Human acupoint positioning analysis system and method based on image recognition
By using an acupoint location system based on image recognition and impedance measurement, combined with neural networks and multimodal verification, the problem of reliance on physician experience in traditional acupuncture acupoint location has been solved, achieving precise and objective acupoint location and improving operational accuracy and safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG UNIV OF TRADITIONAL CHINESE MEDICINE
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-19
AI Technical Summary
Traditional acupuncture point location relies on the physician's experience, which is highly subjective and makes it difficult to achieve precision and objectivity, thus affecting the accuracy and safety of the procedure.
An image recognition-based acupoint location analysis system is adopted, which combines a neural network model, impedance meter and multimodal verification. Acupoints are identified through image processing and impedance difference. The ORB feature detector and federated learning model are used to train the model to achieve accurate acupoint location.
It reduces reliance on physician experience, improves the accuracy and applicability of acupoint location, meets the needs of intelligent acupuncture-assisted operation, and reduces location time and misjudgment rate.
Smart Images
Figure CN122229672A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition technology, specifically to a human acupoint location analysis system and method based on image recognition. Background Technology
[0002] Acupuncture is an important component of traditional Chinese medicine. Through thousands of years of practice and inheritance, it has formed a complete theoretical and operational system, possessing unique value and profound significance in disease treatment, rehabilitation, and health maintenance. Traditional acupuncture point location mainly relies on anatomical landmarks and body measurements, heavily depending on the physician's clinical experience and subjective judgment. Furthermore, individual differences in body shape further complicate manual location, affecting not only the accuracy of acupuncture procedures but also potentially impacting treatment efficacy and safety. Therefore, there is an urgent need for an objective and precise acupuncture point location assistance system to reduce reliance on physician experience and meet the demands of intelligent acupuncture-assisted operations. Summary of the Invention
[0003] The purpose of this invention is to provide a human acupoint location analysis system and method based on image recognition, so as to solve the problems raised in the prior art.
[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The image recognition-based method for locating and analyzing acupoints on the human body includes the following steps: The process involves acquiring patient body images, preprocessing the images, and extracting acupoint locations. A feature detector is used to extract feature points from the patient body images, dividing the images into pixel windows, and establishing a neural network model for acupoint localization. The neural network model is then trained based on the patient body images. This training is performed to obtain a more accurate first confidence score. Lasers, motors, cameras, and projectors were deployed in the acupuncture department. The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images. Feature point matching and registration are performed to obtain the target image corresponding to the target object. The target image is divided into pixel windows, and the first confidence score of each pixel window as an acupoint is obtained based on the trained neural network model. Impedance values at different pixel window locations were collected using an impedance meter. Based on the impedance values, the second confidence level of each pixel window as an acupoint was obtained. The skin resistance at acupoints is lower and conductivity is higher, while the skin resistance in non-acupoint areas is higher and conductivity is lower. Impedance analyzers utilize this impedance difference to assist in acupoint identification. They are devices for detecting the electrical impedance of biological tissues. By inputting a weak and safe detection current into the body surface, they measure the impedance values at different locations. Utilizing the significant impedance difference between acupoint and non-acupoint areas, they assist in the location and verification of acupoints, helping to reduce location time, improve accuracy, and reduce false positives through multimodal verification, making them suitable for diverse clinical scenarios. Pixel windows are merged based on the first and second confidence levels to aggregate the target acupoint areas; A laser emits a laser point, a camera captures the relative position of the target acupoint area and the laser point, and a drive motor adjusts the position of the laser point to align the laser point with the target acupoint.
[0005] Preferably, the neural network model training includes: The ORB feature detector is used to extract feature points from patient body images. Feature points are stable and prominent points on the patient's body surface. The patient body images contain multiple images taken in different poses. The spatial distance between acupoints and feature points in each patient body image is calculated, and key feature points corresponding to acupoints are screened and determined. Set the pixel window size and window sliding step size, divide the patient's body image into several pixel windows, take the pixel window corresponding to the acupoint location area as the acupoint window, and obtain the window vector pointing to each pixel window from the key feature points. The neural network model is trained based on each window vector. The trained neural network model is used to determine the reliability of the pixel window pointed to by the input window vector as an acupoint.
