Pulse detection method, system and device based on micro motion amplification parameter optimization

Abstract

The invention discloses a pulse detection method, system and device based on micro motion amplification parameter optimization. The method comprises the following steps: acquiring an image pulse signal through a binocular vision pulse image acquisition system; a phase motion amplification algorithm in a micro motion amplification technology is applied to amplify the out-of-plane displacement of the contact film and enhance the signal intensity of micro vibration on the contact film; space coordinate calculation and V4 algorithm interpolation curved surface fitting are carried out through the feature points to reconstruct a three-dimensional pulse signal; pulse condition recognition and classification are carried out on the reconstructed pulse signals; through the phase amplification technology and the three-dimensional pulse signal reconstruction method, the out-of-plane displacement of the contact film is enhanced in the three-dimensional pulse signal detection process, the specified frequency of the pulse signal is accurately amplified, and the three-dimensional pulse signal is accurately reconstructed.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a pulse detection method, system and device based on optimizing micro-motion amplification parameters. Background Art

[0002] The pulse signal is an important physiological signal that reflects the activity of the cardiovascular system. The main source of the pulse signal is the beating of the heart, which is the process of pushing oxygen and blood throughout the body with each beat of the heart. These pulsations are transmitted by the arteries and eventually produce perceptible pulse signals at specific locations in the body. Studies have shown that the pulse signal carries a wealth of physiological and pathological information, and the physiological information extracted from the pulse signal can be used to assess the health status of the heart.

[0003] The signal obtained by taking the pulse with the fingertips of traditional Chinese medicine has the three-dimensional properties of "pulse length-pulse width-pulse height". The two-dimensional pulse signal obtained by the traditional detection method loses the pulse length and pulse width information. Therefore, according to the principle of pulse diagnosis in traditional Chinese medicine and the principle of binocular stereoscopic vision measurement, a three-dimensional pulse signal detection system based on binocular vision is designed. The system uses a binocular camera to obtain the weak pulsation of the contact membrane caused by pulse fluctuations, calculates the spatial coordinates of multiple points, and completes the dynamic three-dimensional pulse signal reconstruction of the contact membrane shape through linear interpolation. This innovative method provides new ideas and methods for the objectification of pulse diagnosis in traditional Chinese medicine from the perspective of three-dimensional pulse signals.

[0004] However, since the pulse signal itself is relatively weak, the contact membrane displacement collected by the system is still too small, resulting in a large error in the reconstructed three-dimensional pulse signal. If the three-dimensional pulse signal cannot be accurately reconstructed, it will have a huge impact on the subsequent pulse classification.

[0005] In summary, how to effectively enhance the off-plane displacement of the contact film and accurately reconstruct the three-dimensional pulse signal, and how to accurately identify these reconstructed signals are key issues that need to be solved urgently. Summary of the invention

[0006] The embodiments of the present application provide a pulse detection method, system and device based on micro-motion amplification parameter optimization, thereby solving the technical problems in the prior art of small phase offset of three-dimensional pulse signals and low accuracy of measured pulse signals. Through phase amplification technology and three-dimensional pulse signal reconstruction method, the off-plane displacement of the contact film is enhanced during the three-dimensional pulse signal detection process, and the specified frequency of the pulse signal is accurately amplified and the three-dimensional pulse signal is precisely reconstructed.

[0007] The embodiment of the present application provides a pulse detection method based on optimizing micro-motion amplification parameters, comprising the following steps: Step S1, collecting the continuous contact film image of the traditional Chinese medicine pulse training instrument at the radial artery through a binocular vision pulse image acquisition system, obtaining the image pulse signal, and performing area selection and preprocessing; Step S2, applying the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film; The phase motion amplification algorithm is specifically: Step S21, processing the spatial domain of the video, using a complex controllable pyramid to decompose the video according to the spatial scale, direction and position, and being able to decompose the image into frequency sub-bands of different directions and scales; Step S22, by giving a specified frequency band, independently performing time domain bandpass filtering on the phase at each position, direction and ratio, and selecting the required frequency band signal; Step S23, amplifying or attenuating the phase after bandpass filtering, thereby amplifying the corresponding motion; Step S24, recombining the processed pyramids of various scales to reconstruct the enlarged video image; Step S25, adaptively optimizing the parameters in steps S21-S23 to detect and estimate, and obtain the optimal number of pyramid layers and magnification.

