Human body acupoint recognition method and system based on artificial intelligence image analysis

Abstract

The invention discloses a human body acupoint recognition method and system based on artificial intelligence image analysis. The human body acupoint recognition method carries out recognition through six steps. The method can accurately identify the acupuncture points, and has the advantages of high real-time performance and high adaptability. The beneficial effects of the invention are that: 1, the structure of a multi-task CNN backbone network and a cascade regression device is adopted, and precise positioning and real-time identification of acupuncture points are realized; 2, the method is also combined with a position coding mechanism of anatomical knowledge, so that the accuracy and robustness of recognition are improved; 3, through a multi-scale feature extraction and iterative optimization strategy, adapting to different body forms and attitude changes; a real-time guiding function is achieved, and the efficiency and safety of acupuncture operation are improved; and 5, the system has good expansibility and practicability, and provides technical support for standardization and modernization of acupuncture treatment.

Description

Technical Field

[0001] The present invention relates to the field of computer science and technology, and in particular to a method and system for identifying human acupoints based on artificial intelligence image analysis. Background Art

[0002] As a traditional treatment method that has been passed down for thousands of years, acupuncture has significant effects in treating pain, regulating endocrine, improving sleep, and enhancing immunity. However, the effectiveness of acupuncture treatment depends largely on the accuracy of acupoint positioning. Traditional acupoint positioning methods mainly rely on acupuncturists to complete through palpation and visual inspection, combined with their many years of clinical experience and professional training. This manual positioning method has the following problems: 1. Time-consuming operation: Each acupoint needs to be carefully touched and measured, and the entire positioning process is time-consuming; 2. High subjectivity: The positioning results of different physicians may differ; 3. Susceptible to interference: factors such as the doctor's physical condition and ambient light may affect the positioning accuracy; 4. Individual differences: Each patient has different body shapes and bone structures, which increases the difficulty of accurate positioning.

[0003] In recent years, computer vision and deep learning technologies have made significant breakthroughs in the fields of human key point detection and posture estimation. Models trained based on convolutional neural networks and large-scale data sets can accurately locate human skeletal joints and facial feature points, and effectively learn the spatial relationship between them. However, applying these technologies to the field of acupoint recognition still faces many challenges: 1. The correspondence between acupoint locations and anatomical landmarks is complex, and a more sophisticated feature mapping needs to be established; 2. The relative positions of acupoints in people of different body types vary greatly, which increases the difficulty of generalizing the model; 3. Some acupoints lack obvious visual features and are difficult to locate accurately based on image information alone.

[0004] Therefore, in view of the shortcomings of the existing technology, it is necessary to provide a human acupoint recognition method and system based on artificial intelligence image analysis to solve the shortcomings of the existing technology. Summary of the invention

[0005] The first object of the present invention is to avoid the shortcomings of the prior art and provide a method for identifying human acupoints based on artificial intelligence image analysis. The method for identifying human acupoints based on artificial intelligence image analysis can accurately identify acupoints and has the advantages of strong real-time performance and strong adaptability.

[0006] The above-mentioned purpose of the present invention is achieved by the following technical measures: A method for identifying human acupoints based on artificial intelligence image analysis is provided, which is performed through the following steps: S1, collecting the human body surface image to be processed; S2, preprocessing the human body surface image obtained in S1; S3, extracting visual feature maps and human posture parameters from the human body surface image pre-processed by S2 through the multi-task CNN backbone network in the trained acupoint recognition model; the trained acupoint recognition model is obtained by training the acupoint recognition model with the acupoint database, wherein the acupoint recognition model is composed of a multi-task CNN backbone network, a 3D human meridian acupoint template and a cascade regressor; S4: Generate the initial estimated position of the acupoints on the human body surface image based on the 3D human meridian acupoint template in the trained acupoint recognition model and the human body posture parameters obtained in S3 ; S5, the initial estimated position of S4 obtained by the cascade regressor in the trained acupoint recognition model The visual feature map obtained by S3 is processed to obtain the coordinates of the acupuncture points in the human body surface image; S6. Post-process the acupoint coordinates obtained in S5 to obtain an acupoint recognition result.

[0007] Preferably, the acupoint database includes basic acupoint data.

[0008] Preferably, the process of acquiring the above basic acupoint data is as follows: A1. Collect multiple images of the human body surface; A2. Preprocess the human body surface image obtained in A1 and mark the acupoint information, wherein the marked acupoint information includes the standard name of the acupoint, the anatomical location description of the acupoint, the meridian system to which the acupoint belongs, the main function of the acupoint and the indication of the acupoint, so as to obtain the basic acupoint data.

[0009] Preferably, the acupoint database also includes spatial positioning data.

[0010] Preferably, the process of acquiring the above-mentioned spatial positioning data is as follows: B1. Scan multiple human body 3D data; B2. Annotate the three-dimensional coordinate information of acupoints, the positional relationship of the human anatomical landmarks of acupoints, the relative distance between different acupoints, and the angle data between different acupoints on the three-dimensional data of the human body to obtain the spatial positioning data.

[0011] Preferably, the above-mentioned 3D human meridian acupoint template is constructed by the spatial positioning data.