[0006] Preferably, the key feature points corresponding to acupoints are screened and identified, including: The positions of the same feature point in several patient body images and the corresponding acupoints in each image are obtained respectively. The spatial distance between the feature point and the acupoint is calculated, and the variance between the spatial distances is obtained. If the variance is less than the preset variance threshold, the feature point is regarded as the key feature point of the acupoint.
[0007] Human acupoints usually have a fixed relative positional relationship with specific parts of the body surface. The location of acupoints needs to rely on the part as a reference. The location of the part is an important reference for locating acupoints, which is the key feature point in this solution. Therefore, this solution needs to first determine the key feature points related to acupoint location from the image, and then realize acupoint location based on the relative positional relationship between the key feature points and acupoints.
[0008] Preferably, training the neural network model includes: Establish a federated network model, treating each patient as a source domain client, and obtain the local models obtained by parallel training of each source domain client in the federated network model, as well as the global model obtained by aggregating each local model; Federated network models are characterized by data remaining local and distributed collaborative training. By training models through federated learning, the direct transmission and centralized aggregation of raw data are avoided, which can effectively protect user data privacy and data security while ensuring training effectiveness.
[0009] Obtain the locations of all acupoint windows and key feature points in the patient's body image, and obtain the window vector pointing to each pixel window from the key feature point. If the window pointed to by the window vector is an acupoint window, set the target value of the window vector to 1; if the window pointed to by the window vector is not an acupoint window, set the target value of the window vector to 0. Using each window vector as input and the target value of each window vector as output, a training set and a validation set are set, and local model training is performed through iterative optimization using the cross-entropy loss function. Based on each trained local model, the models are summarized and merged to form a new global model. The trained federated network model is used to obtain the target value of the window vector based on the input window vector, and the output target value is used as the reliability of the corresponding pixel window as the acupoint.
[0010] Preferably, obtaining the target image corresponding to the target object includes: The ORB feature detector extracts feature points from the surface of the target object in each projected image. Based on feature point matching, the spatial alignment of multiple projected images is completed. Then, the Poisson fusion algorithm is used to perform pixel-level fusion processing on the aligned images to achieve seamless image stitching and generate the target image corresponding to the target object.
[0011] Preferably, the first confidence level of each pixel window as an acupoint is obtained, including: Extract each pixel window and key feature point from the target image to obtain the window vector pointing to each pixel window. Input each window vector into the trained neural network model to obtain the reliability of the corresponding pixel window as an acupoint, which is used as the first confidence level of the pixel window as an acupoint.
[0012] Preferably, the second confidence level for each pixel window as an acupoint is obtained, including: An impedance meter is pre-deployed on the target object. The impedance meter is used to collect the impedance values of each pixel window corresponding to the body surface location. The impedance values are normalized and converted into a second confidence level for each pixel window as an acupoint.
[0013] Preferably, pixel windows are merged to converge the target acupoint area, including: The weights of the first confidence level and the second confidence level are preset. The total confidence level of each pixel window is calculated according to the weights, and the pixel windows with a total confidence level greater than the preset confidence level threshold are marked. Traverse each pixel window. If the number of marked windows in the eight neighboring windows of a pixel window is greater than a preset threshold, then merge the pixel window and its eight neighboring windows into the target acupoint region, and thus obtain the final target acupoint region.
[0014] Preferably, aligning the laser point with the target acupoint includes: A laser point is emitted by a laser to form a mark on the human body surface. The camera captures the position of the laser point in real time, extracts the center coordinates of the laser point, and calculates the offset between the laser point and the target acupoint area. Based on the PID control algorithm, the drive motor adjusts the position of the laser point. With the offset as input, the motor step size is adjusted through proportional-integral-derivative to align the laser point with the target acupoint.
[0015] The human acupoint location analysis system based on image recognition includes a model training module, a target image extraction module, a confidence calculation module, and a target acupoint alignment module. Model training module: used to acquire patient body images, preprocess the body images, and extract acupoint locations; use a feature detector to extract feature points in the patient body images, divide the pixel windows in the patient body images, establish a neural network model for locating human acupoints, and train the neural network model based on the patient body images. Target image extraction module: used for deploying lasers, motors, cameras, and projectors in acupuncture departments; The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images. Feature point matching and registration are performed to obtain the target image corresponding to the target object. Confidence calculation module: used to divide the pixel windows in the target image and obtain the first confidence score of each pixel window as an acupoint based on the trained neural network model; Impedance values at different pixel window locations were collected using an impedance meter. Based on the impedance values, the second confidence level of each pixel window as an acupoint was obtained. Target acupoint alignment module: used to merge pixel windows and aggregate target acupoint regions based on the first confidence level and the second confidence level; A laser emits a laser point, a camera captures the relative position of the target acupoint area and the laser point, and a drive motor adjusts the position of the laser point to align the laser point with the target acupoint.
