Automatic cupping robot

Abstract

The invention provides an automatic cupping robot, which belongs to the technical field of robots, and is characterized in that a first acquisition module scans and reconstructs a three-dimensional contour of a target cupping area, a second acquisition module extracts user interaction input information to generate an additional point location map, and a fusion module compares the two with a standard acupoint contour to obtain a positioning contour; the control module automatically generates a cupping scheme including modes such as cupping reserving and cupping moving based on the positioning contour and user data, and drives the mechanical arm to execute operation; the prevention module collects skin images and user feedback in real time, safety protection is achieved, automation, individuation and safety of the cupping process are achieved, and the cupping precision, adaptability and safety are improved.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to an automatic cupping robot. Background Technology

[0002] With the increasing awareness of health, cupping, as a traditional Chinese medicine therapy, is widely used in daily health care and as an adjunct treatment for diseases due to its significant effects in relieving muscle soreness and unblocking meridians. However, existing cupping methods still have many shortcomings: First, manual cupping relies on the experience and judgment of professional physicians, and the accuracy of acupoint location is greatly affected by the physician's skill level. Ordinary users find it difficult to operate on their own, and the uneven distribution of professional physician resources cannot meet the public's demand for convenient cupping. Second, existing cupping equipment is mostly fixed in mode, lacking adaptability to individual user differences. For example, different users have different physical conditions, pain perception, and fat thickness, and uniform cupping parameters are difficult to achieve ideal relief effects, and may even cause discomfort to users due to improper parameters. Third, existing cupping processes lack effective real-time safety monitoring mechanisms, making it impossible to capture changes in the user's skin condition and subjective discomfort feedback in a timely manner, which can easily lead to problems such as excessive skin redness and swelling, and severe bruising, thus requiring improvement in safety. Fourth, existing technologies lack accurate conversion and utilization of user interactive input information, making it difficult to effectively combine the user's subjective needs such as pain perception and expected effects with objective cupping operations, resulting in insufficient personalization of cupping plans. Furthermore, existing equipment lacks AI vision technology, making it impossible to achieve dynamic and precise acupoint positioning through intelligent image analysis, and also hindering real-time adjustments to operational strategies based on visual data. Simultaneously, the lack of AI vision's ability to quantitatively assess skin condition further limits the intelligence and safety of the cupping process. To address these technical problems, this invention provides an automated cupping robot. Summary of the Invention

[0003] This invention provides an automatic cupping robot to solve the aforementioned technical problems.

[0004] This invention provides an automatic cupping robot, comprising: The first acquisition module is used to scan the target cupping area of ​​the target cupping patient to obtain point cloud data, and to reconstruct the three-dimensional region contour from the point cloud data. The second data acquisition module is used to receive the interactive input of the target cupping subject to the cupping questionnaire survey, extract multi-dimensional information from the interactive input, and perform transformation analysis on each dimension of information to obtain an additional point map. The interactive input includes: the target cupping subject's pain perception and expected relief effect on the target cupping area, the target cupping subject's physical condition and individual preferences, and the additional point map is related to the explicit additional contour points of the corresponding dimension information. The fusion module is used to fuse the three-dimensional region contour with the additional point map in each dimension to obtain the additional region contour, and compare and analyze it with the standard human acupoint contour to obtain the first positioning contour. The control module is used to automatically generate a cupping plan based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception, and to control the cupping robotic arm to perform corresponding cupping operations on each cupping acupoint. The cupping operations include: residual cupping, moving cupping and flash cupping. The prevention module is used to control the third acquisition module to acquire skin images of the corresponding cupping acupoints when the cupping robotic arm performs the cupping operation, and to activate the voice interaction device to collect the cupping feedback information of the target cupping user during the cupping process, so as to carry out safety protection.

[0005] Preferably, the first acquisition module includes: An initial construction unit is used to select the cross-sectional direction from the structure-section lookup table according to the skeleton structure of the target cupping area, and select the first data from the point cloud data to construct the initial framework based on the azimuth angle between the cross-sectional direction and the center line of the skeleton structure. The first intermediate unit is used to sequentially supplement the remaining point cloud data into the initial frame according to the scanning path to obtain the first intermediate contour. The second intermediate unit is used to input the point cloud data into a position registration model that matches the skeleton structure of the target cupping area, to obtain a view based on the direction of each section, and to arrange all views in directional order to obtain the second intermediate contour. The line deviation unit is used to obtain the relative path of the center point of the scanning spot based on the center point of the target cupping area during the movement of the center point of the scanning spot along the scanning path, and to obtain the relative deviation line with the scanning path, wherein the thickness of each position point in the relative deviation line is consistent with the deviation magnitude of the corresponding position point. The contour deviation processing unit is used to perform position deviation processing on the second intermediate contour to obtain a three-dimensional region contour based on the first deformation vector of adjacent point cloud data in the same intermediate contour and the second deformation vector of the first intermediate contour and the second intermediate contour based on the same contour position, and in combination with the undulation and jumping process and relative deviation line of the target cupping area.

