CN116089797A - Pulse condition identification method and system based on convolutional neural network - Google Patents

Pulse condition identification method and system based on convolutional neural network Download PDF

Info

Publication number
CN116089797A
CN116089797A CN202310082129.0A CN202310082129A CN116089797A CN 116089797 A CN116089797 A CN 116089797A CN 202310082129 A CN202310082129 A CN 202310082129A CN 116089797 A CN116089797 A CN 116089797A
Authority
CN
China
Prior art keywords
pulse
neural network
pulse condition
convolutional neural
time domain
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310082129.0A
Other languages
Chinese (zh)
Inventor
丁泽
张枫宁
张奕扬
李晓雨
王凤霞
陈涛
孙立宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou University
Original Assignee
Suzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou University filed Critical Suzhou University
Priority to CN202310082129.0A priority Critical patent/CN116089797A/en
Publication of CN116089797A publication Critical patent/CN116089797A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/48—Other medical applications
    • A61B5/4854—Diagnosis based on concepts of traditional oriental medicine
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7253—Details of waveform analysis characterised by using transforms
    • A61B5/7257—Details of waveform analysis characterised by using transforms using Fourier transforms
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Pathology (AREA)
  • Public Health (AREA)
  • Mathematical Physics (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Fuzzy Systems (AREA)
  • Alternative & Traditional Medicine (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Cardiology (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

The invention provides a pulse condition identification method and a system based on a convolutional neural network, which are characterized in that signals of three-dimensional pulse conditions of a plurality of positions of a cun-guan ruler are acquired through a flexible sensor, the wearable acquisition of the pulse conditions is realized through a designed circuit board, the acquired pulse conditions are converted into frequency domain signals through fast Fourier change, and the preprocessed signals are trained through the convolutional neural network to generate a neural network model so as to realize the real-time pulse condition identification of follow-up wearable equipment. The method utilizes the acquired three-dimensional pulse condition data of the cunguan ruler to perform pulse condition identification based on a convolutional neural network, and utilizes the characteristic that the convolutional neural network performs information transmission layer by simulating a human neuron structure to realize automatic extraction of characteristics and finally realize pulse condition identification. Meanwhile, the pulse condition training model provided by the invention can be used in the fields of pulse condition identification, pulse analysis and the like. The invention designs a flexible pressure sensor based on ionic gel, which has good stability, excellent sensitivity and good biocompatibility.