[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a human acupoint location analysis system and method based on image recognition, including: acquiring patient body images, extracting acupoint locations, extracting feature points from the patient body images, establishing and training a neural network model; deploying a laser, a motor, a camera, and a projector, and using the projector to project crosswise from different angles to obtain a target image corresponding to the target object; dividing the pixel windows in the target image to obtain the first and second confidence scores of the pixel windows as acupoints; merging the pixel windows to aggregate the target acupoint regions; and having the camera capture the relative position of the target acupoint region and the laser point to align the laser point with the target acupoint. This invention, through the analysis of patient body images and the integration of parameters from multiple instruments and equipment, achieves human acupoint location, which helps reduce reliance on physician experience, improves the accuracy and applicability of acupoint location, and meets the needs of intelligent acupuncture-assisted operation. Attached Figure Description
[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating the human acupoint location analysis method based on image recognition of the present invention. Detailed Implementation
[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.
[0020] Example: Figure 1 As shown, this invention provides a technical solution for a human acupoint location analysis method based on image recognition, including the following steps: The patient's body image is acquired, preprocessed, and acupoint locations are extracted. In this embodiment, the preprocessing includes size normalization and histogram equalization to eliminate size differences caused by image resolution and shooting distance, and to reduce the interference of uneven lighting and shadows on subsequent recognition, thus facilitating subsequent feature point extraction and model training. The acupoint locations are determined based on the confirmed locations of acupoints by experienced physicians. The specific determination process is existing technology and will not be described in detail here.
[0021] The ORB feature detector is used to extract feature points from patient body images. Feature points are stable and prominent points on the patient's body surface. ORB (Oriented FAST and Rotated BRIEF) is a visual feature extraction algorithm that combines FAST corner detection with BRIEF feature descriptors. It has the characteristics of high computational efficiency and strong robustness to changes in illumination, scale and rotation. It is suitable for image feature point extraction and matching scenarios. Using the ORB feature detector in this scheme, feature points on the surface of the target object can be extracted stably and reliably.
[0022] The patient's body imaging includes multiple images taken in different poses and postures. The spatial distance between acupoint locations and various feature points is calculated in each patient's body image. Key feature points corresponding to acupoints are then selected and identified. Specifically: The positions of the same feature point in several patient body images and the corresponding acupoints in each image are obtained respectively. The spatial distance between the feature point and the acupoint is calculated, and the variance between the spatial distances is obtained. If the variance is less than the preset variance threshold, the feature point is regarded as the key feature point of the acupoint.
[0023] It should be noted that acupoints on the human body usually have a fixed relative positional relationship with specific parts of the body surface. The location of acupoints relies on these parts as a reference. The location of these parts is an important reference for locating acupoints, which is the key feature point in this method. For example, the acupoint is located above the key feature point (such as the Quchi acupoint, which is directly above the elbow crease), or the acupoint is located at a certain distance along a specific direction from the key feature point (such as the Waiguan acupoint, which is located a fixed distance away from the elbow crease). Therefore, this method requires first identifying the key feature points related to acupoint location from the images, and then locating the acupoint based on the relative positional relationship between the key feature points and the acupoint.
[0024] The pixel window size and window sliding step are set to divide the patient's body image into several pixel windows. The pixel window corresponding to the acupoint location area is taken as the acupoint window, and the window vector pointing to each pixel window by the key feature point is obtained. In this embodiment, the pixel window is a square with a side length of 5 pixels, that is, the pixel window is 5*5 pixels in size, and the window sliding step is 5 pixels.
[0025] A neural network model for locating acupoints on the human body is established. The model is trained using each window vector. The trained model is then used to determine the reliability of the pixel window pointed to by the input window vector as an acupoint. Specifically: Establish a federated network model, treating each patient as a source domain client, and obtain the local models obtained by parallel training of each source domain client in the federated network model, as well as the global model obtained by aggregating each local model; Federated network models are characterized by data remaining local and distributed collaborative training. By training models through federated learning, the direct transmission and centralized aggregation of raw data are avoided, which can effectively protect user data privacy and data security while ensuring training effectiveness.