[0012] Preferably, the above human body posture parameters are obtained by the following steps: S3.1. Calculate the rotation matrix of Euler angles , the rotation matrix It is obtained by combining three single-axis rotation matrices in a specific order. The rotation matrix From equation (1) to equation (4), we can get: ...Formula (1); ...Formula (2); ...Formula (3); ...Formula (4); in, , and They are three Euler angles respectively; S3.2, the human body posture parameters are composed of Euler angle rotation parameters and translation vectors, expressed by formula (5), and ; ...Formula (5); in, is the rotation parameter, and the rotation parameter The first three vectors in are the rotation matrix , is the translation vector, are human body posture parameters.

[0013] Preferably, the initial estimated position Obtained by the following steps: S4.1, obtain the intrinsic parameter matrix K used to describe the internal parameters of the camera, and the extrinsic parameter matrix E used to describe the position and direction of the camera relative to the world coordinate system, The internal parameter matrix K is calculated by formula (6), and the external parameter matrix E is calculated by formula (7). ...Formula (6); in, and is the focal length, is the optical center coordinate; ...Formula (7); S4.2, according to formula (8), the total projection matrix P is obtained, thereby mapping the three-dimensional world coordinates to the coordinates of the human body surface image to be processed; ...Formula (8); S4.3, according to the 3D acupuncture point X on the 3D human meridian acupuncture point template projected to the coordinates of the human body surface image to be processed, obtain the initial estimated position corresponding to the 2D image acupuncture point ,and , .

[0014] Preferably, the above S4.3 includes the following steps: S4.3.1. Representing the 3D acupuncture points on the 3D human meridian acupuncture point template as homogeneous coordinates ,and , where the homogeneous coordinates It is expressed by formula (9); ...Formula (9); Where x, y, and z are homogeneous coordinates The coordinate point of T Transpose the matrix; S4.3.2, transforming the 3D acupuncture points on the 3D human meridian acupuncture point template from the world coordinate system to the camera coordinate system through the external parameter matrix E, which is specifically represented by formula (10); ...Formula (10); in, is a 3D point in the camera coordinate system.

[0015] S4.3.3, using the internal reference matrix K to project the 3D acupuncture points on the 3D human meridian acupuncture point template onto the two-dimensional image plane, and normalize them, as specifically represented by formula (11); ...Formula (11); in, are homogeneous image coordinates, and , x h , y h and z h The homogeneous image coordinates are The coordinates of the points; S4.3.4, through the homogeneous image coordinates Normalize to get the initial estimated position , specifically expressed by formula (12), and the initial estimated position By the complete projection function Represents that the projection function From formula (13), we can get: ...Formula (12); ...Formula (13).

[0016] Preferably, the above-mentioned cascade regressor is provided with a multi-step cascade regressor, and each step regressor is provided with a feature extraction module, a GAT processing module of a graph attention network, and a decoder.

[0017] Preferably, the processing of each step of the regressor includes the following steps: S5.1, the feature extraction module in the current step regressor extracts from the feature map F The local features of the window are resampled to 7×7×256 and then convolved into two layers to obtain a 512-dimensional visual feature vector; at the same time, the relative position vector between each feature point in the feature map F is calculated , encoded as position features through a multi-layer perceptron , add the relative position vector and position feature to get the fusion feature ; When the current step regressor is the first step regressor, the feature map F is the visual feature map; when the current step regressor is a regressor other than the first step regressor, the feature map F is the fusion feature map embedded in the previous step regressor Visual feature map of S5.2. Fusion features obtained in S5.1 The estimated position is input into the GAT processing module in the current step regressor, and multiple steps are iterated through the decoder MLP network and activation function of the current step regressor, and finally the acupoint position correction vector is output. , get the updated acupoint coordinates, where the updated acupoint coordinates are ,in is the estimated position; when the current step regressor is the first step regressor and the GAT processing module is the first layer graph attention network, the estimated position is the initial estimated position ; When the current step regressor is a regressor other than the first step regressor, the estimated position is the updated acupoint coordinates of the last graph attention network; when the current step regressor is the first step regressor but the GAT processing module is not the first layer graph attention network, the estimated position is the updated acupoint coordinates of the last graph attention network.

[0018] Preferably, the above S6 specifically comprises: defining the acupoint coordinates updated for the last time in the cascade regressor as the final predicted acupoint coordinates, and performing validity judgment and smoothing processing on the final predicted acupoint coordinates to obtain the final acupoint coordinates; then superimposing the acupoint points corresponding to the final acupoint coordinates on the human body surface image of S1, and outputting the corresponding acupoint names and positioning information at the same time.

[0019] Preferably, the training method of the acupoint recognition model is trained through a two-stage strategy, specifically, in the first stage, only the multi-task CNN backbone network is trained; in the second stage, the multi-task CNN backbone network is frozen, and the cascade regressor is trained at the same time, and finally the trained acupoint recognition model is obtained.

[0020] Preferably, the acupoint database also includes clinical application data; the clinical application data includes acupoint acupuncture depth, acupoint angle, acupoint contraindication information, acupoint combination scheme, and precautions for use of acupoints for special populations.

[0021] The second object of the present invention is to avoid the shortcomings of the prior art and provide a human acupoint recognition system that can accurately recognize acupoints and has the advantages of strong real-time performance and strong adaptability.

[0022] The above-mentioned purpose of the present invention is achieved by the following technical measures: A human acupoint recognition system is provided, which adopts the human acupoint recognition method based on artificial intelligence image analysis.

[0023] The human body acupoint recognition system of the present invention is provided with: Multi-angle image acquisition module - collects human body surface images; Acupoint database storage module: stores the acupoint database; Preprocessing and labeling module: preprocessing the human body surface image and labeling the labeled acupoint information of the human body surface image in the acupoint database; Big data model learning module: training the acupoint recognition model according to the acupoint database; and identifying the human body surface image to be processed through the trained acupoint recognition model to obtain the acupoint recognition result; Interactive interface module - displays the acupoint recognition results and performs human-computer interaction.

[0024] The present invention provides a method and system for human acupoint recognition based on artificial intelligence image analysis, wherein the human acupoint recognition method is performed by the following steps: S1, collecting a human body surface image to be processed; S2, preprocessing the human body surface image obtained by S1; S3, extracting a visual feature map and human body posture parameters in the human body surface image preprocessed by S2 through a multi-task CNN backbone network in a trained acupoint recognition model; the trained acupoint recognition model is obtained by training the acupoint recognition model with an acupoint database, wherein the acupoint recognition model is composed of a multi-task CNN backbone network, a 3D human meridian acupoint template and a cascade regressor; S4, generating an initial estimated position of the acupoint on the human body surface image according to the 3D human meridian acupoint template in the trained acupoint recognition model and the human body posture parameters obtained by S3 ; S5, the initial estimated position obtained by S4 through the cascade regressor in the trained acupoint recognition model And the visual feature map obtained in S3 is processed to obtain the coordinates of the acupuncture points in the human body surface image; S6, the acupuncture point coordinates obtained in S5 are post-processed to obtain the acupuncture point recognition result. The present invention can accurately identify acupuncture points, and has the advantages of strong real-time performance and strong adaptability. The beneficial effects of the present invention: 1. The multi-task CNN backbone network and the cascade regressor architecture are adopted to realize the precise positioning and real-time recognition of acupuncture points; 2. The present invention also combines the position encoding mechanism of anatomical knowledge to improve the accuracy and robustness of recognition; 3. Through multi-scale feature extraction and iterative optimization strategies, it can adapt to different body shapes and posture changes; 4. It has real-time guidance function, which improves the efficiency and safety of acupuncture operation; 5. It has good scalability and practicality, and provides technical support for the standardization and modernization of acupuncture treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention is further described with reference to the accompanying drawings, but the contents in the accompanying drawings do not constitute any limitation to the present invention.

[0026] Figure 1 The present invention is a flowchart of a method for identifying human acupoints based on artificial intelligence image analysis.

[0027] Figure 2 This is a flow chart of the recognition method of the trained acupoint recognition model. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is further described in conjunction with the following embodiments.

[0029] Example 1 A method for identifying human acupoints based on artificial intelligence image analysis, such as Figure 1 As shown, proceed as follows: S1, collecting the human body surface image to be processed; S2, preprocessing the human body surface image obtained in S1; S3, extracting visual feature maps and human posture parameters from the human body surface image preprocessed by S2 through the multi-task CNN backbone network in the trained acupoint recognition model; the trained acupoint recognition model is obtained by training the acupoint recognition model with the acupoint database, wherein the acupoint recognition model is composed of a multi-task CNN backbone network, a 3D human meridian acupoint template and a cascade regressor; S4: Generate the initial estimated position of the acupoints on the human body surface image based on the 3D human meridian acupoint template in the trained acupoint recognition model and the human body posture parameters obtained in S3 ; S5, the initial estimated position of S4 obtained by the cascade regressor in the trained acupoint recognition model The visual feature map obtained by S3 is processed to obtain the coordinates of the acupuncture points in the human body surface image; S6. Post-process the acupoint coordinates obtained in S5 to obtain an acupoint recognition result.

[0030] It should be noted that the multi-task CNN backbone network of the present invention takes a 256×256 human body surface image as input, downsamples it to 64×64 through the initial encoder, and then processes it through 4 serially connected hourglass modules. Each hourglass module includes an encoder-decoder structure in which the feature map is reduced to 8×8 at the bottleneck layer, and is provided with two branch tasks of 3D posture parameter estimation and feature point heat map, and finally outputs posture parameters and a 64×64 visual feature map. Multi-task learning is a machine learning method that allows the model to learn multiple related tasks simultaneously during training, thereby improving the generalization ability and robustness of the model. Specifically, while the network of the present invention is learning the human body posture parameter estimation, it is also learning the acupoint recognition task, so that the global features extracted from the posture estimation task can be used to assist the acupoint recognition task, thereby improving the overall recognition accuracy.

[0031] The 3D human meridian acupoint template is a universal template. Through S4 processing, the posture parameters predicted by the multi-task CNN backbone network are projected onto the 3D human meridian acupoint template to obtain the initial estimated position. This 3D human meridian acupoint template projection utilizes the geometric information of the 3D model to improve the accuracy of initial positioning.

[0032] It should be noted that the 3D human meridian acupoint template of the present invention can be constructed from the spatial positioning data in the acupoint database. The 3D human meridian acupoint template can also be derived from a public anatomical database and a professionally annotated medical image data set. The 3D human meridian acupoint template is a statistical average of the distribution of real human anatomical structures, ensuring the accuracy and reliability of the model. This embodiment is explained by taking the construction of a 3D human meridian acupoint template using the spatial positioning data in the acupoint database as an example.

[0033] Then the acupoint database of this embodiment includes basic acupoint data, spatial positioning data and clinical application data. The clinical application data includes the acupuncture depth of the acupoint, the angle of the acupoint, the contraindication information of the acupoint, the acupoint combination scheme, and the precautions for the use of the corresponding acupoints for special groups (such as pregnant women and the elderly). The role of the clinical application data of the present invention is to output the clinical application data corresponding to the acupoint when outputting the acupoint recognition result, thereby enriching the acupoint information.

[0034] The process of obtaining basic acupoint data is as follows: A1. Collect multiple images of the human body surface; A2. Preprocess the human body surface image obtained in A1 and annotate the acupoint information. The annotated acupoint information includes the standard acupoint name (such as Chinese name, pinyin name, English name), the anatomical location description of the acupoint, the meridian system to which the acupoint belongs, the main function of the acupoint and the indication of the acupoint, so as to obtain basic acupoint data.

[0035] It should be noted that the preprocessing of the human body surface image obtained by A1 and the human body surface image obtained by S1 of the present invention is to adjust the brightness, adjust the contrast, correct the color balance, remove noise, unify the image size and resolution, align the posture, correct the posture, compensate for the illumination, and remove unqualified images. Among them, the brightness and contrast are adjusted to improve the image clarity, the color balance correction is performed to ensure the accurate restoration of the skin color, the Gaussian filter and other methods are used to remove noise, and at the same time, the image size and resolution are unified, the posture alignment and correction are performed, and the illumination compensation is used to eliminate the influence of uneven illumination, and the unqualified quality images such as blur, overexposure, underexposure, etc. are removed, and repeated or redundant data are removed.

[0036] When collecting human body surface images of A1 and S1, a surround-arranged high-resolution camera array (≥12 million pixels) can be used to capture subtle features and anatomical landmarks on the human body surface in all directions. The bracket of the camera array adopts an electric lifting design, which can flexibly adjust the height and shooting angle according to the body shape of the subject to ensure complete coverage of all parts of the body. The camera of the present invention is also equipped with intelligent exposure control and LED fill light array, which can provide uniform lighting in different light environments and highlight the surface features. The high-precision autofocus system cooperates with the anti-shake bracket to effectively eliminate shooting jitter and ensure image clarity. This multi-angle, high-resolution image acquisition solution provides a reliable data basis for subsequent acupoint identification.

[0037] In the acupuncture point information of A2, acupuncture experts with rich clinical experience divide the human body into major anatomical areas such as the head, trunk, and limbs. At the same time, they mark the main anatomical landmarks of the human body (such as bone joints, muscle directions, etc.), record the relative distance and angle relationship between the acupuncture points and the surrounding anatomical landmarks, and mark the recommended needle insertion depth and angle of the acupuncture points. At the same time, the marking of acupuncture point information can also be carried out for quality control. Specifically, different experts use a multi-person cross-validation method to repeatedly mark to ensure the accuracy of the marking, establish a standardized review process to ensure the quality of the marking, and continuously optimize and update the marking data based on clinical feedback.

[0038] The process of obtaining spatial positioning data is as follows: B1. Scan multiple human body 3D data; B2. The three-dimensional coordinate information of the acupuncture points, the positional relationship of the human anatomical landmarks of the acupuncture points, the relative distance between different acupuncture points, and the angle data between different acupuncture points are annotated on the three-dimensional data of the human body to obtain the spatial positioning data.

[0039] It should be noted that the three-dimensional data of the human body can be obtained through structured light, 3D scanning equipment, CT scanner, etc., and the labeling method of the three-dimensional data of the human body is a conventional technology in this field and will not be described here one by one.

[0040] The human body posture parameters of the present invention are obtained by the following steps: S3.1. Calculate the rotation matrix of Euler angles , the rotation matrix The rotation matrix is ​​obtained by combining three single-axis rotation matrices in a specific order. From equation (1) to equation (4), we can get: ...Formula (1); ...Formula (2); ...Formula (3); ...Formula (4); in, , and They are three Euler angles respectively; S3.2, the human body posture parameters are composed of Euler angle rotation parameters and translation vectors, which are expressed by formula (5), and ; ...Formula (5); in, is the rotation parameter, and the rotation parameter The first three vectors in are the rotation matrices , is the translation vector, are human body posture parameters.

[0041] The initial estimated position of the present invention Obtained by the following steps: S4.1. Obtain an intrinsic parameter matrix K for describing the internal parameters of the camera and an extrinsic parameter matrix E for describing the position and orientation of the camera relative to the world coordinate system. The intrinsic parameter matrix K is calculated by equation (6) and the extrinsic parameter matrix E is calculated by equation (7). The intrinsic parameter matrix K is used to describe the internal parameters of the camera, including focal length, optical center and pixel ratio. The intrinsic parameter matrix K is usually expressed as a 3×3 matrix: ...Formula (6); in, and is the focal length, is the optical center coordinate; ...Formula (7); S4.2. Obtain the total projection matrix P according to formula (8), thereby mapping the three-dimensional world coordinates to the coordinates of the human body surface image to be processed; wherein the dimension of the total projection matrix P is 3×4; ...Formula (8); S4.3, according to the 3D acupuncture point X on the 3D human meridian acupuncture point template projected to the coordinates of the human body surface image to be processed, obtain the initial estimated position corresponding to the 2D image acupuncture point ,and , .

[0042] Among them, S4.3 includes the following steps: S4.3.1. Representing the 3D acupuncture points on the 3D human meridian acupuncture point template as homogeneous coordinates ,and , where the homogeneous coordinates It is expressed by formula (9); ...Formula (9); Where x, y, and z are homogeneous coordinates The coordinate point of T Transpose the matrix; S4.3.2, transform the 3D acupuncture points on the 3D human meridian acupuncture point template from the world coordinate system to the camera coordinate system through the external parameter matrix E, which is specifically expressed by formula (10); ...Formula (10); in, is a three-dimensional point in the camera coordinate system; S4.3.3, using the internal reference matrix K, project the 3D acupuncture points on the 3D human meridian acupuncture point template onto the 2D image plane and normalize them, as specifically expressed by formula (11); ...Formula (11); in, are homogeneous image coordinates, and , x h , y h and z h The homogeneous image coordinates are The coordinates of the points; S4.3.4. By homogeneous image coordinates Normalize to get the initial estimated position , specifically expressed by formula (12), and the initial estimated position By the complete projection function Represents that the projection function From formula (13), we can get: ...Formula (12); ...Formula (13).

[0043] The cascade regressor is provided with a multi-step cascade regressor, and each step regressor is provided with a feature extraction module, a GAT processing module of a graph attention network, and a decoder.

[0044] It should be noted that the cascade regressor adopts a multi-step iterative optimization structure. The cascade regressor usually uses three steps to achieve the ideal effect. This embodiment takes a three-step cascade regressor as an example. The GAT processing module has a total of 4 layers, each of which includes operations such as query / key / value vector generation, attention weight calculation, feature aggregation, and residual connection. The multi-layer structure of the GAT processing module allows the model to observe globally first, and then gradually focus on the local information most relevant to the target. The initial layer pays more attention to global structural information, such as unobstructed areas, and subsequent layers focus on local precise adjustment. This strategy helps to capture more accurate features at different scales, as well as feature information at different levels. The role of the decoder is to calculate the relative position of each acupoint relative to other anatomical landmarks, generate a position vector containing its relative position and geometric features, and use the embedding layer to convert the position vector into a high-dimensional feature representation so that it can be combined with other features to integrate the spatial relationship information of the acupoint into the feature representation. Feature aggregation combines the local visual features (such as color, texture) of the acupoint and its relative position on the human body to form a complete feature representation. This can help the model understand the relative position relationship between acupoints and gradually improve the feature expression ability from coarse to fine by extracting features at different scales.

[0045] Among them, Figure 2 , the processing of each step of the regressor includes the following steps: S5.1, the feature extraction module in the current step regressor extracts from the feature map F The local features of the window are resampled to 7×7×256 and then convolved into two layers to obtain a 512-dimensional visual feature vector; at the same time, the relative position vector between each feature point in the feature map F is calculated , encoded as position features through a multi-layer perceptron , add the relative position vector and position feature to get the fusion feature ; When the current step regressor is the first step regressor, the feature map F is the visual feature map; when the current step regressor is a regressor other than the first step regressor, the feature map F is the fusion feature embedded in the previous step regressor Visual feature map of S5.2. Fusion features obtained in S5.1 The estimated position is input into the GAT processing module in the current step regressor, and multiple steps are iterated through the decoder MLP network and activation function of the current step regressor, and finally the acupoint position correction vector is output. , get the updated acupoint coordinates, where the updated acupoint coordinates are ,in is the estimated position; when the current step regressor is the first step regressor and the GAT processing module is the first layer graph attention network, the estimated position is the initial estimated position ; When the current step regressor is a regressor other than the first step regressor, the estimated position is the updated acupoint coordinates of the last graph attention network; when the current step regressor is the first step regressor but the GAT processing module is not the first layer graph attention network, the estimated position is the updated acupoint coordinates of the last graph attention network.

[0046] It should be noted that the activation function of the present invention is a conventional activation function in the art, such as Sigmoid function, ReLU function, Leaky ReLU function, Softmax function, etc. These activation functions can be used as the activation function of the present invention.

[0047] S6 is specifically as follows: defining the last updated acupoint coordinates in the cascade regressor as the final predicted acupoint coordinates, and performing validity judgment and smoothing on the final predicted acupoint coordinates to obtain the final acupoint coordinates; then superimposing the acupoint points corresponding to the final acupoint coordinates on the human body surface image of S1, and outputting the corresponding acupoint names and positioning information at the same time.

[0048] It should be noted that the validity judgment is to delete the wrong acupoint coordinates, such as acupoints that are close to each other, overlapping acupoints, etc.; and the coordinate smoothing process is a conventional operation in the art and is not the focus of the invention and will not be described in detail.

[0049] The training method of the acupoint recognition model is trained through a two-stage strategy. Specifically, in the first stage, only the multi-task CNN backbone network is trained; in the second stage, the multi-task CNN backbone network is frozen and the cascade regressor is trained at the same time, and finally the trained acupoint recognition model is obtained.

[0050] It should be noted that the first stage training termination condition of the present invention is reaching the number of training rounds, and the number of training rounds can be 500 rounds or other numbers. After the first stage training, the second stage training is carried out, and the second stage training termination condition is reaching the number of training rounds or convergence. The specific training termination condition can be determined according to the actual situation. The training method of the acupoint recognition model of the present invention uses an adaptive moment estimation (Adam, Adaptive MomentEstimation) optimizer, the initial learning rate is set to 0.001, and the cosine annealing learning rate scheduling strategy is adopted, which decays once every 100 rounds. The optimizer and The parameters are set to 0.9 and 0.999 respectively, and the weight decay coefficient is 0.0001.

[0051] The beneficial effects of the human acupoint recognition method based on artificial intelligence image analysis are as follows: 1. The multi-task CNN backbone network and cascade regressor architecture are adopted to achieve accurate positioning of acupoints. At the same time, the image acquisition and processing speed of the present invention reaches more than 30 frames per second, and the acupoint recognition response time is controlled within 100ms for real-time display and prompting without lag, thereby enabling real-time recognition; 2. The present invention also combines the position encoding mechanism of anatomical knowledge to improve the accuracy and robustness of recognition; 3. Through multi-scale feature extraction and iterative optimization strategies, it adapts to acupoint recognition and posture changes of different body shapes, ages, and genders; 4. It has a real-time guidance function to improve the efficiency and safety of acupuncture operations; 5. It has good scalability and practicality, and provides technical support for the standardization and modernization of acupuncture treatment.

[0052] Example 2 Application of the method for identifying human acupoints based on artificial intelligence image analysis in Example 1. Among them, the human body surface images of S1 and A1 are collected in real time by the camera of the patient or the saved image files are loaded. The human body surface image is collected using a high-definition camera with a resolution of not less than 1080p, equipped with a multi-angle adjustable bracket and an optional infrared camera. The acquisition parameters are set to a frame rate of 30-60fps, and the autofocus, autoexposure and white balance functions are enabled. The acquisition process performs 360-degree surround shooting of the target area, captures multiple images at key angles such as 0°, 45°, 90°, 135°, 180°, etc., and maintains an optimal shooting distance of 20-50cm. The present invention detects image clarity and lighting conditions in real time, automatically filters blurred or unqualified images, and ensures acquisition quality.

[0053] The human body surface image of this embodiment needs to be standardized preprocessed, which specifically includes adjusting the image size to the standard input size of 256×256 and RGB color space conversion, while removing noise through Gaussian filtering, enhancing contrast through histogram equalization, eliminating salt and pepper noise through median filtering, and performing image enhancement and illumination correction processing such as adaptive gamma correction, local contrast enhancement and shadow removal. For input images of different sizes, bilinear interpolation method is used for resampling, maintaining the aspect ratio of the image, and zero padding is performed on the edges to meet the input requirements of the network. Other processes are the same as in Example 1.

[0054] This embodiment has accurate recognition: training based on deep learning and graph attention network; position encoding mechanism combined with anatomical knowledge; multi-scale feature extraction and iterative optimization strategy Strong real-time performance: image acquisition and processing speed reaches more than 30 frames per second; acupoint recognition response time is controlled within 100ms; real-time display and prompts without lag Strong adaptability: supports acupoint recognition for different body shapes, ages, and genders; adapts to various lighting and shooting angle conditions; has robustness against occlusion and posture changes; supports personalized parameter adjustment Intelligent interaction: intuitive graphical operation interface; support for multiple interaction methods; switching between professional and simple modes; remote guidance and teaching functions.

[0055] Example 3 Application of the method for identifying human acupoints based on artificial intelligence image analysis in Example 2. This example is used to identify acupoints on the human hand. The operator first uses a camera to take multiple human surface images of the hand. The system processes the images and identifies acupoints such as "Hegu" and "Neiguan" through a big data model. When performing acupuncture, the camera will monitor the position of the hand in real time. When the needle tip approaches a certain acupoint, the system will prompt the name of the acupoint and its function for the acupuncturist's reference.

[0056] Example 4 Application of the human acupoint recognition method based on artificial intelligence image analysis of Example 2. This embodiment is used for the recognition of multiple parts of the body, and the system can take images of different parts such as the head, torso, limbs, etc. By combining the human body surface images of multiple parts, the present invention can recognize important acupoints such as "Baihui" and "Zusanli", and provide complete acupoint prompts during acupuncture through real-time monitoring of the human body surface images and acupuncture parts.

[0057] Example 5 A human acupoint recognition system adopts the human acupoint recognition method based on artificial intelligence image analysis as in embodiment 1 or 2 to perform human acupoint recognition.

[0058] Wherein, the human body acupoint recognition system of the present invention is provided with: Multi-angle image acquisition module - collects human body surface images; Acupoint database storage module: stores acupoint database; Preprocessing and annotation module: preprocess the human body surface image and annotate the annotated acupoint information of the human body surface image in the acupoint database; Big data model learning module: train the acupoint recognition model according to the acupoint database; at the same time, the trained acupoint recognition model recognizes the human body surface image to be processed and obtains the acupoint recognition result; Interactive interface module - displays acupoint recognition results and performs human-computer interaction.

[0059] The beneficial effects of the human acupoint recognition system are as follows: 1. The multi-task CNN backbone network and cascade regressor architecture are adopted to achieve accurate positioning of acupoints. At the same time, the image acquisition and processing speed of the present invention reaches more than 30 frames per second, and the acupoint recognition response time is controlled within 100ms for real-time display and prompting without lag, thereby enabling real-time recognition; 2. The present invention also combines the position encoding mechanism of anatomical knowledge to improve the accuracy and robustness of recognition; 3. Through multi-scale feature extraction and iterative optimization strategies, it adapts to acupoint recognition and posture changes of different body shapes, ages, and genders; 4. It has a real-time guidance function to improve the efficiency and safety of acupuncture operations; 5. It has good scalability and practicality, and provides technical support for the standardization and modernization of acupuncture treatment.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for identifying human acupoints based on artificial intelligence image analysis, characterized in that: Proceed as follows: S1, collecting the human body surface image to be processed; S2, preprocessing the human body surface image obtained in S1; S3, extracting visual feature maps and human posture parameters from the human body surface image pre-processed by S2 through the multi-task CNN backbone network in the trained acupoint recognition model; the trained acupoint recognition model is obtained by training the acupoint recognition model with the acupoint database, wherein the acupoint recognition model is composed of a multi-task CNN backbone network, a 3D human meridian acupoint template and a cascade regressor; S4: Generate the initial estimated position of the acupoints on the human body surface image based on the 3D human meridian acupoint template in the trained acupoint recognition model and the human body posture parameters obtained in S3 ; S5, the initial estimated position of S4 obtained by the cascade regressor in the trained acupoint recognition model The visual feature map obtained by S3 is processed to obtain the coordinates of the acupuncture points in the human body surface image; S6. Post-process the acupoint coordinates obtained in S5 to obtain an acupoint recognition result.

2. The method for identifying human acupoints based on artificial intelligence image analysis according to claim 1, characterized in that: The acupoint database includes basic acupoint data; The process of obtaining the basic acupoint data is as follows: A1. Collect multiple images of the human body surface; A2. Preprocess the human body surface image obtained in A1 and mark the acupoint information, wherein the marked acupoint information includes the standard name of the acupoint, the anatomical location description of the acupoint, the meridian system to which the acupoint belongs, the main function of the acupoint and the indication of the acupoint, so as to obtain the basic acupoint data.

3. The method for identifying human acupoints based on artificial intelligence image analysis according to claim 2, characterized in that: The acupoint database also includes spatial positioning data; The process of obtaining the spatial positioning data is as follows: B1. Scan multiple human body 3D data; B2. Annotating the three-dimensional coordinate information of acupoints, the positional relationship of the acupoint anatomical landmarks, the relative distance between different acupoints, and the angle data between different acupoints on the three-dimensional data of the human body to obtain the spatial positioning data; The 3D human meridian acupoint template is constructed based on the spatial positioning data.

4. The method for identifying human acupoints based on artificial intelligence image analysis according to claim 2 or 3, characterized in that: The human body posture parameters are obtained by the following steps: S3.

1. Calculate the rotation matrix of Euler angles , the rotation matrix It is obtained by combining three single-axis rotation matrices in a specific order. The rotation matrix From equation (1) to equation (4), we can get: ...Formula (1); ...Formula (2); ...Formula (3); ...Formula (4); in, , and They are three Euler angles respectively; S3.2, the human body posture parameters are composed of Euler angle rotation parameters and translation vectors, expressed by formula (5), and ; ...Formula (5); in, is the rotation parameter, and the rotation parameter The first three vectors in are the rotation matrix , is the translation vector, are human body posture parameters.

5. The method for human acupoint recognition based on artificial intelligence image analysis according to claim 4, characterized in that: The initial estimated position Obtained by the following steps: S4.