Description

Pulse condition identification method and system based on convolutional neural network
Technical Field
The invention relates to the technical field of traditional Chinese medicine pulse diagnosis and the field of computer data processing, in particular to a pulse condition identification method based on a convolutional neural network.
Background
The four diagnosis methods of "Wang, shang, mi, and Chen" are the methods of diagnosing patients in Chinese national traditional medicine, wherein "Cheng" generally represents pulse diagnosis, and traditional Chinese doctors feel the physique of the human body by pressing the pulse of three positions of the radial artery cun guan chi with fingers, so that the pulse diagnosis of traditional Chinese medicine has important clinical value. However, the accuracy is mainly dependent on subjective judgment and experience of doctors, and lacks objective diagnosis indexes, so that the accuracy is necessary to be combined with modern computer technology in the modern development process of traditional Chinese medicine so as to promote the integration of the traditional Chinese medicine with modern and scientific development.
There are many studies currently on the acquisition of pulse signals, and the sensor principles are generally photoelectric, pressure, piezoelectric, capacitive and triboelectric. In order to truly simulate the fingertip feeling when a doctor presses a pulse, the principle of the pressure sensor is most similar to that. In the field of signal analysis, many scholars have studied the method, and the analysis method mainly comprises a time domain analysis method, a frequency domain analysis method and the like. The time domain analysis method is to define a plurality of time domain features with physiological significance according to pulse signals, wherein the time domain features mainly comprise main peak value of pulse, dicrotic wave, amplitude of canyon wave, pulse time and the like; the frequency domain analysis method mainly converts a time domain signal into a frequency domain signal through fast Fourier transform, and performs frequency domain feature extraction; for preprocessed signals, the prior art discloses a pulse condition signal identification method based on a thresholdless recursion chart and a convolution neural network, and the method converts pulse condition signals into the recursion chart, so that nonlinear characteristics are mapped into a two-dimensional plane, and learning and classification of the pulse condition signal characteristics are realized through multi-layer convolution calculation of the convolution neural network, so that the pulse condition signal identification method has the capability of distinguishing different pulse condition types.
In pulse signal pattern recognition research, the prior art mainly adopts convolutional neural networks, counter-propagating neural networks, LSTM neural networks and the like to judge and analyze pulse signals. However, the pulse condition identification accuracy rate in the prior art is not high, or the adopted technical means are complex, and the popularization and the application are difficult.
Disclosure of Invention
The aim of the invention is achieved by the following technical scheme.
According to a first aspect of the present invention, there is provided a pulse condition recognition system based on a convolutional neural network, comprising:
acquiring pulse time domain characteristics to be identified:
the pulse time domain characteristics are sent to a pulse identification model based on a convolutional neural network to carry out pulse identification, and a pulse identification result is obtained;
the pulse condition identification model is obtained through the following steps:
step one: collecting pulse condition training data
Collecting atrial fibrillation, coronary heart disease, hypertension and radial artery multichannel pulse time domain data of patients with cardiac pacemakers;
collecting pulse condition time domain data of healthy people;
performing label classification on all the pulse time domain signals to obtain a training data set;
step two: building pulse condition identification convolutional neural network
The pulse condition identification convolutional neural network comprises: an input layer, a convolution layer, a maximum pooling layer, a flat layer, a full connection layer and an output layer;
step three: convolutional neural network for training pulse condition identification
And converting the multichannel pulse condition time domain signal into a frequency domain signal through Fourier transformation, inputting the frequency domain signal serving as an input layer into the pulse condition recognition convolutional neural network, training the pulse condition recognition convolutional neural network by using the training data set, and obtaining the pulse condition recognition model after training.
Further, the first step specifically includes:
collecting the pulse condition data of patients with cardiovascular diseases, and classifying and calibrating the data according to the traditional Chinese medicine pulse condition characteristics and western medicine electrocardiogram diagnosis characteristics;
collecting pulse condition time domain data of a healthy person, wherein the pulse condition time domain data comprises two parameters of time and voltage;
and carrying out label classification on all the pulse time domain signals to obtain a training data set.
Further, the method for acquiring the pulse time domain data comprises the following steps:
the method comprises the steps of collecting pulse 9X 4 channel data of three positions of a radial artery cun guan chi of a patient and a healthy person through a flexible sensing module, wherein each position is provided with 4X 3 sensors, each position is provided with four sensors in the pulse width direction, and the length direction is provided with three sensors; the flexible sensing module comprises a flexible sensing array and an air bag;
and sending the channel data to a signal acquisition module to obtain the pulse time domain data.
Further, the signal acquisition module comprises an FPGA acquisition circuit, wherein the FPGA acquisition circuit is a four-layer circuit board and comprises the following electronic components: the system comprises an operational amplifier circuit, an ADS chip, an FPGA processing chip and an RS485 serial port.
Further, in the training process of the pulse condition recognition convolutional neural network by using the training data set, a random gradient descent algorithm is used for iterating and updating the convolutional kernel state to operate, and a random gradient descent is used for searching a global optimal solution.
Further, the pulse condition identification convolutional neural network is based on a TensorFlow framework.
Further, the number of the convolution layers is multiple, the convolution layers are connected through convolution filters, and the definition of the convolution filters is that the number of channels is multiplied by the signal format and the type of the filter.
Further, the flexible sensing array includes a flexible pressure sensor prepared from an ion gel-based.
Further, the method further comprises the steps of:
the pulse condition identification result is sent to a display module for display, and the display module displays the pulse condition identification result for a raspberry group display or a mobile phone APP.
According to a second aspect of the present invention, there is provided a pulse condition recognition system based on a convolutional neural network, comprising:
the flexible sensing module comprises a flexible sensing array and an air bag;
the signal acquisition module comprises an FPGA acquisition circuit and an RS485 transmission module;
the intelligent sensing module comprises a neural network module and a display module;
the pulse time domain characteristics to be identified are obtained through the flexible sensing module and the signal acquisition module:
and sending the pulse time domain characteristics to a pulse recognition model based on a convolutional neural network in the neural network module for pulse recognition, so as to obtain a pulse recognition result.
The invention has the advantages that: the method utilizes the acquired three-dimensional pulse condition data of the cunguan ruler to perform pulse condition identification based on a convolutional neural network, and utilizes the characteristic that the convolutional neural network performs information transmission layer by simulating a human neuron structure to realize automatic extraction of characteristics and finally realize pulse condition identification. Meanwhile, the pulse condition training model provided by the invention can be used in the fields of pulse condition identification, pulse analysis and the like. The invention designs a flexible pressure sensor based on ionic gel, which has good stability, excellent sensitivity and good biocompatibility.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
FIG. 1 is a schematic flow chart of a pulse condition recognition method based on a convolutional neural network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram showing the steps of pulse condition collection and classification in the embodiment of the present invention;
FIG. 3 is a diagram of a multi-dimensional pulse acquired by the present invention;
FIG. 4 is a schematic diagram of a convolutional neural network framework employed in the present invention;
FIG. 5 is a schematic diagram showing the components of the pulse condition recognition system according to the present invention;
FIG. 6 is a flow chart of the system for diagnosing pulse conditions according to the present invention;
FIG. 7 is a schematic diagram of a pulse condition sensor according to the present invention;
FIG. 8 is a circuit diagram of an operational amplifier of the present invention;
FIG. 9 is a diagram of an FPGA acquisition circuit of the present invention;
FIG. 10 is a circuit diagram of a pulse FPGA of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The innovation point of the invention is that the flexible sensor designed and manufactured based on the traditional Chinese medicine theory is used for collecting signals of three-dimensional pulse conditions of multiple parts of the cun guan chi, the designed circuit board is used for realizing wearable collection of the pulse conditions, the collected pulse conditions are converted into frequency domain signals through fast Fourier change, the preprocessed signals are trained through a convolutional neural network, a neural network model is generated, and the real-time pulse condition identification of the follow-up wearable equipment is realized.
The invention uses the convolutional neural network to classify and train the pulse signals, because the convolutional neural network has excellent capability in signal identification. In addition, the invention designs a flexible pressure sensor based on ionic gel, which has good stability, excellent sensitivity and good biocompatibility. The system can accurately distinguish cardiovascular disease pulse conditions such as: atrial fibrillation, coronary heart disease, hypertension, etc.