[0026] The system acquires all acupoint windows and key feature point locations from the patient's body image, obtaining the window vector pointing to each pixel window from the key feature point. If the window vector points to an acupoint window, the target value of the window vector is set to 1; otherwise, it is set to 0. Each pixel window corresponds to a window vector and a target value. Typically, there is a fixed correlation between acupoint locations and key feature point locations. Therefore, there is a corresponding mapping relationship between the window vector and the target value. In this scheme, the neural network model can be trained based on this mapping relationship. In this embodiment, the model consists of an input layer, two fully connected layers, and an output layer. The ratio of training set to validation set is 7:3. It is optimized iteratively using the cross-entropy loss function. The optimizer is the Adam optimizer, the learning rate is set to 0.001, and the number of iterations is 60. The specific training process is existing technology and will not be described in detail here.
[0027] Using each window vector as input and its target value as output, a training set and a validation set are set. Local models are trained iteratively using a cross-entropy loss function. Each trained local model is then aggregated and merged to form a new global model. The trained federated network model is used to obtain the target value of the input window vector, and the output target value is used as the reliability of identifying the corresponding pixel window as an acupoint. In this embodiment, the reliability level ranges from 0 to 1; the closer the level is to 1, the greater the probability that the pixel window is an acupoint.
[0028] Lasers, motors, cameras, and projectors were deployed in the acupuncture department. The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images, and feature point matching and registration are performed to obtain the target image corresponding to the target object.
[0029] The projector projects images from different angles to eliminate blind spots, supports autofocus and keystone correction, and uses an ORB feature detector to extract feature points from the surface of the target object in each projected image. Based on feature point matching, spatial alignment of multiple projected images is achieved. Then, a Poisson fusion algorithm is used to perform pixel-level fusion processing on the aligned images, achieving seamless image stitching and generating the target image corresponding to the target object. The Poisson fusion algorithm used here is a current technology.
[0030] Extract each pixel window and key feature point from the target image to obtain the window vector pointing to each pixel window. Input each window vector into the trained neural network model to obtain the reliability of the corresponding pixel window as an acupoint, which is used as the first confidence level of the pixel window as an acupoint.
[0031] Since the reliability level ranges from 0 to 1, the first confidence level also ranges from 0 to 1.
[0032] An impedance meter is pre-deployed on the target object. The impedance meter is used to collect the impedance values of each pixel window corresponding to the body surface location. The impedance values are normalized and converted into a second confidence level for each pixel window as an acupoint.
[0033] The skin resistance at acupoints is lower and conductivity is higher, while the skin resistance in non-acupoint areas is higher and conductivity is lower. Impedance analyzers utilize this impedance difference to assist in acupoint identification. They are devices used to detect the electrical impedance of biological tissues. By inputting a weak and safe detection current into the body surface, they measure the impedance values at different locations. By leveraging the significant impedance difference between acupoint and non-acupoint areas, they assist in the location and verification of acupoints, helping to reduce location time, improve accuracy, and reduce false positives through multimodal verification, making them suitable for diverse clinical scenarios.
[0034] Normalization is a function that maps a value to 0 to 1. There are various ways to implement it, which will not be elaborated here. In this embodiment, the smaller the impedance value, the larger the normalized impedance value and the higher the second confidence level; the larger the impedance value, the smaller the normalized impedance value and the lower the second confidence level.
[0035] Based on the first and second confidence levels, pixel windows are merged to aggregate the target acupoint regions, specifically as follows: The weights of the first confidence level and the second confidence level are preset. The total confidence level of each pixel window is calculated according to the weights, and the pixel windows with a total confidence level greater than the preset confidence level threshold are marked. For example, for a certain pixel window, the first confidence level is D1, the second confidence level is D2, and according to the preset first confidence level weight W1 and second confidence level weight W2, the total confidence level is D = W1 * D1 + W2 * D2.
[0036] Traverse each pixel window. If the number of marked windows in the eight neighboring windows of a pixel window is greater than a preset threshold, then merge the pixel window and its eight neighboring windows into the target acupoint region, and thus obtain the final target acupoint region.
[0037] The eight neighboring windows refer to the top, bottom, left, right, top left, top right, bottom left, and bottom right pixel windows of a pixel window. In this embodiment, the preset number threshold is 5. When there are 6 or more pixel windows as marker windows in the eight neighboring windows, the pixel window and its eight neighboring windows are merged into the target acupoint area.
