Study On Pulse Analysis Method Based On Convolution Neural Network

With the maturity of deep learning technology in recent years,deep neural network as the core of deep learning technology,its application of the model has gradually attracted people's attention.As an excellent algorithm model in deep learning domain,convolution neural network has developed rapidly in recent years,and its efficient identification method has attracted a great deal of attention.It has become one of the hot topics in many fields of science.Convolution neural network is developed on the basis of multi-layer neural network for the classification and recognition of two-dimensional images designed algorithm,which includes the input layer,output layer,hidden layer.The special structure used in convolutional neural networks not only speeds up training,but also provides such a multi-layered structure with a significant advantage in accuracy.Pulse is caused by the heart beat pressure changes caused by the aortic vessel wall vibration,the vibration along the arterial wall to the outer wall to pass the formation.Pulse has an important clinical significance.As an important body parameter,traditional Chinese medicine can diagnose the health of the body according to the direction and strength of the pulse and call it "cut-off pulse." Therefore,the use of modern data processing pulse data extraction feature extraction for analysis is of great significance.This paper presents an important analysis method for pulse non-stationary signal.Firstly,the pulse signal is transformed by continuous wavelet transform to get the two-dimensional wavelet coefficient matrix which characterizes the pulse time-frequency characteristics.Then,the wavelet coefficient matrix is used as the input of convolution neural network Layer,and then use convolutional neural network analysis.For the continuous wavelet transform of pulse signal,in order to find the suitable wavelet basis function for wavelet transform,this paper compares several commonly used basis functions of continuous wavelet transform and finds the wavelet transform suitable for input layer of convolutional neural network,that is generalized Morse Analyze wavelet transform.As an important analytical method for analyzing non-stationary signals,wavelet has many excellent features.The generalized Morse wavelet has been effectively used in the fields of seismic signal analysis and time-frequency analysis.In summary,the paper first proposed the wavelet transform of pulse signal,using the generalized Morse analytic wavelet to wavelet transform to get the wavelet coefficient matrix,and then the wavelet transform coefficient matrix obtained as convolutional neural network input layer Analysis,the pulse of different people to distinguish between,and then distinguish between different states of the pulse.Article content can be divided into the following four parts:1.Analyze the characteristics of pulse data to find the appropriate pulse data cutting method and make the cut pulse data contain the same data points.2.Using the eox mobile intelligent oximeter provided by Nanjing Pulse Electronic Technology Co.,Ltd.to collect the pulse and collect the pulse of more than a dozen young volunteers.3.Select the wavelet base which is the most similar to the pulse waveform and perform continuous wavelet transform on the pulse data after cutting to obtain the transformed wavelet coefficient matrix.4.Set up an appropriate model of convolution neural network,and use the coefficient matrix that characterizes the pulse signal as the input of convolutional neural network to analyze the pulse data.