Head acupuncture semi-supervised intelligent acupoint selection method and system based on visual positioning

Abstract

The invention discloses a head acupuncture semi-supervised intelligent acupoint selection method and system based on visual positioning, and belongs to the technical field of intelligent auxiliary medical treatment. The method comprises the steps of collecting a face image and extracting face standard two-dimensional feature points; constructing a first training data set by using the three-dimensional space coordinates of the face standard two-dimensional feature points and the three-dimensional space coordinates of the head acupuncture point feature points, and training a multi-layer perceptron mapping network; training a Yolo two-dimensional mapping network by using a second training data set constructed by the virtual face image and the two-dimensional coordinates of the head acupoint feature points; and inputting an unmarked face image into the trained multilayer perceptron mapping network and the trained Yolo network, and carrying out weighted fusion on two-dimensional coordinates output by the two networks to complete head acupuncture intelligent auxiliary acupoint selection. According to the invention, semi-supervised accurate recognition and positioning of face acupuncture points are realized, so that the accuracy and safety of self-service acupuncture treatment of a patient are improved on the premise of depending on a small amount of labeled data.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent assisted medical technology, and specifically relates to a semi-supervised intelligent acupoint selection method and system for head acupuncture based on visual positioning. Background Art

[0002] Traditional acupuncture treatment requires going to the hospital for registration and treatment, which takes a lot of time for patients. Therefore, self-help assisted acupuncture is a method with strong clinical application value. However, for head acupuncture, due to the dense facial acupuncture points and the complex surrounding tissue structure, the accuracy and safety of acupuncture points are extremely high, and the acupuncture points located on the scalp are more difficult to obtain, which brings difficulties to self-help assisted acupuncture points for head acupuncture.

[0003] Existing acupuncture point selection auxiliary tools and methods, such as acupuncture atlases and body surface landmark positioning methods, have certain subjectivity and uncertainty. For example, although acupuncture atlases provide the approximate location of acupoints, due to individual differences in patients and changes in facial morphology, the actual acupoint selection still needs to rely on the doctor's experience and touch for fine-tuning; the body surface landmark positioning method is easily affected by factors such as changes in facial expressions and skin sagging, resulting in inaccurate acupoint selection. On the other hand, existing artificial intelligence diagnosis and treatment technologies often rely on a large amount of labeled data, but the human resources of doctors with professional knowledge are limited, making it difficult to provide a sufficient number of labeled samples in the traditional sense.

[0004] Therefore, it is particularly important to develop a method that can accurately and quickly locate facial acupuncture points by relying on a small amount of annotated data. With the rapid development of computer vision and deep learning technology, these technologies provide new solutions for acupuncture point selection, which can achieve semi-supervised accurate recognition and positioning of facial acupuncture points, thereby improving the accuracy and safety of self-help acupuncture treatment for patients while relying on a small amount of annotated data. Summary of the invention

[0005] In order to solve the problem of self-help assisted acupuncture point positioning on the head, the present invention proposes a semi-supervised intelligent acupuncture point positioning method and system for the head based on visual positioning, which uses neural network and computer vision technology to perform intelligent assisted acupuncture point positioning through three-dimensional coordinate inference and two-dimensional coordinate mapping of acupuncture points, supporting intelligent assisted treatment methods such as acupuncture treatment for retinal optic nerve damage.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] The semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning includes the following steps:

[0008] Step S1, in the training stage of the present invention, firstly, the Dlib face standard feature point recognition library is used to identify 68 standard facial feature points through the input face image . Use the 3DMM model to model the three-dimensional spatial coordinates of standard facial feature points , as shown below:

[0009] ,

[0010] in, is the average position of the standard two-dimensional feature points on the face, For the human face shape base, is the facial shape feature coefficient, For human facial expression base, is the facial expression feature coefficient, m and n represent the total number of facial shape bases and facial expression bases respectively.

[0011] Construct the reprojection error objective function, the objective function is as follows:

[0012] ,

[0013] In the formula, is the standard projection matrix, is the face pose and rotation matrix relative to the camera, is the translation vector of the face relative to the camera posture. By minimizing the above equation, we can solve , , , , thus obtaining the face pose relative to the camera, as well as the 3D coordinates of the standard feature points of the face and the characteristic coefficients of the patient's 3DMM model. Then, using the obtained 3DMM characteristic coefficients, the 3D point cloud of the face is obtained.