[0038] A laser emits a laser point, a camera captures the relative position of the target acupoint area and the laser point, and a drive motor adjusts the position of the laser point to align the laser point with the target acupoint.
[0039] A laser point is emitted by a laser to form a mark on the human body surface. The camera captures the position of the laser point in real time (the center coordinates of the laser point can be extracted using Canny edge detection). The center coordinates of the laser point are extracted, and the offset between the laser point and the target acupoint area is calculated. Based on the PID control algorithm, the drive motor adjusts the position of the laser point. With the offset as input, the motor step size is adjusted through proportional-integral-derivative to align the laser point with the target acupoint, so as to better display the location of the target acupoint.
[0040] This embodiment also provides a human acupoint location analysis system based on image recognition, including a model training module, a target image extraction module, a confidence calculation module, and a target acupoint alignment module; Model training module: used to acquire patient body images, preprocess the body images, and extract acupoint locations; use a feature detector to extract feature points in the patient body images, divide the pixel windows in the patient body images, establish a neural network model for locating human acupoints, and train the neural network model based on the patient body images. Target image extraction module: used for deploying lasers, motors, cameras, and projectors in acupuncture departments; The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images. Feature point matching and registration are performed to obtain the target image corresponding to the target object. Confidence calculation module: used to divide the pixel windows in the target image and obtain the first confidence score of each pixel window as an acupoint based on the trained neural network model; Impedance values at different pixel window locations were collected using an impedance meter. Based on the impedance values, the second confidence level of each pixel window as an acupoint was obtained. Target acupoint alignment module: used to merge pixel windows and aggregate target acupoint regions based on the first confidence level and the second confidence level; A laser emits a laser point, a camera captures the relative position of the target acupoint area and the laser point, and a drive motor adjusts the position of the laser point to align the laser point with the target acupoint.
[0041] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0042] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0043] The above are merely preferred embodiments of the present invention and are 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 within the protection scope of the present invention.
Claims
1. An image recognition-based human acupoint positioning analysis method, characterized in that, Includes the following steps: Acquire patient body images, preprocess the images, and extract acupoint locations; Using a feature detector, feature points are extracted from the patient's body image, pixel windows are divided in the patient's body image, a neural network model for locating acupoints is established, and the neural network model is trained based on the patient's body image. Lasers, motors, cameras, and projectors were deployed in the acupuncture department. The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images. Feature point matching and registration are performed to obtain the target image corresponding to the target object. The target image is divided into pixel windows, and the first confidence score of each pixel window as an acupoint is obtained based on the trained neural network model. Impedance values at different pixel window locations were collected using an impedance meter. Based on the impedance values, the second confidence level of each pixel window as an acupoint was obtained. Pixel windows are merged based on the first and second confidence levels to aggregate the target acupoint areas; A laser emits a laser point, and a camera captures the relative position of the target acupoint area and the laser point. The camera then drives a motor to adjust the position of the laser point, thus aligning the laser point with the target acupoint.
2. The image recognition-based human acupoint positioning analysis method according to claim 1, characterized in that, Implementing neural network model training includes: The ORB feature detector is used to extract feature points from patient body images. The feature points are stable and prominent points on the patient's body surface. The patient body images include multiple images taken in different postures. The spatial distance between the acupoints and each feature point in each patient body image is calculated, and the key feature points corresponding to the acupoints are screened and determined. Set the pixel window size and window sliding step size, divide the patient's body image into several pixel windows, take the pixel window corresponding to the acupoint location area as the acupoint window, and obtain the window vector pointing to each pixel window from the key feature points. The neural network model is trained based on each window vector. The trained neural network model is used to determine the reliability of the pixel window pointed to by the input window vector as an acupoint.
3. The image recognition-based human acupoint positioning analysis method according to claim 2, characterized in that, Screening and identifying key feature points corresponding to acupoints, including: The positions of the same feature point in several patient body images and the corresponding acupoints in each image are obtained respectively. The spatial distance between the feature point and the acupoint is calculated, and the variance between the spatial distances is obtained. If the variance is less than a preset variance threshold, the feature point is regarded as the key feature point of the acupoint.