In one embodiment, as shown in fig. 1, the pulse condition identification method based on the convolutional neural network provided by the invention has an overall flow chart shown in fig. 1. The pulse condition identification method based on the convolutional neural network in the embodiment mainly comprises two parts: one part is the assembly of system hardware and the other part is the generation of neural network models. The neural network model generation mainly comprises the following steps: firstly, standard data of pulse conditions of healthy people and patients with cardiovascular diseases are collected; and secondly, building a neural network model, inputting the acquired data into the neural network model, performing training test, and finally generating a pulse condition identification model. The identification of the cardiovascular disease pulse type can be realized by loading the pulse identification model into the intelligent perception module shown in fig. 5.
In the process of training the pulse condition identification convolutional neural network by using the training data set, a random gradient descent algorithm is used for iterating and updating the convolutional kernel state to operate, and a random gradient descent is used for searching a global optimal solution. The pulse condition identification convolutional neural network is based on a TensorFlow framework.
The pulse condition acquisition process is shown in figure 2, the flexible sensor is used for acquiring the pulse condition data of the cardiovascular disease patient, and meanwhile, the data are classified and marked according to the traditional Chinese medicine pulse condition characteristics and western medicine electrocardiogram diagnosis characteristics such as atrial fibrillation, coronary heart disease, hypertension and the like. The invention collects 250 groups of data of patients suffering from atrial fibrillation and coronary heart disease. The healthy patient data collected by the present invention is 500 groups. Classifying and calibrating the acquired data to obtain a healthy person data set, a disease data set and secondarily classifying the disease data set, wherein the healthy person data set, the disease data set and the disease data set comprise an atrial fibrillation patient data set, a coronary heart disease patient data set, a hypertension data set and a heart stent-mounted patient data set.
The data acquired by the pulse condition is shown in figure 3, the invention acquires the data of three pulse conditions of cunguan and chi at the radial artery by using the 4X 9 multichannel flexible sensor, and the spatial multidimensional pulse condition information is formed by fitting the acquired data of a plurality of sensing units, wherein common pulse conditions include slippery pulse, flat pulse and wiry pulse. In the process of marking data, not only the acquired pulse data of the cardiovascular disease patient are marked, but also the pathological characteristics of Western medicine such as atrial fibrillation, coronary heart disease and the like are marked. Inputting the acquired pulse condition into a convolutional neural network model, training and finally obtaining the neural network model.
The neural network model framework provided by the invention is shown in fig. 4, the convolutional neural network is trained under a TensorFlow framework, and the neural network has a multi-layer convolutional structure. Before standard cardiovascular disease pulse condition data and healthy human pulse condition data are input into a neural network, firstly preprocessing the data, and firstly calculating root mean square error of the data:
Figure BDA0004067744750000051
where Xi is a set of pulse condition data, pi is the average value of the set of data, and n is the number of the set of data. Then the fast Fourier transform is utilized to convert the time domain signal into the frequency domain signal, and the principle of Fourier series transform for the continuous periodic signal of the pulse wave is as follows:
Figure BDA0004067744750000061
Figure BDA0004067744750000062
Figure BDA0004067744750000063
Figure BDA0004067744750000064
where x (t) is a Fourier transformed function, a 0 Representing the mean value of the signals, and calculating the mean value when each group of signals are input, a n 、b n Representing the amplitude, w, of the signal 0 Representing the frequency of the signal, T is the period of the signal, T is the time of the signal, and the normalization process is carried out on the signal. And then, in the convolutional neural network of the data input value, the neural network framework is shown in fig. 4, the input layer is a three-layer one-dimensional convolutional neural network, the number of the convolutional layers is multiple, the multiple convolutional layers are connected through convolutional filters, and the definition of the convolutional filters is that the number of channels is multiplied by the type of the signal format multiplied by the filter. The filters (filters) are all 4, the kernel (kernel) size is 5, and the activation functions used are all ReLU activation functions.
The array of each input neural network of the activation function f (x) =max (0, x) is 36×576, wherein 36 is the number of sensors, data acquired by the flexible sensor known as 9×4 array is 576 is the length of the array, and the volume of data input each time is H 1 *W 1 *D 1 Wherein H is 1 For the height of the input data, the value is fixed at 576, W 1 For the input data width, its value is fixed at 36, D 1 The depth of the input data is here 1. Outputting according to the convolution neural network formula
Figure BDA0004067744750000065
Figure BDA0004067744750000066
D 2 =3. Wherein W is 2 、H 2 、D 2 The width, the height and the depth of output data are respectively, F is the size of Filter, and P is the size of zero padding.
