Method for establishing traditional Chinese medicine pulse condition recognition model based on time sequence convolutional network

Abstract

The invention discloses a method for establishing a traditional Chinese medicine pulse condition recognition model based on a time sequence convolutional network. The method comprises the steps that a pressure pulse wave signal and a volume pulse wave signal at a wrist radial artery are collected through a detection bracelet, a volume pulse wave signal at a finger tip is collected through a detection finger clip, an original pulse wave signal is obtained, and single-cycle waveforms of the original pulse wave signals are correspondingly obtained; the lengths of the single-cycle waveforms are normalized to obtain a pulse condition time sequence data set with the same length; the data set is divided into a training set, a verification set and a test set, time sequence convolutional network calculation is carried out on the training set and the verification set, verification and hyper-parameter adjustment are carried out, and finally a pulse condition recognition model is obtained; and then the test set is imported into the pulse condition recognition model for recognition and verification, and a prediction result of pulse condition diagnosis and recognition is obtained. The pulse condition recognition method has relatively high accuracy and can provide reference for diagnosis of the health state of the cardiovascular system of the human body.

Description

technical field [0001] The invention relates to the technical field of pulse condition recognition, in particular to a method for establishing a TCM pulse condition recognition model based on a time series convolutional network. Background technique [0002] The pulse signal of the human body is directly related to the heartbeat, the opening of the pulse channel and the profit and loss of Qi and blood, and has the characteristics of time-varying and non-linear. Different types of pulse signals will show obvious morphological differences. [0003] In the analysis and recognition of pulse signal, the feature extraction of pulse wave is generally carried out first, and then the classification model of pulse signal is established. Physiologically meaningful time-domain features of the pulse wave are extracted by the feature point method. This analysis method is more intuitive and widely used, but usually only reflects part of the pulse wave information; the literature uses the ...

AI Technical Summary

Problems solved by technology

Physiologically meaningful time-domain features of the pulse wave are extracted by the feature point method. This analysis method is more intuitive and widely used, but usually only reflects part of the pulse wave information; the literature uses the frequency-domain analysis method, from the statistical The frequency domain characteristics of different pulse conditions are obtained from different angles, but the shape change information in the time dimension is missing; the time-frequency domain analysis method used combines time and spectrum information, but it is more inclined to describe the characteristics of the local state of the pulse condition

It is not difficult to find that the above-mentioned analysis methods usually need to use machine learning and other algorithms to learn the feature data set after feature extraction to establish a pulse signal classification model, but the extracted features are difficult to fully reflect the shape of the pulse signal in the time domain Changes may lead to the loss of some detailed information, thereby reducing the recognition accuracy of the model

Images