[0006] Preferably, the second acquisition module includes: The representation determination unit is used to extract all information features of each dimension and obtain the semantic representation of each information feature. The correlation degree and coefficient determination unit is used to match the semantic expression with the physiological region description items in the physiological semantic association library, determine the physiological dimension correlation degree corresponding to the semantic expression, and identify the fuzzy expression items in the semantic expression, and determine the dimension deviation coefficient of the semantic expression based on the fuzziness level of the fuzzy expression items. The accuracy determination unit is used to determine the accuracy of the semantic representation according to the physiological dimension correlation degree and the dimension deviation coefficient. The preliminary delineation unit is used to initially delineate the initial range corresponding to the semantic expression based on the physiological region description item corresponding to the semantic expression and the skeletal structure features of the cupping area of ​​the target cupping patient. At the same time, it marks the area to be refined in the initial range that corresponds to the expression precision. The refinement unit is used to extract the adjacent region association rules of the cupping acupoints corresponding to the region to be refined, and combine the association feature items of the semantic expression to perform coarse positioning selection of the region to be refined, and refine the outline of the coarse selection according to the expression accuracy to obtain the corresponding region coverage. The vector construction unit is used to match each information feature in the same dimension with the predefined terms of the corresponding dimension one by one to obtain the feature-confidence vector of each predefined term; The bitmap acquisition unit is used to construct a separate initial layer based on each predefined item according to the region coverage of each element in each feature-confidence vector and in combination with the corresponding confidence, and to perform a first rendering process on the separate initial layer according to the dimension attribute of the corresponding dimension and a second rendering process on the additional points in the separate initial layer to obtain a separate rendering layer as an additional bitmap.

[0007] Preferably, the point map acquisition unit includes: The array sub-unit is used to determine the confidence level of each initial point in the individual initial layer, obtain the rendering coefficient and point type of each initial point according to the result of a rendering process, and obtain the value array of the corresponding initial point. When the initial point is located in the refined area of ​​the corresponding area coverage, the corresponding point type is regarded as a fine type; otherwise, the corresponding point type is regarded as a coarse type. The function construction subunit is used to perform cluster analysis on all value arrays in the single initial layer with the number of conventional cupping acupoints in the target cupping area as the number of clusters, and to construct a distance trend map of the corresponding cluster analysis results to obtain the concentrated distance density function based on the distance between each core feature in the corresponding dimension information and the cluster feature corresponding to the cluster center cluster of each cluster analysis result. The first point determines the sub-unit, which is used to extract matching features and the first point corresponding to the matching features from all core features based on the central distance density function of each cluster analysis result; The second point determines the sub-unit, which is used to analyze the maximum length step change, average length step change, and minimum length step change of adjacent lengths after being sorted in order of distance length under the lumped density function, and based on the first ratio of the average length step change to the maximum length step change and the second ratio of the minimum length step change to the average length step change. Based on the first ratio and the second ratio, the central point of the concentrated distance density function is corrected to obtain a concentrated representative point as the second point; Significantly additional subunits are used to make the first point and the second point explicit additional contour points in the separate initial layer.

[0008] Preferably, the fusion module includes: Point marking unit, used to mark the explicit additional contour points in the additional point map of each dimension in the three-dimensional region contour to obtain the additional region contour; The comparison unit is used to compare the outline of the additional area with the outline of the standard human acupoints one by one to obtain the first positioning outline.

[0009] Preferably, the control module includes: The effective point determination unit is used to determine effective acupoints based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception. The parameter set determination unit is used to determine the cupping parameter set for each effective acupoint based on the distribution of effective acupoints and the theoretical interconnection effect of acupoints. The scheme correction unit is used to obtain a preliminary scheme based on the cupping parameter set, and to correct it according to the cupping parameter anomaly set of adjacent effective acupoints in the preliminary scheme to obtain the cupping scheme.