[0014] Step S2: A professional doctor selects acupuncture point feature points on the 3D point cloud of the patient's head obtained in step S1. ,in is the number of acupuncture points. Construct the first training data set for the element, and train the multilayer perceptron mapping network with the following formula as the objective function ;

[0015] ,

[0016] Trained multilayer perceptron mapping network Input the 3D coordinates of the standard facial feature points and output the 3D coordinates of the acupuncture points.

[0017] Step S3, input the three-dimensional point cloud of the patient's head and the three-dimensional coordinates of acupuncture points obtained in step S1 into the UnrealEngine virtual engine, evenly set multiple virtual cameras in the common acquisition position area around the patient's head point cloud, and the virtual engine automatically generates a virtual face image and the corresponding two-dimensional coordinates of acupuncture points, thereby using a small amount of professional doctor annotation data to generate an expanded second training set. The Yolo structure network is trained with the expanded second training set so that the input face image outputs the two-dimensional coordinates of acupuncture points.

[0018] Step S4, in the inference test phase, for general patients who are not labeled. Use a camera to collect the patient's facial image, use the Dlib library to identify 68 standard feature points similar to step S1, and use the 3DMM model to obtain the three-dimensional coordinates of the patient's 68 standard feature points and the camera pose. Then use the two-branch inference network to infer the two-dimensional coordinates of the patient's acupuncture points. Branch 1: Input the three-dimensional coordinates of the 68 standard feature points into the multi-layer perceptron trained in step S2 Generate the 3D coordinates of acupuncture feature points, and then use the camera pose information to calculate the 2D coordinates of the supervised branch acupuncture points: ; Branch 2: Input the patient image into the trained Yolo structure network branch of step S3, and output the two-dimensional reference coordinates of the semi-supervised branch acupuncture points: Finally, the reference coordinates of the acupuncture points generated by the two branches are weighted according to the following formula to generate the final two-dimensional reasoning coordinates of the acupuncture points.

[0019] ,

[0020] in, , is a hyperparameter, determined through parameter adjustment experiments. The final two-dimensional coordinates of the acupuncture points.

[0021] On the other hand, the present invention provides a semi-supervised intelligent acupoint selection system for head acupuncture based on visual positioning, the system comprising:

[0022] 3DMM model solving module, used to collect face images and extract standard two-dimensional facial feature points, and use the 3DMM model to solve the camera posture of the face image, the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional point cloud of the face;

[0023] A multi-layer perceptron mapping network training module is used to select the positions of acupuncture point feature points on the three-dimensional point cloud of the face, construct a first training data set using the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional spatial coordinates of the acupuncture point feature points, and train a multi-layer perceptron mapping network from the three-dimensional spatial coordinates of the standard two-dimensional facial feature points to the three-dimensional spatial coordinates of the acupuncture point feature points;

[0024] A two-dimensional mapping network training module is used to input the three-dimensional spatial coordinates of the face three-dimensional point cloud and the standard two-dimensional facial feature points into a virtual engine, and evenly set multiple virtual cameras around the face three-dimensional point cloud to generate virtual face images at different distances and postures and the corresponding two-dimensional coordinates of the head acupuncture point feature points, and generate a second training data set constructed with the virtual face images and the two-dimensional coordinates of the head acupuncture point feature points, wherein the second training data set is used to train a two-dimensional mapping network from face images to acupuncture point feature points with the Yolo network as the basic structure;

[0025] The intelligent assisted acupoint selection module is used to input the unlabeled face image captured by the camera into the trained multi-layer perceptron mapping network to infer the three-dimensional spatial coordinates of the acupuncture point feature points, use the inferred three-dimensional spatial coordinates of the acupuncture point feature points and the camera posture projection to calculate the coordinates of the acupuncture point feature points on the two-dimensional image, input the unlabeled face image into the trained Yolo network to generate the two-dimensional reference coordinates of the acupuncture point feature points; weighted fusion of the coordinates on the two-dimensional image and the two-dimensional reference coordinates to complete the intelligent assisted acupoint selection of head acupuncture.

[0026] In a third aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning.

[0027] In a fourth aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the aforementioned semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning.

[0028] The beneficial effects of the present invention are:

[0029] The present invention can be widely used in the field of autonomous acupuncture treatment, providing patients with reliable self-help acupoint positioning support, improving the convenience of acupuncture treatment, and saving patients the time cost of frequent medical treatment. The semi-supervised method using the virtual engine of the present invention can effectively reduce the workload of doctors in annotating data, which is conducive to saving doctors' manpower costs and improving the application speed of the system in different types of diseases. In addition, the method proposed in the present invention can locate acupoints in the visible area of ​​the face, as well as invisible acupoints under the scalp, expanding the scope of application of this acupuncture-assisted positioning method. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the training phase of the semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning of the present invention;

[0031] Figure 2 This is a flow chart of the reasoning and application phase of the semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] A semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning is used to assist patients in self-help acupuncture treatment. The method is divided into a training stage and an inference application stage.

[0034] like Figure 1 As shown, in the training stage, multiple patient face images are collected, and 68 standard feature points are extracted from the two-dimensional images. The 3DMM model is used to solve the camera posture of the patient image, the three-dimensional spatial coordinates of the 68 standard facial feature points and the three-dimensional point cloud of the patient's head; the doctor marks the positions of the acupoints on the generated three-dimensional point cloud of the patient's head, and the 68 standard facial feature points and acupuncture points of each patient form the first training data set; the data of all the labeled patients are used to train the first supervised acupoint two-dimensional projection inference branch network: the first branch network is a multilayer perceptron neural network that maps the three-dimensional coordinates of the acupoints to the three-dimensional coordinates of the acupuncture points In addition, the 3D point cloud of the patient's head is input into the Unreal Engine virtual engine, and multiple virtual camera positions are set around the 3D head model to generate virtual face images and 2D image point coordinates of acupoints under each virtual camera. The virtual face images and 2D point coordinates of acupoints constitute a second extended training data set, and the second extended training data set is used to train the Yolo network, which inputs real or virtual face images and outputs semi-supervised 2D mapping of acupoints.

[0035] like Figure 2 As shown in the figure, in the inference application stage, the facial image of the unlabeled patient is collected through the camera and the 68 standard feature points of the face are detected. The first inference branch uses the 3DMM model to solve the three-dimensional spatial coordinates of the camera pose of the patient's collected image and the 68 standard feature points of the face; the acupoint three-dimensional mapping multi-layer perceptron trained in the training stage is used The three-dimensional coordinates of acupuncture points of unlabeled patients are inferred; the supervised acupuncture point two-dimensional projection is calculated using the inferred three-dimensional coordinates of acupuncture points and the patient image camera pose projection. The second reasoning branch is the semi-supervised acupuncture point two-dimensional mapping branch, which inputs the patient's real image and uses the Yolo network trained in the training phase to generate the semi-supervised acupuncture point two-dimensional projection. Finally, the prediction results of the two branches are weighted and fused to complete the intelligent assisted acupuncture point selection on the head and obtain the acupuncture point location coordinates.

[0036] The method specifically comprises the following steps:

[0037] Step S1, in the training stage of the present invention, firstly, the Dlib face standard feature point recognition library is used to identify 68 standard facial feature points through the input patient image . Use the 3DMM model to model the three-dimensional space coordinates of standard feature points , as shown below:

[0038] ,

[0039] in, is the average position of the standard two-dimensional feature points on the face, For the human face shape base, is the facial shape feature coefficient, For human facial expression base, is the facial expression feature coefficient, m and n represent the total number of facial shape bases and facial expression bases respectively.

[0040] Construct the reprojection error objective function, the objective function is as follows:

[0041] ,

[0042] In the formula, is the standard projection matrix, is the rotation matrix of the face relative to the camera, is the translation vector of the face relative to the camera posture. By minimizing the above equation, we can solve , , , , thus obtaining the face posture relative to the camera, as well as the three-dimensional coordinates of the standard feature points of the face and the characteristic coefficients of the patient's 3DMM model. Then, using the obtained 3DMM characteristic coefficients, the dense three-dimensional point cloud of the patient's head is obtained.

[0043] Step S2: A professional doctor selects acupuncture point feature points on the 3D point cloud of the patient's head obtained in step S1. ,in is the number of acupuncture points. Construct the first training data set for the element, and train the multilayer perceptron mapping network with the following formula as the objective function ;

[0044] ,

[0045] The trained perceptron Input the 3D coordinates of standard facial feature points and output the 3D coordinates of acupuncture points.

[0046] Step S3, input the three-dimensional point cloud of the patient's head and the three-dimensional coordinates of acupuncture points obtained in step S1 into the UnrealEngine virtual engine, evenly set multiple virtual cameras in the common acquisition position area around the patient's head point cloud, and the virtual engine automatically generates a virtual face image and the corresponding two-dimensional coordinates of acupuncture points, thereby using a small amount of professional doctor annotation data to generate an expanded second training set. The Yolo structure network is trained with the expanded second training set so that the input face image outputs the two-dimensional coordinates of acupuncture points.