The invention also adds a maximum pooling layer in the one-dimensional convolution network, the activation function is a ReLU activation function, and the output layer consists of the following parts: flat layer and full tie layer. According to the convolutional neural network-based output layer, the random neuron discarding layer dropout is added, and the discarding coefficient is 0.5. Training the neural network by using TensorFlow to obtain a final pulse condition identification neural network model.
The hardware structure diagram of the pulse condition identification system provided by the invention is shown in figure 5, and mainly comprises the following 3 modules; the system comprises an ion gel-based flexible sensing module, an FPGA-based signal acquisition module and an intelligent sensing module based on a convolutional neural network. The flexible sensing module based on ionic gel is obtained by heating ionic liquid and is integrated with a flexible circuit board to form a flexible sensing array, and FIG. 7 is a schematic diagram of the flexible sensing array of the invention; based on the three pulse conditions 9×4 channel data of the radial artery cun guan chi captured by the flexible sensing array, each position has 4×3 sensors, each position has four sensors in pulse width and three sensors in length direction. The flexible perception module is composed of two parts: the flexible sensing array and the air bag are arranged on the acquisition device by a tester in the acquisition process, and the air bag is pressurized by the air pump to achieve the purpose of sinking pulse; in the signal acquisition module, an FPGA acquisition circuit is a four-layer circuit board and mainly comprises the following electronic components: the high-precision digital signal processing device comprises an operational amplifier circuit, a plurality of high-precision ADS chips (figure 9), an FPGA processing chip and an RS485 serial port. The FPGA processing chip is as follows: the atlas EP3C10E144A7N (fig. 10) has an acquisition frequency of 50 hz, and the signal needs to be amplified and filtered before entering the circuit board, the amplifying chip is MCP6002, the amplification factor is 121 times, and the operational amplifier circuit is shown in fig. 8.
In fig. 8, two MCP6002 op-amps are included, with capacitors C1, C2 of 01.U, C3 of 10nF, C4, C5 of 10U, C of 0.22U, C7 of 0.22U. The resistors R1, R2, R3, R5, R7 and R8 are 10K, R4 is 47K and R6 is 100K. One end of the C3 is grounded, and the other end of the C3 is connected with the positive input end of the U1A operational amplifier; one end of R4 is grounded, and the other end is connected with the positive input end of the U1A operational amplifier; one end of C5 is grounded, and the other end is connected with the negative input end of U1A through R8; r6 and C6 are connected in parallel, one end of the R6 is connected with the negative input end of U1A, and the other end of the R6 is connected with the output end of U1A; one end of C4 is connected with the output end of U1A, and the other end is connected with the negative input end of U1B operational amplifier through R5; r7 and C7 are connected in parallel, one end of the R7 is connected with the negative input end of the U1B, and the other end of the R7 is connected with the output end of the U1B; c2 and R2 are connected in parallel, one end is grounded, and the other end is connected with the positive input end of U1B; one end of the resistor R1 is connected with the power VCC, and the other end is connected with the positive input end of the U1B; r3 is connected with the output end of U1B.
Fig. 9 is an ADS1256 chip and its peripheral circuits, and fig. 10 is a schematic diagram of a pipe teaching connection mode of EP3C10E144A7N, which is a more conventional usage mode in the art, and will not be described herein.
The intelligent recognition module comprises a neural network model, different types of neural network models are adopted according to different display modules, if the neural network model is displayed on a computer end, the trained convolutional neural network model is directly used, if the neural network model is displayed on a smart phone end or a raspberry pie end, the trained convolutional neural network model is converted into a TensorFlow Lite model, and then the TensorFlow Lite model is placed in the intelligent recognition module.
The cardiovascular pulse condition identification system based on the convolutional neural network provided by the invention has the advantages that the signals are acquired and identified in the use process as shown in figure 6. The wrist of the tester is placed on the acquisition device, three pulse signals of the radius artery cun guan chi of the tester are captured by the flexible sensor, and the signals are time domain signals of time and voltage; the signals enter the ADS chip, analog signals are converted into digital signals, the FPGA transmits the signals to the neural network model through a serial port, time domain signals are converted into frequency domain signals, normalization processing is carried out on the frequency domain signals, and the frequency domain signals and the trained neural network model are compared to obtain diagnosis results; the results appear to be healthy or to have cardiovascular disease, specific categories of cardiovascular disease are the following: atrial fibrillation, coronary heart disease, hypertension and patients with heart stents.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. A pulse condition identification method based on a convolutional neural network is characterized by comprising the following steps:
acquiring pulse time domain characteristics to be identified:
the pulse time domain characteristics are sent to a pulse identification model based on a convolutional neural network to carry out pulse identification, and a pulse identification result is obtained;