[0010] Preferably, the prevention module includes: The skin vector construction unit is used to extract features from the skin image to obtain the proportion of redness area, skin wrinkling degree and redness texture density of the cupping area, and construct the first input vector; The feedback vector construction unit is used to analyze the semantic negativity and emotional anxiety of cupping feedback information and construct the second input vector. The model analysis unit is used to input the first input vector and the second input vector into the dual-vector analysis model to obtain safety protection measures, and then send them to the corresponding cupping robotic arm to perform control operations.

[0011] Preferred options also include: The display module is used to display the operation information for each cupping acupoint, including: negative pressure value, cupping duration, and cupping mode.

[0012] Compared with the prior art, the beneficial effects of this application are as follows: Extreme positioning accuracy: Through a dynamic deviation correction algorithm linked to breathing frequency, the positioning error of acupoints is reduced, and users of different ages, physical conditions and breathing states can obtain accurate positioning; The integration of AI vision technology enables intelligent scanning and feature extraction of the three-dimensional contour of the target cupping area. Combined with comparative analysis of the standard acupoint database, the intelligence and accuracy of acupoint positioning are further improved, effectively avoiding the experience bias of manual positioning.

[0013] Personalized and Dynamic Adaptation: Real-time feedback and linked semantic mapping, along with adaptive parameter dynamic adjustment, enhance personalization. AI vision can capture individual characteristics such as the user's skin contour curves and fat thickness distribution, providing visual data support for the dynamic adjustment of cupping parameters, making the treatment plan more suitable for the user's actual physical condition.

[0014] Synergistic Effects: The synergistic optimization logic of acupoint combinations significantly enhances the effects of cupping; at the same time, AI vision monitors the skin condition in real time during cupping, forming two-way data support with user feedback, ensuring that the cupping operation maximizes the relief effect within a safe range.

[0015] In summary, the goal is to automate, personalize, and ensure the safety of the cupping process, thereby improving its accuracy, adaptability, and safety.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of an automatic cupping robot according to an embodiment of the present invention; Figure 2 This is a theoretical structural diagram of the automatic cupping robot in an embodiment of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] This invention provides an automatic cupping robot, such as Figure 1 As shown, it includes: The first acquisition module is used to scan the target cupping area of ​​the target cupping patient to obtain point cloud data, and to reconstruct the three-dimensional region contour from the point cloud data. The second data acquisition module is used to receive the interactive input of the target cupping subject to the cupping questionnaire survey, extract multi-dimensional information from the interactive input, and perform transformation analysis on each dimension of information to obtain an additional point map. The interactive input includes: the target cupping subject's pain perception and expected relief effect on the target cupping area, the target cupping subject's physical condition and individual preferences, and the additional point map is related to the explicit additional contour points of the corresponding dimension information. The fusion module is used to fuse the three-dimensional region contour with the additional point map in each dimension to obtain the additional region contour, and compare and analyze it with the standard human acupoint contour to obtain the first positioning contour. The control module is used to automatically generate a cupping plan based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception, and to control the cupping robotic arm to perform corresponding cupping operations on each cupping acupoint. The cupping operations include: residual cupping, moving cupping and flash cupping. The prevention module is used to control the third acquisition module to acquire skin images of the corresponding cupping acupoints when the cupping robotic arm performs the cupping operation, and to activate the voice interaction device to collect the cupping feedback information of the target cupping user during the cupping process, so as to carry out safety protection.

[0021] In this embodiment, the target cupping user refers to a person who needs to undergo cupping to relieve physical discomfort or achieve health care purposes.

[0022] In this embodiment, a 3D laser scanner is used to scan the surface of the target cupping area to obtain point cloud data.

[0023] In this embodiment, interactive input refers to the information related to cupping that the user inputs through the interactive methods provided by the robot. These interactive methods include completing a questionnaire, answering questions via voice, and selecting options on a touchscreen. For example, if a user selects a question on the robot's touchscreen, such as "I have soreness on the left side of my lower back, I want to relieve muscle stiffness, I have a cold constitution, and I prefer a gentle cupping intensity," these selections and input constitute the interactive input.

[0024] Pain perception refers to the discomfort felt by the target cupping patient in the target cupping area, including the location, intensity, type, and duration of pain. Expected relief effect refers to the improvement in physical condition that the target cupping patient hopes to achieve through cupping.

[0025] Physical condition refers to the basic physical condition of the person receiving cupping, including body type, skin sensitivity, presence of skin diseases, fat thickness, and presence of underlying diseases.

[0026] Individual preferences refer to the preferences of the target cupping patient during the cupping process, including cupping intensity, cupping duration, cupping mode, etc.