[0008] Step S3, after the pulse signal image is magnified, the least square method is combined to eliminate the baseline interference, the centroid of each small grid is selected as the feature point, and the spatial coordinates of the feature points are calculated and the V4 algorithm interpolation surface fitting is performed to reconstruct the three-dimensional pulse signal; Step S4, based on the random forest classification model, extract the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divide the test set and the training set, build a pulse recognition machine learning model, and perform pulse recognition and classification on the reconstructed pulse signal.

[0009] Preferably, the method of time domain bandpass filtering in step S22 includes: Step S221, Fourier transform: perform Fourier transform on the spatial domain of each scale and each direction to convert it from the spatial domain to the frequency domain; Step S222, bandpass filtering: applying a bandpass filter to the frequency domain in each scale and direction to selectively pass signals in a specific frequency band; Step S223, inverse Fourier transform: convert the signal of the specific frequency band back to the spatial domain to obtain the required frequency band signal.

[0010] Preferably, the parameter adaptive optimization detection estimation in step S25 includes: Step S251, setting the initial values ​​of the pyramid layers and magnification, and initializing them; Step S252, selecting mean square error (MSE) and peak signal-to-noise ratio (PSNR) as image quality evaluation indicators; Step S253, using the trust region optimization algorithm, find a value of the number of pyramid layers and the magnification factor so that the mean square error (MSE) is minimized and the peak signal-to-noise ratio (PSNR) is maximized.

[0011] Preferably, the method for reconstructing the three-dimensional pulse signal in step S3 includes: Step S31, dividing the amplified video into frames, and performing enhancement and Niblack binarization processing on each frame image; Step S32, calculating the centroid of each small grid of the contact film image after binarization, and taking it as a feature point; Step S33, using binocular vision principle and combining calibrated camera parameters to calculate the three-dimensional space coordinates; Step S34, using least squares fitting to remove the baseline; Step S35, using the V4 algorithm of the scattered Voronoi diagram for interpolation to obtain a smooth surface.

[0012] Preferably, the method for enhancing each frame of image in step S31 includes: performing median filtering on it using a 2x2 median filter.

[0013] Preferably, the least squares fitting and baseline removal method in step S34 includes: Step S341, store all z-axis coordinates of the three-dimensional space coordinates in the data set In, it can be expressed as ,in It is the first three-dimensional space coordinate z-axis data, It is the last three-dimensional space coordinate z-axis data, corresponding to The value is ,in It corresponds to The x-axis data, It corresponds to Find a linear function for the x-axis data:

[0014] Where m is the slope and b is the intercept; Step S342, calculating the values ​​of m and b using the least squares method; Step S343, use all the z-axis coordinates minus The new z-axis coordinate is obtained.

[0015] Preferably, the method of interpolating the scattered Voronoi diagram using the V4 algorithm in step S35 includes: Generate a grid matrix and create two matrices and , which represent all points on the grid and Coordinates are interpolated on the grid using the V4 algorithm of the scattered Voronoi diagram. In the Voronoi diagram, each region is represented as a convex hull formed by the generating point and its adjacent points; the Voronoi unit formula in the Voronoi diagram is defined as:

[0016] in: is a generating point, belonging to a given point set , each generating point Each corresponds to a Voronoi cell , Yes To point The Euclidean distance of Yes To point The Euclidean distance of and is a point in the set of generating points.

[0017] Preferably, the method for pulse signal to carry out pulse condition identification classification in step S4 comprises: Step S41: according to the waveform change of the three-dimensional pulse signal, 10 features are extracted, namely: peak value, mean value, root mean square value, standard deviation, effective value, variance, peak factor, pulse factor, waveform factor, and margin factor; Step S42: using a two-sample KS test to select features; Step S43: using a random forest classifier to perform pulse recognition and classification on the reconstructed pulse signal, and selecting accuracy, non-weighted average precision, non-weighted average F1 score and kappa coefficient as evaluation indicators of the classification performance of the classifier; Step S44: extract characteristic data from the reconstructed pulse signal according to the classification of five pulse types: deep, intermittent, surging, slippery and rapid.

