart sound time-frequency domain features,this paper constructs deep heart sound classification models suitable for high-performance servers and lightweight heart sound classification models suitable for small instruments.Firstly,by analyzing the characteristics of the heart sound signal,this article standardizes the heart sound signal data,processes low-frequency and high-frequency noise and segmented it by equal-time-length.The processed heart sound signal is suitable for the subsequent feature extraction and classification of the heart sound signal.Secondly,this paper designed two time-frequency feature extraction and fusion architectures.Methods such as long and short-term memory structure and residual structure are used to construct two deep models based on time-frequency feature fusion.Then,according to the time-frequency feature fusion method,this paper studies deep separable convolution,fusion grouped convolution,gated recurrent unit and attention mechanism methods,constructs two lightweight models based on time-frequency feature fusion for heart sound signals,and deploys the lightweight model in the mobile device.Finally,through the index and parameter amount of the model,the heart sound data set is tested with the heart sound classification model.The results show that the accuracy rate of the deep model can reach more than 97%,with a large amount of parameters.Compared with the deep model,the lightweight model has fewer parameters,and the accuracy rate can reach more than 95%.In summary,the heart sound classification method based on time-frequency feature fusion proposed in this paper can achieve better heart sound classification results,and has important research significance for the auxiliary diagnosis of cardiovascular diseases.