Research On Heart Sound Classification Method Based On Noise Detection Network

The heart is a very important organ of the human body,and its activity often reflects the physiological or pathological conditions of our body.Heart sound is an important clue to assess heart disease.It is a guide for advanced diagnostic tests and is often used for early diagnosis of cardiovascular disease.Auscultation of heart sounds is an important part of physical examination and many pathological heart diseases can be known.Active prevention of cardiovascular disease and reliable monitoring play an important role in inhibiting the spread of cardiovascular disease and are also important measures for the treatment of cardiovascular disease.Because manual auscultation has the characteristics of simple operation,low cost,significant auscultation effect and easy transfer of auscultation environment,stethoscope auscultation has become the most commonly used method for diagnosing heart disease.However,manual auscultation requires doctors with rich clinical experience and excellent professional knowledge.In addition,manual auscultation is often susceptible to clinical environmental noise.Therefore,it is very necessary to provide a method for automatically classifying heart sound signals,in particular,to classify mixed heart sound signals in a complex clinical environment.In order to simulate the complex clinical auscultation environment,different pollution intensity and pollution ratio are used to express the degree of environmental noise pollution of the heart sound signal.In this paper,the noise data set of the laboratory is used to randomly pollute the heart sound signal according to different pollution levels to obtain different pollution.The degree of mixed heart sound signals.Because environmental noise can cause great interference to heart sound classification,if the classification of the heart sound signal is used directly by the classification network,the classification effect will be unsatisfactory.Therefore,it is necessary to add a pre-detection network that can eliminate the environmental noise in the heart sound signal before classifying the network.This paper proposes a heart sound classification method for pre-detection network + classification network dual network structure.The main contents are as follows:1)The heart sound segment contaminated by environmental noise is removed by the pre-detection network to reduce the influence of environmental noise on the classification effect of the classification network.The noise recognition module uses a noise recognition algorithm based on a convolutional neural network and a bidirectional long-term memory network.2)Classification of heart sound signals.The model structure used for the classification of mixed heart sound signals is the pre-detection network + classification network dual network structure,wherein the classification network is based on convolutional neural network and single bidirectional long-term memory network classification model.The simulation experiment of mixed heart sound signal classification is carried out for the pre-detection network + classification network dual network structure and the classification network single network structure,and then the simulation experiment results are analyzed.The experiment proves that the pre-detection network + classification network dual network structure pair is used.The classification effect of mixed heart sound signals is better than that of classified networks.It shows that the pre-detection network + classification network model has better anti-interference ability,stronger robustness and more effective classification effect in complex clinical auscultation environment.