[0018] The present application also proposes a pulse detection system based on optimizing micro-motion amplification parameters, comprising: The image signal acquisition module collects the continuous contact membrane image of the traditional Chinese medicine pulse training instrument at the radial artery through the binocular vision pulse image acquisition system, obtains the image pulse signal, and performs area selection and preprocessing; Phase motion amplification module, which uses the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film; The pulse image reconstruction module is a pulse signal image that has been amplified and processed. The least square method is used to eliminate baseline interference, and the centroid of each small grid is selected as a feature point. The time dimension is introduced, and the spatial coordinates of the feature points are calculated and the V4 algorithm interpolation surface is fitted to reconstruct the three-dimensional pulse signal. The pulse signal recognition module, based on the random forest classification model, extracts the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divides the test set and the training set, constructs a pulse recognition machine learning model, and performs pulse recognition and classification on the reconstructed pulse signal.

[0019] The present application also proposes a pulse detection device based on micro-motion amplification parameter optimization, comprising: a host computer and a slave computer; the host computer and the slave computer are connected and communicated via a data signal line, a ground line, and a control line; The host computer is used to execute the above-mentioned pulse detection method based on optimizing the micro-motion amplification parameters; The lower computer comprises: a base and a pulse training instrument; the base is equipped with a lifting bracket; the lifting bracket is equipped with a binocular camera and an airbag probe; The binocular camera and the airbag probe are adjusted to a suitable height by adjusting the height of the lifting bracket; the pressure of the airbag probe is adjusted so that the contact membrane of the airbag probe can generate a deformation measurement pulse signal as the pulse beats.

[0020] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: This application adopts a phase motion amplification algorithm, which effectively solves the technical problem of low measurement accuracy due to small phase offset in the pulse signal detection process. This application is suitable for enhancing certain frequency components in the signal and can amplify signals in a specific frequency range without affecting other frequency components. This algorithm not only improves the detectability and identifiability of the image pulse signal, but also has better accuracy in three-dimensional pulse image reconstruction than existing technologies.

[0021] 2. This application proposes an adaptive parameter optimization method for the number of pyramid layers and amplification factors, which makes the amplified signal closer to the original signal. The original signal is convolved with a bandpass filter to decompose the signal, which has the advantage of shorter decomposition time and reduces errors.

[0022] 3. This application also proposes a pulse classification and recognition method, specifies various data for feature extraction and evaluation indicators for classification performance, so that the pulse feature information extracted from the three-dimensional pulse signal is more accurate and objective, and provides a new perspective and method for the objective research of pulse diagnosis and pulse analysis, making pulse signal analysis standardized and intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a pulse detection method based on optimizing micro-motion amplification parameters in Embodiment 1 of the present application; Figure 2 This is a flow chart of the phase motion amplification algorithm in Example 1 of the present application; Figure 3 This is a general flow chart of three-dimensional pulse signal reconstruction in the first embodiment of the present application; Figure 4 Reconstructing three-dimensional pulse signal images of different frames in the first embodiment of the present application; Figure 5 Schematic diagram of a pulse detection system based on optimizing micro-motion amplification parameters in Embodiment 2 of the present application; Figure 6 Schematic diagram of a three-dimensional pulse image detection device for micro-motion amplification in Embodiment 3 of the present application; Figure 7 This is a schematic diagram of the structure of the lower computer in Example 3 of the present application. DETAILED DESCRIPTION

[0024] The present invention provides a pulse detection method, system and device based on optimizing micro-motion amplification parameters, utilizes a phase video amplification algorithm to amplify the micro-signals of interest, focuses on high-precision measurement of contact membrane deformation, obtains more pulse feature information from the three-dimensional reconstruction of pulse fluctuations, classifies and identifies common pulse conditions in combination with machine learning methods, and thereby achieves more accurate three-dimensional model reconstruction.

[0025] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0026] Embodiment 1

[0027] like Figure 1 As shown, a pulse detection method based on optimizing micro-motion amplification parameters proposed by the present invention comprises the following steps: Step S1, using a binocular vision pulse image acquisition system, collects the continuous contact membrane image of the traditional Chinese medicine pulse training instrument at the radial artery, obtains the image pulse signal, and performs region selection and preprocessing.

[0028] Step S2, applying the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film.