[0027] In this embodiment, multidimensional information extraction refers to extracting relevant effective information from the user's interactive input according to preset dimensions such as pain perception, expected relief effect, physical fitness, and individual preferences. For example, from the user's interactive input, the following can be extracted: pain perception dimension: left side lumbar pain; expected relief effect dimension: relief of muscle stiffness; physical fitness dimension: cold constitution; and individual preference dimension: mild intensity.

[0028] In this embodiment, the additional point map refers to a graph containing explicit additional contour points related to the information of each dimension, obtained after transforming and analyzing the information of each dimension.

[0029] Fusion processing refers to overlaying and integrating the 3D region contour with additional point maps from various dimensions, so that the explicit additional contour points in the additional point maps accurately correspond to the corresponding positions in the 3D region contour, forming an additional region contour containing more information. Standard human acupoint contours refer to a pre-constructed standard 3D contour model containing all cupping-related acupoints on the human body, based on human anatomy and traditional Chinese medicine acupoint theory. This model clearly defines the standard location and range of each acupoint.

[0030] The first positioning contour refers to the contour obtained by comparing the additional area contour with the standard human acupoint contour to accurately locate the relevant acupoints for cupping. This contour includes both the individual regional characteristics of the target cupping patient and accurately corresponds to the location of the standard acupoints. For example, by comparing the additional area contour of the user's waist with the standard waist acupoint contour, the actual location of acupoints such as Shenshu and Dachangshu on the user's waist can be accurately located, and the resulting contour is the first positioning contour.

[0031] In this embodiment, the global acupoint refers to all points corresponding to cupping-related acupoints contained in the first positioning contour, covering all potential cupping acupoints in the target cupping area.

[0032] In this embodiment, a fat thickness prediction model is established based on the depth information of the three-dimensional region contour and combined with the statistical data of human fat distribution. The predicted fat thickness is obtained by inputting acupoint location information and user physical fitness data.

[0033] In this embodiment, the acupoint contour curve refers to the local skin contour curve at the location of each acupoint, reflecting the skin undulation characteristics at that location. For example, the skin contour curve at the Zusanli acupoint on the leg is relatively flat.

[0034] In this embodiment, based on the user's pain perception and expected relief effect, the corresponding key acupoints are selected from the global acupoint database according to the symptoms-acupoint correspondence database. For example, if a user has lower back pain, according to traditional Chinese medicine theory, the Shenshu, Dachangshu, and Weizhong acupoints in the lower back are key acupoints for relieving lower back pain.

[0035] In this embodiment, based on the matching degree analysis between pain perception information and acupoint efficacy, acupoints with high matching degree are selected from global acupoints as strongly related acupoints. For example, if a user has soreness on the right side of their neck, acupoints such as Fengchi (right side of the neck) and Jianjing (connection between the shoulder and neck) that match the pain perception are strongly related acupoints.

[0036] A cupping plan refers to a complete operational procedure determined based on effective acupoints, including information such as cupping parameters, cupping operation type, and cupping sequence for each effective acupoint. For example, a cupping plan for a user with lower back pain would be: Kidney Shu (BL23): negative pressure 0.04 MPa, cupping time 8 minutes; Large Intestine Shu (BL25): negative pressure 0.035 MPa, cupping time 7 minutes; Weizhong (BL40): negative pressure 0.03 MPa, flash cupping time 5 minutes; the cupping sequence would start from Kidney Shu (BL23), followed by Large Intestine Shu (BL25) and Weizhong (BL40).

[0037] A cupping robotic arm refers to a mechanical structure in a robot used to perform cupping operations. According to the instructions of the cupping plan, it moves to the designated acupoint and performs the corresponding cupping operation. For example, the robotic arm carries cupping cups, moves to the Shenshu acupoint on the user's waist according to the instructions, adjusts the angle, applies a preset negative pressure value, and performs the cupping operation.

[0038] Retention cupping refers to the cupping operation method in which the cupping device is attached to the acupoint and kept still for a fixed time; moving cupping refers to the cupping operation method in which the cupping device is attached to the skin surface and moved slowly along a preset path; flash cupping refers to the cupping operation method in which the cupping device is quickly attached to the skin surface and immediately removed, and repeated multiple times.

[0039] The third data acquisition module uses a high-definition camera to capture real-time images of the skin condition at the cupping acupoints on the user's waist.

[0040] Skin images refer to the skin surface images of the cupping acupoints and surrounding areas captured by the third acquisition module, reflecting the skin's color (such as the degree of redness), whether there is redness, swelling, bruising, damage, etc.; voice interaction devices refer to the devices in the robot used to collect user voice feedback and provide voice prompts, including microphones, speakers, etc.

[0041] Cupping feedback information refers to the cupping-related sensory information provided by the user during the cupping process via voice interaction devices or other interactive methods. This includes information such as discomfort (e.g., tingling, numbness, excessive tightness), comfort evaluation, and real-time feedback on the effects. In this embodiment, safety protection refers to measures taken based on skin image analysis results and cupping feedback information to prevent injury to the user during cupping. These measures include adjusting cupping parameters (e.g., reducing negative pressure, shortening duration), pausing the cupping operation, and terminating the cupping operation.

[0042] like Figure 2 The diagram shows the theoretical structure of the automatic cupping robot.

[0043] The beneficial effects of the above technical solution are as follows: the first acquisition module realizes accurate three-dimensional contour reconstruction of the target cupping area; the second acquisition module extracts and transforms user individual information in multiple dimensions; the fusion module realizes personalized acupoint positioning; the control module generates customized cupping plans based on multi-dimensional parameters; and the prevention module monitors and ensures cupping safety in real time. Overall, the cupping operation is automated, personalized, and safe, effectively improving the accuracy of acupoint positioning and the adaptability of cupping plans. It avoids the problems of experience dependence and inappropriate parameters in manual cupping, and reduces the risk of cupping through real-time safety protection, thereby improving the user experience and cupping effect.

[0044] This invention provides an automatic cupping robot, wherein the first data acquisition module includes: An initial construction unit is used to select the cross-sectional direction from the structure-section lookup table according to the skeleton structure of the target cupping area, and select the first data from the point cloud data to construct the initial framework based on the azimuth angle between the cross-sectional direction and the center line of the skeleton structure. The first intermediate unit is used to sequentially supplement the remaining point cloud data into the initial frame according to the scanning path to obtain the first intermediate contour. The second intermediate unit is used to input the point cloud data into a position registration model that matches the skeleton structure of the target cupping area, to obtain a view based on the direction of each section, and to arrange all views in directional order to obtain the second intermediate contour. The line deviation unit is used to obtain the relative path of the center point of the scanning spot based on the center point of the target cupping area during the movement of the center point of the scanning spot along the scanning path, and to obtain the relative deviation line with the scanning path, wherein the thickness of each position point in the relative deviation line is consistent with the deviation magnitude of the corresponding position point. The contour deviation processing unit is used to process the first deformation vector of adjacent point cloud data in the same intermediate contour and the second deformation vector of the first and second intermediate contours based on the same contour position, and in combination with the undulation and jumping process of the target cupping area and the relative deviation. The difference line is used to process the positional deviation of the second intermediate contour to obtain the three-dimensional region contour.

[0045] In this embodiment, based on human anatomical data, a skeletal structure model of each body part is established and stored in the robot database. Through the three-dimensional contour analysis of the first acquisition module, the skeletal structure corresponding to the target cupping area is determined. The skeletal structure refers to the human skeletal support structure corresponding to the target cupping area, including the distribution, shape, connection relationship and other features of the bones.

[0046] In this embodiment, the structure-section lookup table refers to a pre-established lookup table that records the skeletal structure of different body parts and the corresponding matching cross-sectional directions. Some of the contents are shown in Table 1: Table 1 Structure-Section Comparison Table In this embodiment, the cross-sectional direction refers to the cross-sectional cutting angle and direction used when constructing the contour of the target cupping area, which is used to obtain the contour features of the area from different angles. For example, when constructing the contour of the waist, a horizontal cross-section (parallel to the ground) and a vertical cross-section (perpendicular to the ground) are used as the cross-sectional direction.

[0047] In this embodiment, the centerline refers to the central axis of the skeletal structure of the target cupping area. For example, the centerline of the lumbar skeletal structure is the longitudinal central axis of the lumbar vertebrae, which extends vertically. The azimuth angle refers to the angle between the cross-sectional direction and the centerline of the skeletal structure. For example, the centerline of the lumbar skeletal structure is vertical, the azimuth angle between the horizontal cross-sectional direction and the centerline is 90°, and the azimuth angle between the vertical cross-sectional direction and the centerline is 0°.

[0048] The first data refers to the core point cloud data selected from the point cloud data that matches the cross-sectional direction and azimuth angle and is used to construct the initial framework. For example, in the waist point cloud data, the point cloud data that matches the horizontal and vertical cross-sections is the first data.