1. Obtain an intrinsic parameter matrix K for describing the internal parameters of the camera, and an extrinsic parameter matrix E for describing the position and orientation of the camera relative to the world coordinate system, wherein the intrinsic parameter matrix K is calculated by formula (6), and the extrinsic parameter matrix E is calculated by formula (7). ...Formula (6); in, and is the focal length, is the optical center coordinate; ...Formula (7); S4.2, according to formula (8), the total projection matrix P is obtained, thereby mapping the three-dimensional world coordinates to the coordinates of the human body surface image to be processed; ...Formula (8); S4.3, according to the 3D acupuncture point X on the 3D human meridian acupuncture point template projected to the coordinates of the human body surface image to be processed, obtain the initial estimated position corresponding to the 2D image acupuncture point ,and , .

6. The method for human acupoint recognition based on artificial intelligence image analysis according to claim 5, characterized in that: The S4.3 comprises the following steps: S4.3.

1. Representing the 3D acupuncture points on the 3D human meridian acupuncture point template as homogeneous coordinates ,and , where the homogeneous coordinates It is expressed by formula (9); ...Formula (9); Where x, y, and z are homogeneous coordinates The coordinate point of T Transpose the matrix; S4.3.2, transforming the 3D acupuncture points on the 3D human meridian acupuncture point template from the world coordinate system to the camera coordinate system through the external parameter matrix E, which is specifically represented by formula (10); ...Formula (10); in, is a three-dimensional point in the camera coordinate system; S4.3.3, using the internal reference matrix K to project the 3D acupuncture points on the 3D human meridian acupuncture point template onto the two-dimensional image plane, and normalize them, as specifically represented by formula (11); ...Formula (11); in, are homogeneous image coordinates, and , x h , y h and z h The homogeneous image coordinates are The coordinates of the points; S4.3.4, through the homogeneous image coordinates Normalize to get the initial estimated position , specifically expressed by formula (12), and the initial estimated position By the complete projection function Represents that the projection function From formula (13), we can get: ...Formula (12); ...Formula (13).

7. The method for identifying human acupoints based on artificial intelligence image analysis according to claim 2 or 3, characterized in that: The cascade regressor is provided with a multi-step cascade regressor, and each step regressor is provided with a feature extraction module, a GAT processing module with a multi-layer graph attention network, and a decoder; The processing of each regressor step includes the following steps: S5.1, the feature extraction module in the current step regressor extracts from the feature map F The local features of the window are resampled to 7×7×256 and then convolved into two layers to obtain a 512-dimensional visual feature vector; at the same time, the relative position vector between each feature point in the feature map F is calculated , encoded as position features through a multi-layer perceptron , add the relative position vector and position feature to get the fusion feature ; When the current step regressor is the first step regressor, the feature map F is the visual feature map; when the current step regressor is a regressor other than the first step regressor, the feature map F is the fusion feature map embedded in the previous step regressor Visual feature map of S5.

2. Fusion features obtained in S5.1 The estimated position is input into the GAT processing module in the current step regressor, and multiple steps are iterated through the decoder MLP network and activation function of the current step regressor, and finally the acupoint position correction vector is output. , get the updated acupoint coordinates, where the updated acupoint coordinates are ,in is the estimated position; when the current step regressor is the first step regressor and the GAT processing module is the first layer graph attention network, the estimated position is the initial estimated position ; When the current step regressor is a regressor other than the first step regressor, the estimated position is the updated acupoint coordinates of the last graph attention network; when the current step regressor is the first step regressor but the GAT processing module is not the first layer graph attention network, the estimated position is the updated acupoint coordinates of the last graph attention network.

8. The method for identifying human acupoints based on artificial intelligence image analysis according to claim 2 or 3, characterized in that: The S6 specifically comprises: defining the last updated acupoint coordinates in the cascade regressor as the final predicted acupoint coordinates, and performing validity judgment and smoothing processing on the final predicted acupoint coordinates to obtain the final acupoint coordinates; then superimposing and displaying the acupoint points corresponding to the final acupoint coordinates on the human body surface image of S1, and outputting the corresponding acupoint names and positioning information at the same time; The training method of the acupoint recognition model is trained through a two-stage strategy, specifically, in the first stage, only the multi-task CNN backbone network is trained; in the second stage, the multi-task CNN backbone network is frozen, and the cascade regressor is trained at the same time, and finally the trained acupoint recognition model is obtained; The acupoint database also includes clinical application data; the clinical application data includes acupoint acupuncture depth, acupoint angle, acupoint contraindication information, acupoint combination scheme, and precautions for the use of acupoints for special populations.

9. A human acupoint recognition system, characterized in that: Human acupoint recognition is performed using the human acupoint recognition method based on artificial intelligence image analysis as described in any one of claims 1 to 8.

10. The human acupoint recognition system according to claim 9, characterized in that: The settings are: Multi-angle image acquisition module - collects human body surface images; Acupoint database storage module: stores the acupoint database; Preprocessing and labeling module: preprocessing the human body surface image and labeling the labeled acupoint information of the human body surface image in the acupoint database; Big data model learning module: training the acupoint recognition model according to the acupoint database; and identifying the human body surface image to be processed through the trained acupoint recognition model to obtain the acupoint recognition result; Interactive interface module - displays the acupoint recognition results and performs human-computer interaction.

Images