Analysis And Classification Of Feminine Menstrual Circle Based On Pulse Signal

With the progress of society and the development of the times,the health condition of feminine menstrual cycle has been paid more and more attention and becomes a common concern of people.The regularity of their physiological cycle can directly reflect the health status of women’s internal secretion system,which has attracted the attention of researchers.In recent years,some relevant software has also appeared on the market to calculate the corresponding date range of different physiological stages in the next cycle by means of independent records from users,but the prediction process is not scientific and the accuracy is difficult to guarantee.And from a medical point of view,detecting physiological indicators in hospital is more accurate,but the process is tedious and costly,which is not conducive to convenience.Inspired by traditional Chinese medicine pulse diagnosis,this paper is to combine the tradition and modern data processing methods to response to this status quo,that proposing the data preprocessing technique and feature depth mining method to identify different feminine menstrual stages by pulse.The main research results are as follows:1.Aiming at the problem of residual noise in pulse signal preprocessing,a combined preprocessing technique based on wavelet threshold denoising method and waveform morphology is proposed.The joint treatment scheme eliminates different types of noises and abnormal waveform which is too wide or too narrow,and remove the related pulse with poor quality according to the strategy of quality detection,ensuring the high reliability and high quality of the signal.2.Aiming at the feature extraction of pulse signals,a feature mining technique based on wavelet transform and ResNet of deep learning model is proposed.Specifically,the time-domain of pulse signal is converted into the time-frequency domain by wavelet transform,extending the local characteristics with low dimension to high dimensional space,which realizes the basic feature enrichment for the subsequent deep learning model feature mining.Then,based on the idea of the convolutional neural network of ResNet,multiple sub-models suitable for identifying the characteristics of pulse signals in different menstrual periods are built,and the sub-models are fused by Voting strategy during the application process,so as to achieve the purpose of identifying the different menstrual stages through pulse signals.3.Based on the preprocessing and feature extraction methods mentioned above,this paper designs three sets of experiments according to different application scenarios in order to verify the accuracy of the methodology.The results show that the pulse contains the general characteristics and personalized characteristics that distinguish the luteal phase(premenstrual phase),menstrual period(mid-menstruation)and follicular phase(post-menstruation),and that the pulse characteristics on the ovulatory day of the menstrual cycle are different from those on the non-ovulatory day.