[0049] In this embodiment, the initial framework refers to a preliminary outline framework that reflects the core skeletal structure of the target cupping area, constructed based on the first data. For example, a preliminary framework containing the core outline of the lumbar vertebrae and the connection position of the sacrum is constructed based on the first data of the waist.

[0050] In this embodiment, the scanning path refers to the movement path of the scanning spot when the first acquisition module (such as a 3D laser scanner) scans the target cupping area. For example, when scanning the back, the scanning path is a longitudinal parallel path from below the neck to above the waist, with a spacing of 0.5cm between each path to ensure full coverage of the back area.

[0051] In this embodiment, the first intermediate contour refers to the preliminary complete contour obtained after the remaining point cloud data is sequentially added to the initial frame according to the scanning path. It includes the core structure of the skeleton and some surface details. For example, the remaining point cloud data of the waist (point clouds corresponding to muscle and skin details) is sequentially added to the initial frame according to the longitudinal parallel scanning path to obtain the first intermediate contour that can reflect the waist skeleton and preliminary muscle contour.

[0052] In this embodiment, detailed construction information of the location registration model is as follows: Model Structure: An improved CNN architecture is adopted, consisting of an input layer, four convolutional layers (Conv1-Conv4), two pooling layers (Pool1-Pool2), one fully connected layer (FC1), and an output layer. Specifically, Conv1 uses 3×3 convolutional kernels (64 kernels); Conv2 uses 3×3 convolutional kernels (128 kernels); Conv3 uses 5×5 convolutional kernels (256 kernels); and Conv4 uses 5×5 convolutional kernels (512 kernels). Pool1 and Pool2 both use 2×2 max pooling. FC1 has 1024 nodes. The output layer is a regression layer, outputting the view coordinate mapping relationship for each cross-section.

[0053] Training data: Sample type: 2,000 sets of point cloud data and corresponding cross-sectional view annotation data of target cupping areas covering people of different ages (18-65 years old) and body types (thin / medium / fat), involving four core cupping areas: shoulder and neck, waist, back and legs.

[0054] Annotation method: Three senior TCM physicians, in conjunction with human anatomical atlases, annotated the acupoint locations and contour boundaries of the cross-sectional views corresponding to each point cloud data, namely horizontal, vertical, and 45° tilt, with an annotation accuracy error ≤0.1cm.

[0055] Sample size: 500 sets for the shoulder and neck area, 600 sets for the waist, 500 sets for the back, and 400 sets for the legs, divided into training set, validation set, and test set in a ratio of 7:2:1.

[0056] Training process: Loss function: The mean squared error (MSE) loss function is used. ,in, These are the actual view coordinates. The coordinates are predicted by the model, and N is the number of samples.

[0057] Optimizer: The Adam optimizer is selected with an initial learning rate of 0.001, a decay rate of 10% every 100 rounds, and a weight decay coefficient of 0.0001.

[0058] Number of iterations and convergence condition: 300 iterations, training stops when the loss function value on the validation set is less than 0.0005 for 10 consecutive iterations.

[0059] Input and output parameters: Input: Point cloud data of the target cupping area, in the format of an XYZ three-dimensional coordinate matrix with dimensions of 1024×3, and skeletal structure type such as lumbar spine skeleton, shoulder and neck scapula skeleton, etc.

[0060] Output: A two-dimensional view of each cross-section, with a resolution of 512×512 pixels, including outline boundaries and acupoint candidate region markers.

[0061] In this embodiment, a view refers to a two-dimensional image of the corresponding cross-sectional direction obtained after the point cloud data is converted through a position registration model. For example, the horizontal cross-sectional view of the waist is a two-dimensional image parallel to the ground, which can clearly show the cross-section of the lumbar vertebrae and the thickness distribution of the surrounding muscles under the cross-section.

[0062] In this embodiment, the azimuth angle between each cross-section direction and the centerline is calculated, and the azimuth order is determined by the order of the azimuth values ​​from smallest to largest or from largest to smallest.

[0063] In this embodiment, a view stitching and 3D reconstruction algorithm is used to integrate the views arranged in orientation order to construct the second intermediate contour.

[0064] In this embodiment, the scanning spot refers to the spot emitted by the first acquisition module (such as a 3D laser scanner) for scanning the target area, and the movement trajectory of the center point is the scanning path; the relative path refers to the trajectory of the position change of the center point of the scanning spot relative to the center point of the target cupping area during the movement according to the scanning path.