the pulse condition identification model is obtained through the following steps:
step one: collecting pulse condition training data
Collecting atrial fibrillation, coronary heart disease, hypertension and radial artery multichannel pulse time domain data of patients with cardiac pacemakers;
collecting pulse condition time domain data of healthy people;
performing label classification on all the pulse time domain signals to obtain a training data set;
step two: building pulse condition identification convolutional neural network
The pulse condition identification convolutional neural network comprises: an input layer, a convolution layer, a maximum pooling layer, a flat layer, a full connection layer and an output layer;
step three: convolutional neural network for training pulse condition identification
And converting the multichannel pulse condition time domain signal into a frequency domain signal through Fourier transformation, inputting the frequency domain signal serving as an input layer into the pulse condition recognition convolutional neural network, training the pulse condition recognition convolutional neural network by using the training data set, and obtaining the pulse condition recognition model after training.
2. The pulse condition identification method based on convolutional neural network according to claim 1, wherein,
the first step specifically comprises the following steps:
collecting the pulse condition data of patients with cardiovascular diseases, and classifying and calibrating the data according to the traditional Chinese medicine pulse condition characteristics and western medicine electrocardiogram diagnosis characteristics;
collecting pulse condition time domain data of a healthy person, wherein the pulse condition time domain data comprises two parameters of time and voltage;
and carrying out label classification on all the pulse time domain signals to obtain a training data set.
3. The pulse condition identification method based on convolutional neural network according to claim 1, wherein,
the method for collecting pulse time domain data comprises the following steps:
the method comprises the steps of collecting pulse 9X 4 channel data of three positions of a radial artery cun guan chi of a patient and a healthy person through a flexible sensing module, wherein each position is provided with 4X 3 sensors, each position is provided with four sensors in the pulse width direction, and the length direction is provided with three sensors; the flexible sensing module comprises a flexible sensing array and an air bag;
and sending the channel data to a signal acquisition module to obtain the pulse time domain data.
4. The pulse condition identification method based on convolutional neural network according to claim 1, wherein,
the signal acquisition module comprises an FPGA acquisition circuit, wherein the FPGA acquisition circuit is a four-layer circuit board and comprises the following electronic components: the system comprises an operational amplifier circuit, an ADS chip, an FPGA processing chip and an RS485 serial port.
5. The pulse condition identification method based on convolutional neural network according to claim 1, wherein,
and in the process of training the pulse condition recognition convolutional neural network by using the training data set, iterating and updating the state of the convolutional kernel by using a random gradient descent algorithm to operate, and searching for a global optimal solution by using random gradient descent.
6. The pulse condition identification method based on convolutional neural network according to claim 1, wherein,
the pulse condition identification convolutional neural network is based on a TensorFlow framework.
7. A pulse condition identification method based on convolutional neural network as defined in claim 3, wherein,
the number of the convolution layers is multiple, the convolution layers are connected through convolution filters, and the definition of the convolution filters is that the channel number is multiplied by the signal format and the type of the filter is multiplied.
8. The pulse condition identification method based on convolutional neural network according to claim 3, wherein,
the flexible sensing array includes a flexible pressure sensor made from an ion gel-based.
9. The convolutional neural network-based pulse condition recognition method of claim 1, further comprising:
the pulse condition identification result is sent to a display module for display, and the display module displays the pulse condition identification result for a raspberry group display or a mobile phone APP.
10. A pulse condition recognition system based on convolutional neural network, comprising:
the flexible sensing module comprises a flexible sensing array and an air bag;
the signal acquisition module comprises an FPGA acquisition circuit and an RS485 transmission module;
the intelligent sensing module comprises a neural network module and a display module;
the pulse time domain characteristics to be identified are obtained through the flexible sensing module and the signal acquisition module:
and sending the pulse time domain characteristics to a pulse recognition model based on a convolutional neural network in the neural network module for pulse recognition, so as to obtain a pulse recognition result.
CN202310082129.0A 2023-02-08 2023-02-08 Pulse condition identification method and system based on convolutional neural network Pending CN116089797A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310082129.0A CN116089797A (en) 2023-02-08 2023-02-08 Pulse condition identification method and system based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310082129.0A CN116089797A (en) 2023-02-08 2023-02-08 Pulse condition identification method and system based on convolutional neural network