[0047] Step S4, in the inference test phase, for general patients who are not labeled. Use a camera to collect the patient's facial image, use the Dlib library to identify 68 standard feature points similar to step S1, and use the 3DMM model to obtain the three-dimensional coordinates of the patient's 68 standard feature points and the camera pose. Then use the two-branch inference network to infer the two-dimensional coordinates of the patient's acupuncture points. Branch 1: Input the three-dimensional coordinates of the 68 standard feature points into the multi-layer perceptron trained in step S2 Generate the 3D coordinates of acupuncture feature points, and then use the camera pose information to calculate the 2D coordinates of the supervised branch acupuncture points: ; Branch 2: Input the patient image into the trained Yolo structure network branch of step S3, and output the two-dimensional reference coordinates of the acupuncture points of the semi-supervised branch: Finally, the reference coordinates of the acupuncture points generated by the two branches are weighted according to the following formula to generate the final two-dimensional reasoning coordinates of the acupuncture points.

[0048] ,

[0049] in, , is a hyperparameter, determined through parameter adjustment experiments. The final two-dimensional coordinates of the acupuncture points.

[0050] On the other hand, the present invention provides a semi-supervised intelligent acupoint selection system for head acupuncture based on visual positioning, and each module included in the system can implement each step of the aforementioned method. Specifically, the system includes:

[0051] 3DMM model solving module, used to collect face images and extract standard two-dimensional facial feature points, and use the 3DMM model to solve the camera posture of the face image, the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional point cloud of the face;

[0052] A multi-layer perceptron mapping network training module is used to select the positions of acupuncture point feature points on the three-dimensional point cloud of the face, construct a first training data set using the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional spatial coordinates of the acupuncture point feature points, and train a multi-layer perceptron mapping network from the three-dimensional spatial coordinates of the standard two-dimensional facial feature points to the three-dimensional spatial coordinates of the acupuncture point feature points;

[0053] A two-dimensional mapping network training module is used to input the three-dimensional spatial coordinates of the face three-dimensional point cloud and the standard two-dimensional facial feature points into a virtual engine, and evenly set multiple virtual cameras around the face three-dimensional point cloud to generate virtual face images at different distances and postures and the corresponding two-dimensional coordinates of the head acupuncture point feature points, and generate a second training data set constructed with the virtual face images and the two-dimensional coordinates of the head acupuncture point feature points, wherein the second training data set is used to train a two-dimensional mapping network from face images to acupuncture point feature points with the Yolo network as the basic structure;

[0054] The intelligent assisted acupoint selection module is used to input the unlabeled face image captured by the camera into the trained multi-layer perceptron mapping network to infer the three-dimensional spatial coordinates of the acupuncture point feature points, use the inferred three-dimensional spatial coordinates of the acupuncture point feature points and the camera posture projection to calculate the coordinates of the acupuncture point feature points on the two-dimensional image, input the unlabeled face image into the trained Yolo network to generate the two-dimensional reference coordinates of the acupuncture point feature points; weighted fusion of the coordinates on the two-dimensional image and the two-dimensional reference coordinates to complete the intelligent assisted acupoint selection of head acupuncture.

[0055] In a third aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the aforementioned semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning.

[0057] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable EPROM, an electrically erasable programmable EEPROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are executed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0058] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular sequence or order.

[0059] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0060] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning, characterized in that: The method comprises the following steps: Step S1, collecting a face image and extracting standard two-dimensional facial feature points, using a 3DMM model to solve the camera pose of the face image, the three-dimensional spatial coordinates of the standard two-dimensional facial feature points, and the three-dimensional point cloud of the face; Step S2, selecting the positions of acupuncture point feature points on the three-dimensional point cloud of the face, constructing a first training data set using the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional spatial coordinates of the acupuncture point feature points, and training a multi-layer perceptron mapping network from the three-dimensional spatial coordinates of the standard two-dimensional facial feature points to the three-dimensional spatial coordinates of the acupuncture point feature points; Step S3, inputting the three-dimensional spatial coordinates of the face three-dimensional point cloud and the standard two-dimensional facial feature points into a virtual engine, and evenly arranging multiple virtual cameras around the face three-dimensional point cloud, generating virtual face images at different distances and postures and corresponding two-dimensional coordinates of head acupuncture point feature points, generating a second training data set constructed with the virtual face images and the two-dimensional coordinates of the head acupuncture point feature points, and the second training data set is used to train a two-dimensional mapping network from face images to acupuncture point feature points with Yolo network as the basic structure; Step S4, input the unlabeled face image captured by the camera into the trained multi-layer perceptron mapping network to infer the three-dimensional spatial coordinates of the acupuncture point feature points, use the inferred three-dimensional spatial coordinates of the acupuncture point feature points and the camera posture projection to calculate the coordinates of the acupuncture point feature points on the two-dimensional image, input the unlabeled face image into the trained Yolo network to generate the two-dimensional reference coordinates of the acupuncture point feature points; weighted fusion of the coordinates on the two-dimensional image and the two-dimensional reference coordinates to complete the intelligent assisted acupuncture point selection on the head.