[0029] like Figure 2 As shown, the phase motion amplification algorithm is specifically: Step S21, decomposing the video into a series of frequency sub-bands of different frequencies and scales through a complex steerable pyramid, and obtaining amplitude and phase information of each scale; For spatial domain processing of videos, a complex controllable pyramid is used to decompose the video according to the spatial scale, direction and position. The image can be decomposed into frequency sub-bands of different directions and scales. The formula is as follows:

[0030] In the formula, is a frame of image signal, It is the direction and scale The bandpass filter, It is the frequency sub-bands in different directions and scales after decomposition.

[0031] Step S22, performing time-domain bandpass filtering on the image of each scale to obtain the required frequency band signal; By filtering the phase independently at each position, direction and scale given a specified frequency band, the required part is selected: Step S221, Fourier transform: for each scale, each direction Perform a Fourier transform to convert it from the spatial domain to the frequency domain:

[0032] In the formula, is an image The frequency domain representation of .

[0033] Step S222, bandpass filtering: frequency representation of each scale and direction Apply a bandpass filter to selectively pass signals in a specific frequency band:

[0034] In the formula, is the frequency domain representation after filtering, is the transfer function of the bandpass filter.

[0035] Step S223, inverse Fourier transform: convert the signal of a specific frequency band back to the spatial domain to obtain the desired frequency band signal:

[0036] In the formula, It is the image signal after time domain bandpass filtering.

[0037] Step S23, amplifying or attenuating the phase after bandpass filtering, thereby amplifying the corresponding motion; Convert the video input signal from the spatial domain to the frequency domain. Perform Fourier transform, which is expressed as follows:

[0038] Each subband here corresponds to a frequency , It represents the phase signal.

[0039] Multiply the phase by a factor , and then superimposed back to the original signal, we can get:

[0040] By performing the same processing on each component, we can obtain the amplification The motion can be amplified by processing the phase, while the amplitude will not be amplified in this process, which can effectively avoid amplifying the noise.

[0041] Step S24: recombining the processed pyramids of various scales to reconstruct the enlarged video image.

[0042] Step 25, parameter adaptive optimization detection estimation includes: Set initial values ​​for the number of pyramid levels and magnification, and then choose a larger value as the upper bound of the parameter.

[0043] The formulas for mean square error (MSE) and peak signal-to-noise ratio (PSNR) are as follows:

[0044]

[0045] In the formula, is the number of layers in the pyramid, is the number of pixels in the image, They are the original image and the enlarged image. The value of pixels, ZS is the total number of frames.

[0046] Using the trust region optimization algorithm, the following optimization problem can be defined:

[0047] In the formula, is the boundary condition obeyed by the parameters to be optimized, For the lower bound, For the upper bound.

[0048] like Figure 3 As shown, in step S3, the pulse signal image after amplification processing is combined with the least square method to eliminate baseline interference, and the centroid of each small grid is selected as the feature point. By introducing the time dimension, the feature point is calculated for spatial coordinates and the V4 algorithm interpolation surface fitting is performed to reconstruct the three-dimensional pulse signal. Figure 4 These are the reconstructed pulse waves of different frames.

[0049] Step S31, the amplified video is divided into frames, and each frame is enhanced and Niblack binarized. Regarding image enhancement, a 2x2 median filter is first used to perform median filtering, and the filter formula is as follows:

[0050] is the original image at position The pixel value of is the filtered image at position The pixel value of Is the filter in position The element value of and Relative to the current pixel The local coordinates of , where the value is -1 or 0; Niblack binarization uses the weighted sum of the local mean and standard deviation to determine the binarization threshold for each pixel. , the threshold calculation formula is usually:

[0051] in, is an empirical constant used to control the weight of contrast in the binarization process. is the local mean, is the standard deviation.

[0052]

[0053] Step S32, calculate the centroid of each small grid of the contact film image after binarization, and use it as a feature point. The formula is as follows:

[0054]

[0055] in For connected areas midpoint The gray value of is the centroid coordinate of the target area.

[0056] Step S33, using binocular vision principle and combining calibrated camera parameters to calculate the three-dimensional space coordinates; First, the Zhang Youzheng calibration method is used to obtain the internal and external parameters of the camera, and then the three-dimensional space coordinates are calculated based on the binocular vision principle. The formula is as follows:

[0057] in, is the pixel coordinate, Zc The coordinates of the camera coordinate system, is the camera extrinsic parameter, is the camera's internal parameter, are coordinates in world coordinates.

[0058] Step S34, using least squares fitting to remove the baseline, the steps are as follows: Step S341, store all z-axis coordinates in the data set In, it can be expressed as ,in It is the first three-dimensional space coordinate z-axis data, It is the last three-dimensional space coordinate z-axis data, corresponding to The value is ,in It corresponds to The x-axis data, It corresponds to Find a linear function for the x-axis data:

[0059] Where m is the slope and b is the intercept.

[0060] Step S342, calculate the values ​​of m and b using the least squares method:

[0061]

[0062] Step S343, use all the z-axis coordinates minus The new z-axis coordinate is obtained.

[0063] Step S35, using the V4 algorithm of the scattered Voronoi diagram for interpolation to obtain a smooth surface.

[0064] First generate the grid matrix and create two matrices and , which represent all points on the grid x and y Coordinates, and then interpolate on the grid using the V4 algorithm of the scattered Voronoi diagram. In the Voronoi diagram, each region can be represented as a convex hull formed by the generated point and its adjacent points. The Voronoi unit in the Voronoi diagram is usually defined by the following formula:

[0065] in: is a generating point, belonging to a given point set , each generating point Each corresponds to a Voronoi cell , Yes To point The Euclidean distance of Yes To point The Euclidean distance of and is a point in the set of generating points.

[0066] Step S4, based on the random forest classification model, extract the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divide the test set and the training set, build a pulse recognition machine learning model, and perform pulse recognition and classification on the reconstructed pulse signal.

[0067] Step S41: according to the waveform change of the three-dimensional pulse signal, 10 features are extracted, namely: peak value, mean value, root mean square value, standard deviation, effective value, variance, peak factor, pulse factor, waveform factor, and margin factor;

[0068] In the formula, Represents the pulse signal of this frame, is the signal peak.

[0069]

[0070] In the formula, is the total number of elements in this frame signal, is the mean of the signal.

[0071]

[0072] In the formula, It represents the RMS value of the signal.

[0073]

[0074] In the formula, It represents the standard deviation of the signal.

[0075]

[0076] In the formula, It represents the effective value of the signal.

[0077]

[0078] In the formula, It represents the variance of the signal.

[0079]

[0080] In the formula, It represents the crest factor of the signal.

[0081]

[0082] In the formula, It represents the impulse factor of the signal.

[0083]

[0084] In the formula, It represents the waveform factor of the signal.

[0085]

[0086] In the formula, It represents the margin factor of the signal.

[0087] Step S42: using a two-sample KS test to select features; The two-sample KS test was used to select features, and the results of the KS test are shown in Table 1; Table 1 Feature selection results

[0088] Step S43: Using the Random Forest (RF) classifier, the reconstructed pulse signal is classified and the accuracy rate ( Ac ), unweighted average precision ( Pr ) and the unweighted average F1 score ( F1 ) and kappa coefficient ( Kappa ) is used as the evaluation index of the classification performance of the classifier, and the calculation formula is as follows:

[0089]

[0090]

[0091]

[0092] In the formula, is the total number of categories, It's a certain kind. , , , are the true positive number, false positive number, false negative number, and true negative number of the classification results, respectively. is the diagonal element of the confusion matrix in its classification result, It is The sum of the row elements, It is The sum of the column elements, is the classification category, is the number of rows or columns of the confusion matrix. The result of the kappa coefficient , when it reaches 1, the classifier performs best.

[0093] The reconstructed pulse signal is used to extract feature data and classify it. The classification results of the five pulse conditions, namely, deep, intermittent, strong, slippery and rapid, are shown in Table 2.

[0094] Table 2 Classification results of five pulse conditions

[0095] As can be seen from the table, all evaluation indicators including accuracy, non-weighted average precision, non-weighted average recall, non-weighted average F1 score and kappa coefficient are over 96%.

[0096] Embodiment 2 like Figure 5 As shown, the embodiment of the present application proposes a pulse detection system based on micro-motion amplification parameter optimization, including: an image signal acquisition module 10, a phase motion amplification module 20, a pulse image reconstruction module 30, and a pulse signal recognition module 40.

[0097] The image signal acquisition module collects the continuous contact membrane image of the traditional Chinese medicine pulse training instrument at the radial artery through the binocular vision pulse image acquisition system, obtains the image pulse signal, and performs area selection and preprocessing; Phase motion amplification module, which uses the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film; The pulse image reconstruction module is a pulse signal image that has been amplified and processed. The least square method is used to eliminate baseline interference, and the centroid of each small grid is selected as a feature point. The time dimension is introduced, and the spatial coordinates of the feature points are calculated and the V4 algorithm interpolation surface is fitted to reconstruct the three-dimensional pulse signal. The pulse signal recognition module, based on the random forest classification model, extracts the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divides the test set and the training set, constructs a pulse recognition machine learning model, and performs pulse recognition and classification on the reconstructed pulse signal.

[0098] Embodiment 3 like Figure 6 As shown, the embodiment of the present application proposes a pulse detection device based on micro-motion amplification parameter optimization, including: a host computer and a slave computer; the host computer and the slave computer are connected and communicated through a data signal line, a ground line, and a control line; The host computer is used to execute the above-mentioned pulse detection method based on optimizing micro-motion amplification parameters; the host computer can be embedded in a microprocessor, a wearable device, a PC, or a smart phone to realize pulse recognition of a three-dimensional pulse image.

[0099] like Figure 7 As shown, the lower computer includes: a base and a pulse training instrument; the base is equipped with a lifting bracket; the lifting bracket is equipped with a binocular camera and an airbag probe; The binocular camera and the airbag probe are adjusted to a suitable height by adjusting the height of the lifting bracket; the pressure of the airbag probe is adjusted so that the contact membrane of the airbag probe can generate a deformation measurement pulse signal as the pulse beats.

[0100] The embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions in the process. Figure 1A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0101] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A pulse detection method based on optimizing micro-motion amplification parameters, characterized in that: The following steps are involved: Step S1, collecting the continuous contact film image of the traditional Chinese medicine pulse training instrument at the radial artery through a binocular vision pulse image acquisition system, obtaining the image pulse signal, and performing area selection and preprocessing; Step S2, applying the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film; The phase motion amplification algorithm is specifically: Step S21, processing the spatial domain of the video, using a complex controllable pyramid to decompose the video according to the spatial scale, direction and position, and being able to decompose the image into frequency sub-bands of different directions and scales; Step S22, by giving a specified frequency band, independently performing time domain bandpass filtering on the phase at each position, direction and ratio, and selecting the required frequency band signal; Step S23, amplifying or attenuating the phase after bandpass filtering, thereby amplifying the corresponding motion; Step S24, recombining the processed pyramids of various scales to reconstruct the enlarged video image; Step S25, adaptively optimizing the parameters in steps S21 to S23 to obtain the optimal number of pyramid layers and magnification; Step S3, after the pulse signal image is magnified, the least square method is combined to eliminate the baseline interference, the centroid of each small grid is selected as the feature point, and the spatial coordinates of the feature points are calculated and the V4 algorithm interpolation surface fitting is performed to reconstruct the three-dimensional pulse signal; Step S4, based on the random forest classification model, extract the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divide the test set and the training set, build a pulse recognition machine learning model, and perform pulse recognition and classification on the reconstructed pulse signal.

2. The pulse detection method according to claim 1, characterized in that: The method of time domain bandpass filtering in step S22 includes: Step S221, Fourier transform: perform Fourier transform on the spatial domain of each scale and each direction to convert it from the spatial domain to the frequency domain; Step S222, bandpass filtering: applying a bandpass filter to the frequency domain in each scale and direction to selectively pass signals in a specific frequency band; Step S223, inverse Fourier transform: convert the signal of the specific frequency band back to the spatial domain to obtain the required frequency band signal.

3. The pulse detection method according to claim 1, characterized in that: The parameter adaptive optimization detection estimation in step S25 includes: Step S251, setting the initial values ​​of the pyramid layers and magnification, and initializing them; Step S252, selecting mean square error and peak signal-to-noise ratio as evaluation indicators of image quality; Step S253, using the trust region optimization algorithm, find a value of the number of pyramid layers and the magnification factor so that the mean square error is minimized and the peak signal-to-noise ratio is maximized.