Publications (1)

Publication Number Publication Date
CN116089797A true CN116089797A (en) 2023-05-09

Family

ID=86213748

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310082129.0A Pending CN116089797A (en) 2023-02-08 2023-02-08 Pulse condition identification method and system based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN116089797A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116649924A (en) * 2023-06-06 2023-08-29 湖南敬凯投资管理有限公司 Pulse analysis method and device

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116649924A (en) * 2023-06-06 2023-08-29 湖南敬凯投资管理有限公司 Pulse analysis method and device

Similar Documents

Publication Publication Date Title
CN102144916B (en) Multi-channel pulse signal detecting method and device capable of automatically regulating pressure
CN101815466B (en) A non-invasive device NADI TARANGINI useful for quantitative detection of arterial NADI pulse waveform
CN103892818B (en) A kind of non-invasive central arterial blood pressure measuring method and equipment
Li et al. Design of a continuous blood pressure measurement system based on pulse wave and ECG signals
CN112487945B (en) Pulse condition identification method based on double-path convolution neural network fusion
CN105769173A (en) Electrocardiogram monitoring system with electrocardiosignal denoising function
CN112089405B (en) Pulse wave characteristic parameter measuring and displaying device
CN202960481U (en) Traditional Chinese medicine pulse condition acquisition device
CN202920160U (en) Traditional Chinese medicine pulse condition collection system
CN109674464B (en) Multi-lead electrocardiosignal composite feature extraction method and corresponding monitoring system
CN116089797A (en) Pulse condition identification method and system based on convolutional neural network
CN115064246A (en) Depression evaluation system and equipment based on multi-mode information fusion
CN107397542B (en) Dynamic blood pressure monitoring wearable device based on pulse wave sensor and monitoring method
CN105796091B (en) A kind of intelligent terminal for removing electrocardiosignal vehicle movement noise
CN105708441B (en) Wearable fingerstall and electrocardiogram and pulse-tracing collection terminal
CN113413163B (en) Heart sound diagnosis system for mixed deep learning and low-difference forest
CN107970027A (en) A kind of radial artery detection and human body constitution identifying system and method
CN110638482B (en) Real-time monitoring system for bowel sound and abdominal pressure
CN113143270A (en) Bimodal fusion emotion recognition method based on biological radar and voice information
CN109431499B (en) Botanic person home care auxiliary system and auxiliary method
CN107928636B (en) Pulse diagnosis instrument with temperature compensation function
CN113397500B (en) Pulse monitoring device
CN109009005A (en) A kind of wearable Chinese medicine pulse acquisition and analysis system
CN114903475A (en) Health detector based on PCANet neural network and multi-sensor information fusion and method thereof
CN111110207A (en) Pulse diagnosis instrument based on flexible piezoelectric sensor array

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination