Research On Female Graduate Student Condition Based On Pulse Signal

The rapid development of medical technology and artificial intelligence technology brings breakthrough in depth and breadth in the field of health.The health monitoring way of modern people is not limited to hospitals and other medical clinics.With the development of online consultation and self-health testing systems,a growing number of people participate in the test based on health big data,also driven the development of related product technology and industry.Modern Western Medicine uses ECG signal as a main detection evaluation object.This signal can accurately reflect human body condition,and it is easy to make a diagnosis,but the acquisition method is slightly complicated.In China,the pulse diagnosis used in Chinese medicine as an important test method of human health,and the Chinese medicine doctor measured the touch of the pulse signal to make a diagnosis.According to the pulse signal,either strong or weak,the waveform was judged to tell the physical condition of the patient.The pulse signal is collected into digital data while classified into different data sets according to its scene.Deep learning networks use features to classify human condition automatically and it has broad application prospects.This paper studies a deep learning method in different physiological cycles based on finger-tip pulse research and judgment.The paper completed the following work:(1)More than 150 female graduate students of Beijing University of Posts and Telecommunications were selected as volunteers to collect finger-tip pulse,and the pulse classification data sets of different female cycles were formed after data pretreatment and manually labelled.(2)The ResNet and DenseNet network model espacially for pulse signal processing were built to train pulse datasets.Three sets of model and two sets of model were individually built to make three-stage classification process of two female physiological cycles was constructed.In the network design,the residual module of the ResNet and the dense connection of the DenseNet are used to capture more features to optimize network training result.In the experiment,using SMOTE method to enhance the minority data in the data set,using Dropout method to make random deactivation probability of neuron design,both achieved a good optimization effect.