4. The pulse detection method according to claim 1, characterized in that: The method for reconstructing the three-dimensional pulse signal in step S3 includes: Step S31, dividing the amplified video into frames, and performing enhancement and Niblack binarization processing on each frame image; Step S32, calculating the centroid of each small grid of the contact film image after binarization, and taking it as a feature point; Step S33, using binocular vision principle and combining calibrated camera parameters to calculate the three-dimensional space coordinates; Step S34, using least squares fitting to remove the baseline; Step S35, using the V4 algorithm of the scattered Voronoi diagram for interpolation to obtain a smooth surface.

5. The pulse detection method according to claim 4, characterized in that: The method for enhancing each frame of image in step S31 includes: A 2x2 median filter is used to perform median filtering.

6. The pulse detection method according to claim 4, characterized in that: The least square fitting method in step S34 to remove the baseline includes: Step S341, store all z-axis coordinates of the three-dimensional space coordinates in the data set In, it can be expressed as ,in It is the first three-dimensional space coordinate z-axis data, It is the last three-dimensional space coordinate z-axis data, corresponding to The value is ,in It corresponds to The x-axis data, It corresponds to Find a linear function for the x-axis data: ， Where m is the slope and b is the intercept; Step S342, calculating the values ​​of m and b using the least squares method; Step S343, use all the z-axis coordinates minus The new z-axis coordinate is obtained.

7. The pulse detection method according to claim 4, characterized in that: The method of V4 algorithm interpolation of the scattered Voronoi diagram in step S35 includes: Generate a grid matrix and create two matrices and , which represent all points on the grid and Coordinates are interpolated on the grid using the V4 algorithm of the scattered Voronoi diagram. In the Voronoi diagram, each region is represented as a convex hull formed by the generating point and its adjacent points; the Voronoi unit formula in the Voronoi diagram is defined as: ， in: is a generating point, belonging to a given point set , each generating point Each corresponds to a Voronoi cell , Yes To point The Euclidean distance of Yes To point The Euclidean distance of and is a point in the set of generating points.

8. The pulse detection method according to claim 1, characterized in that: The method for pulse signal to carry out pulse condition identification classification in the described step S4 comprises: Step S41: according to the waveform change of the three-dimensional pulse signal, 10 features are extracted, namely: peak value, mean value, root mean square value, standard deviation, effective value, variance, peak factor, pulse factor, waveform factor, and margin factor; Step S42: using a two-sample KS test to select features; Step S43: using a random forest classifier to perform pulse recognition and classification on the reconstructed pulse signal, and selecting accuracy, non-weighted average precision, non-weighted average F1 score and kappa coefficient as evaluation indicators of the classification performance of the classifier; Step S44: extract characteristic data from the reconstructed pulse signal according to the classification of five pulse types: deep, intermittent, surging, slippery and rapid.

9. A pulse detection system based on optimizing micro-motion amplification parameters, characterized in that: include: The image signal acquisition module collects the continuous contact membrane image of the traditional Chinese medicine pulse training instrument at the radial artery through the binocular vision pulse image acquisition system, obtains the image pulse signal, and performs area selection and preprocessing; Phase motion amplification module, which uses the phase motion amplification algorithm in the micro-motion amplification technology to amplify the off-surface displacement of the contact film and enhance the signal strength of the micro-vibration on the contact film; The pulse image reconstruction module is a pulse signal image that has been amplified and processed. The least square method is used to eliminate baseline interference, and the centroid of each small grid is selected as a feature point. The time dimension is introduced, and the spatial coordinates of the feature points are calculated and the V4 algorithm interpolation surface is fitted to reconstruct the three-dimensional pulse signal. The pulse signal recognition module, based on the random forest classification model, extracts the specific change parameters of the three-dimensional pulse signal to form a feature vector set, divides the test set and the training set, constructs a pulse recognition machine learning model, and performs pulse recognition and classification on the reconstructed pulse signal.

10. A pulse detection device based on optimizing micro-motion amplification parameters, characterized in that: include: Upper computer and lower computer; The upper computer and the lower computer are connected and communicated via a data signal line, a ground line, and a control line; The host computer is used to execute the method according to any one of claims 1 to 8; The lower computer comprises: a base and a pulse training instrument; the base is equipped with a lifting bracket; the lifting bracket is equipped with a binocular camera and an airbag probe; The binocular camera and the airbag probe are adjusted to a suitable height by adjusting the height of the lifting bracket; the pressure of the airbag probe is adjusted so that the contact membrane of the airbag probe can generate a deformation measurement pulse signal as the pulse beats.

Images