Research On BCG Signal Classification Based On 1-D Convolutional Neural Network

Due to the aging of the population and the accelerating process of urban construction in China,the impact of the environment and lifestyle of citizens is also deepening,leading to the continuous increase in the number of cardiovascular diseases.In clinic,the medical instruments and methods used for cardiac function test can not meet the requirements of non-contact and non-invasive collection.Although these methods have good results,they will cause some harm to the body in the process of examination.Therefore,it is very important to develop a non-invasive and convenient home heart monitoring system.BCG is a small vibration of the body caused by the impact of blood vessels when the heart contracts and relaxes,reflecting the state of the cardiovascular system.The study of BCG signal classification can effectively prevent cardiovascular disease,and is conducive to the rational allocation of social medical resources.At present,there is a common method for BCG signal classification,feature extraction and training classification.The results show that the performance of BCG signal classification largely depends on the characterization ability of features,and the emergence of convolutional neural network(CNN)can effectively solve the above problems.Therefore,this paper proposes a BCG signal classification method based on 1-D CNN model.The main contents of this paper are as follows:(1)The spectral characteristics of BCG signal slice were analyzed and extracted.By using bispectrum nonparametric direct estimation method,the differences of BCG signals in different states are compared and analyzed.In order to reduce the amount of bispectrum calculation and fit the network input of 1-D CNN in this paper,the slice spectrum completely retains the information of signal amplitude,frequency and phase.A slice spectrum method is proposed to extract the amplitude and phase characteristics of BCG signal and generate one-dimensional vector.(2)Design a 1-D CNN1 multi classification algorithm based on a BCG signal.Aiming at the problem that the classification effect of traditional machine learning algorithm depends on the quality of artificial feature extraction,1-D CNN1 model is used to study the classification effect of BCG signals with different input data length.The results showed that the input 4000 data length was 3.53%,4.57% and 1.38% higher in specificity,sensitivity and accuracy than the 2000 data length.It shows that with the increase of the length of BCG signal data,the more abundant information it contains,so as to obtain better classification results.(3)Design a 1-D CNN2 multi classification algorithm based on slice spectrum feature.In order to solve the problem of large data volume and slow processing results in the 1-D cnn1 algorithm of a BCG signal,128 data length after extracting slice spectrum features is used to replace the original 4000 data length,which saves time and cost.The results showed that 1-D CNN2 improved the accuracy,sensitivity and specificity by 1.03%,0.8%and 0.24% respectively,which indicated that the slice spectrum features can effectively enhance the BCG signal features.