4. The image recognition-based human acupoint positioning analysis method according to claim 2, characterized in that, To train a neural network model, the following steps are included: Establish a federated network model, treating each patient as a source domain client, and obtain the local models obtained by parallel training of each source domain client in the federated network model, as well as the global model obtained by aggregating each local model; Obtain the locations of all acupoint windows and key feature points in the patient's body image, and obtain the window vector pointing to each pixel window from the key feature point. If the window pointed to by the window vector is an acupoint window, set the target value of the window vector to 1; if the window pointed to by the window vector is not an acupoint window, set the target value of the window vector to 0. Using each window vector as input and the target value of each window vector as output, a training set and a validation set are set, and local model training is performed through iterative optimization using the cross-entropy loss function. Based on each trained local model, the models are summarized and merged to form a new global model. The trained federated network model is used to obtain the target value of the window vector based on the input window vector, and the output target value is used as the reliability of the corresponding pixel window as the acupoint.
5. The image recognition-based human acupoint positioning analysis method according to claim 1, characterized in that, Obtain the target image corresponding to the target object, including: The ORB feature detector extracts feature points from the surface of the target object in each projected image. Based on feature point matching, the spatial alignment of multiple projected images is completed. Then, the Poisson fusion algorithm is used to perform pixel-level fusion processing on the aligned images to achieve seamless image stitching and generate the target image corresponding to the target object.
6. The image recognition-based human acupoint positioning analysis method according to claim 2, characterized in that, The first confidence score for each pixel window representing an acupoint is obtained, including: Extract each pixel window and key feature point from the target image to obtain the window vector pointing to each pixel window. Input each window vector into the trained neural network model to obtain the reliability of the corresponding pixel window as an acupoint, which is used as the first confidence level of the pixel window as an acupoint.
7. The image recognition-based human acupoint positioning analysis method according to claim 1, characterized in that, The second confidence level for each pixel window representing an acupoint is obtained, including: An impedance meter is pre-deployed on the target object. The impedance meter is used to collect the impedance values of each pixel window corresponding to the body surface location. The impedance values are normalized and converted into a second confidence level for each pixel window as an acupoint.
8. The method for locating and analyzing human acupoints based on image recognition according to claim 1, characterized in that, Perform pixel window merging to aggregate the target acupoint area, including: The weights of the first confidence level and the second confidence level are preset. The total confidence level of each pixel window is calculated according to the weights. Pixel windows with a total confidence level greater than the preset confidence level threshold are marked. Traverse each pixel window. If the number of marked windows in the eight neighboring windows of a pixel window is greater than a preset threshold, then merge the pixel window and its eight neighboring windows into the target acupoint region, and thus obtain the final target acupoint region.
9. The method for locating and analyzing human acupoints based on image recognition according to claim 1, characterized in that, To align the laser point with the target acupoint, including: A laser point is emitted by a laser to form a mark on the human body surface. The camera captures the position of the laser point in real time, extracts the center coordinates of the laser point, and calculates the offset between the laser point and the target acupoint area. Based on the PID control algorithm, the drive motor adjusts the position of the laser point. With the offset as input, the motor step size is adjusted through proportional-integral-derivative to align the laser point with the target acupoint.
10. A human acupoint location analysis system based on image recognition, used to execute the human acupoint location analysis method based on image recognition as described in any one of claims 1-9, characterized in that, The system includes a model training module, a target image extraction module, a confidence calculation module, and a target acupoint alignment module; Model training module: used to acquire patient body images, preprocess the body images, and extract acupoint locations; Using a feature detector, feature points are extracted from the patient's body image, pixel windows are divided in the patient's body image, a neural network model for locating acupoints is established, and the neural network model is trained based on the patient's body image. Target image extraction module: used for deploying lasers, motors, cameras, and projectors in acupuncture departments; The target object is kept stationary in a fixed posture. A projector is used to project images from different angles. Feature points of the target object are extracted from the projected images. Feature point matching and registration are performed to obtain the target image corresponding to the target object. Confidence calculation module: used to divide the pixel windows in the target image and obtain the first confidence score of each pixel window as an acupoint based on the trained neural network model; Impedance values at different pixel window locations were collected using an impedance meter. Based on the impedance values, the second confidence level of each pixel window as an acupoint was obtained. Target acupoint alignment module: used to merge pixel windows and aggregate target acupoint regions based on the first confidence level and the second confidence level; A laser emits a laser point, and a camera captures the relative position of the target acupoint area and the laser point. The camera then drives a motor to adjust the position of the laser point, thus aligning the laser point with the target acupoint.