2. According to the method of semi-supervised intelligent acupoint selection for head acupuncture based on visual positioning according to claim 1, it is characterized in that: The step S1 includes identifying 68 standard two-dimensional facial feature points on the collected face image. , the 3DMM model is used to solve the 3D spatial coordinates of 68 standard 2D facial feature points : ， in, is the average position of the standard two-dimensional feature points on the face, For the human face shape base, is the facial shape feature coefficient, For human facial expression base, is the facial expression feature coefficient, m and n represent the total number of facial shape bases and facial expression bases respectively; Construct the reprojection error objective function: ， In the formula, is the standard projection matrix, is the face pose and rotation matrix relative to the camera, is the translation vector of the face relative to the camera pose; Minimize the reprojection error objective function and solve , , , , obtain the face posture relative to the camera, the three-dimensional spatial coordinates of the standard two-dimensional facial feature points, and the 3DMM model feature coefficients; The obtained 3DMM feature coefficients are used to obtain the three-dimensional point cloud of the face.

3. According to the method of semi-supervised intelligent acupoint selection for head acupuncture based on visual positioning according to claim 1, it is characterized in that: The step S2 includes selecting acupuncture point feature points on the three-dimensional point cloud of the face. ,in is the number of selected acupuncture point feature points, Construct the first training data set for the elements, As the objective function, train a multilayer perceptron mapping network ; 。 4. The method for semi-supervised intelligent acupuncture point selection for head based on visual positioning according to claim 1, characterized in that: In step S4, the coordinates of the acupuncture point feature points on the two-dimensional image are calculated using the inferred three-dimensional spatial coordinates of the acupuncture point feature points and the camera posture projection: ; The two-dimensional reference coordinates of the characteristic points of acupuncture points are: , the weighted fusion of the coordinates on the two-dimensional image and the two-dimensional reference coordinates to generate the final two-dimensional reasoning coordinates of the acupuncture points include: ， in, , is a hyperparameter, determined through parameter adjustment experiments. The final two-dimensional coordinates of the acupuncture points.

5. A semi-supervised intelligent acupuncture point selection system for head based on visual positioning, characterized in that: The system comprises: 3DMM model solving module, used to collect face images and extract standard two-dimensional facial feature points, and use the 3DMM model to solve the camera posture of the face image, the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional point cloud of the face; A multi-layer perceptron mapping network training module is used to select the positions of acupuncture point feature points on the three-dimensional point cloud of the face, construct a first training data set using the three-dimensional spatial coordinates of the standard two-dimensional facial feature points and the three-dimensional spatial coordinates of the acupuncture point feature points, and train a multi-layer perceptron mapping network from the three-dimensional spatial coordinates of the standard two-dimensional facial feature points to the three-dimensional spatial coordinates of the acupuncture point feature points; A two-dimensional mapping network training module is used to input the three-dimensional spatial coordinates of the face three-dimensional point cloud and the standard two-dimensional facial feature points into a virtual engine, and evenly set multiple virtual cameras around the face three-dimensional point cloud to generate virtual face images at different distances and postures and the corresponding two-dimensional coordinates of the head acupuncture point feature points, and generate a second training data set constructed with the virtual face images and the two-dimensional coordinates of the head acupuncture point feature points, wherein the second training data set is used to train a two-dimensional mapping network from face images to acupuncture point feature points with the Yolo network as the basic structure; The intelligent assisted acupoint selection module is used to input the unlabeled face image captured by the camera into the trained multi-layer perceptron mapping network to infer the three-dimensional spatial coordinates of the acupuncture point feature points, use the inferred three-dimensional spatial coordinates of the acupuncture point feature points and the camera posture projection to calculate the coordinates of the acupuncture point feature points on the two-dimensional image, input the unlabeled face image into the trained Yolo network to generate the two-dimensional reference coordinates of the acupuncture point feature points; weighted fusion of the coordinates on the two-dimensional image and the two-dimensional reference coordinates to complete the intelligent assisted acupoint selection of head acupuncture.

6. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; Wherein, when one or more programs are executed by the one or more processors, the one or more processors implement the semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, which, when executed by a processor, enable the processor to implement a semi-supervised intelligent acupoint selection method for head acupuncture based on visual positioning as described in any one of claims 1-4.

Images