[0065] In this embodiment, the relative deviation line refers to the deviation trajectory between the relative path and the preset scanning path. The thickness of this line reflects the magnitude of the deviation at each position point; that is, the larger the deviation, the thicker the line; the smaller the deviation, the thinner the line. For example, if the preset scanning path is a straight scanning path for the shoulder, and the relative path deviates slightly due to user movement, the resulting relative deviation line may have a deviation of 0.3cm at one position, corresponding to a line thickness of 0.3mm, and a deviation of 0.1cm at another position, corresponding to a line thickness of 0.1mm.

[0066] In this embodiment, the first deformation vector refers to the positional change vector between two adjacent point cloud data within the same intermediate contour. It includes direction and magnitude, reflecting the relative deformation of adjacent point clouds. For example, in the first intermediate contour, for two adjacent point cloud data A and B, where the coordinates of A are (x1, y1, z1) and the coordinates of B are (x2, y2, z2), the first deformation vector is (x2-x1, y2-y1, z2-z1), with a magnitude equal to the distance between the two points and a direction from A to B.

[0067] The second deformation vector refers to the positional change vector between the point cloud data of the first and second intermediate contours at the same contour location. It includes direction and magnitude, reflecting the deformation difference between the two intermediate contours at that location. For example, near a certain position C0 (x0, y0, z0) on the waist contour, the point cloud data coordinates of the first intermediate contour are C1 (x3, y3, z3), and the point cloud data coordinates of the second intermediate contour are C2 (x4, y4, z4). Then the second deformation vector is (x4-x3, y4-y3, z4-z3), with the direction pointing from C1 to C2.

[0068] In this embodiment, the undulating motion process refers to the up-and-down movement and slight shaking of the target cupping area caused by breathing, slight limb movements, etc., during the scanning process. For example, when the user is scanning the waist, breathing causes the waist area to rise and fall with an amplitude of 0.2-0.3cm, which is the undulating motion process.

[0069] Existing cupping robots often employ static deviation compensation for contour reconstruction (e.g., correcting only scan path deviations and point cloud noise), but neglect the dynamic interference of the human body. The target cupping area will experience real-time displacement due to breathing (fluctuation amplitude of 0.1-0.3cm) and limb micro-movements (0.05-0.2cm), resulting in a deviation of 0.2-0.5cm between the statically corrected contour and the actual area, far exceeding the ±0.15cm accuracy requirement for acupoint positioning. Therefore, a dynamic deviation correction model is proposed with the following formula: ,in, This is the total deviation correction amount. This is the correction amount for the first deformation vector; This is the second deformation vector correction amount. The fluctuation influence coefficient is, and , The fluctuation range; Basic influence coefficient; To achieve real-time breathing rate, compared to the original dynamic deviation correction algorithm, the contour reconstruction accuracy is further improved by 20%, and the acupoint positioning error is reduced from ±0.15cm to ±0.12cm. When the user's breathing state fluctuates (such as shallow and rapid breathing due to tension, and deep and slow breathing due to relaxation), the contour stability is improved by 35%, effectively avoiding positioning deviation caused by changes in breathing state.

[0070] It should be noted that, Compensation for static deviation reflects the inherent deviation between point cloud data and contour construction, and is the benchmark equilibrium state of contour reconstruction. It compensates for dynamic deviations and real-time compensation for short-term dynamic disturbances such as breathing and micro-movements, so that the contour always conforms to the dynamic equilibrium state of the actual area. The value range was optimized through an orthogonal experiment involving 30 subjects with different physical conditions to ensure that it is applicable in most scenarios. The correction contribution is reasonable; The values ​​cover the normal adult respiratory rate range (12-18 breaths / minute at rest, 5-20 breaths / minute in extreme conditions).

[0071] The beneficial effects of the above technical solution are as follows: by selecting core data through the initial construction unit to build the initial framework, two complementary intermediate contours are obtained by combining the first intermediate unit and the second intermediate unit respectively, the relative deviation line of the scanning path is obtained through the line deviation unit, and finally the deviation correction is performed by the contour deviation processing unit in combination with multiple types of deformation vectors, undulation and jumping process and relative deviation line. This significantly improves the construction accuracy of the three-dimensional region contour, effectively avoids the influence of factors such as scanning path deviation, human dynamic interference, and point cloud data error on contour reconstruction, and provides high-quality basic data support for subsequent accurate acupoint positioning.

[0072] This invention provides an automatic cupping robot, wherein the second data acquisition module includes: The representation determination unit is used to extract all information features of each dimension and obtain the semantic representation of each information feature. The correlation degree and coefficient determination unit is used to match the semantic expression with the physiological region description items in the physiological semantic association library, determine the physiological dimension correlation degree corresponding to the semantic expression, and identify the fuzzy expression items in the semantic expression, and determine the dimension deviation coefficient of the semantic expression based on the fuzziness level of the fuzzy expression items. The accuracy determination unit is used to determine the accuracy of the semantic representation according to the physiological dimension correlation degree and the dimension deviation coefficient. The preliminary delineation unit is used to initially delineate the initial range corresponding to the semantic expression based on the physiological region description item corresponding to the semantic expression and the skeletal structure features of the cupping area of ​​the target cupping patient. At the same time, it marks the area to be refined in the initial range that corresponds to the expression precision. The refinement unit is used to extract the adjacent region association rules of the cupping acupoints corresponding to the region to be refined, and combine the association feature items of the semantic expression to perform coarse positioning selection of the region to be refined, and refine the outline of the coarse selection according to the expression accuracy to obtain the corresponding region coverage. The vector construction unit is used to match each information feature in the same dimension with the predefined terms of the corresponding dimension one by one to obtain the feature-confidence vector of each predefined term; The bitmap acquisition unit is used to construct a separate initial layer based on each predefined item according to the region coverage of each element in each feature-confidence vector and in combination with the corresponding confidence, and to perform a first rendering process on the separate initial layer according to the dimension attribute of the corresponding dimension and a second rendering process on the additional points in the separate initial layer to obtain a separate rendering layer as an additional bitmap.

[0073] In this embodiment, dimensional information refers to four categories of information extracted from user interaction input, categorized according to pain perception, expected relief effect, physical fitness, and individual preferences. Each category of information is a dimension of dimensional information.

[0074] In this embodiment, information features refer to key information items contained in each dimension that can reflect the core content of that dimension. For example, the information feature of pain perception dimension information, left side neck pain lasting 2 days, is: pain location: left side of neck pain duration: 2 days.

[0075] Semantic representation refers to the content of the representation after describing each information feature in natural language.

[0076] In this embodiment, the physiological semantic association library includes four categories: body part description items (300 items), physiological state description items (200 items), disease description items (250 items), and expected effect description items (150 items). For example, body part description items include: left side of the neck C4-C5 region, left side of the lumbar L3-L4 vertebrae, and a 2cm range around the Jianjing acupoint.

[0077] In this embodiment, the physiological region description item refers to the specific description items stored in the physiological semantic association library used to describe human physiological characteristics, including body part descriptions, physiological state descriptions, and disease descriptions. For example, the body part description item is the middle of the waist and the inner side of the wrist; the physiological state description item is a constitution that is prone to cold and a high tolerance for skin.

[0078] In this embodiment, the cosine similarity algorithm is used to calculate the similarity between the semantic representation and each physiological region description item. The formula is as follows: Where A is the word vector of semantic expression and B is the word vector of descriptive item, and then the physiological dimension correlation is obtained by weighted average.

[0079] In this embodiment, a vague expression refers to a semantic expression whose meaning is unclear or whose scope is uncertain. For example, the semantic expression "a little back pain" (without specifying the exact location) is a vague expression.

[0080] In this embodiment, the fuzziness level refers to the degree of fuzziness of the fuzzy description, which is usually divided into three levels: slight fuzziness, moderate fuzziness, and severe fuzziness. For example, the upper back (a relatively broad range) in "a little soreness in the upper back" is slightly fuzzy.

[0081] In this embodiment, a correspondence table between fuzzy levels and dimensionality deviation coefficients is established, as shown in Table 2: Table 2. Correspondence between fuzzy levels and dimensionality deviation coefficients In this embodiment, the accuracy of the description = physiological dimension correlation degree × dimension deviation coefficient.

[0082] In this embodiment, the approximate body part is determined based on the physiological region description item. Combined with the skeletal structure features of the part, such as bone distribution and muscle range, the initial range is determined by geometric delineation. For example, the physiological region description item corresponding to the semantic expression of soreness on the left side of the neck is the left side of the neck. Combined with the skeletal structure features of the neck (the left cervical spine and surrounding muscle area), the initial range is initially delineated as the area on the left side of the neck from below the angle of the mandible to above the clavicle.

[0083] In this embodiment, the division rules for the region to be refined are set according to the numerical value of the expression precision: when the precision is ≤0.7, the region to be refined is 60% of the initial range; when the precision is 0.7-0.9, the region to be refined is 30%; when the precision is ≥0.9, the region to be refined is 10%. Based on this rule, the region to be refined is marked in the initial range.

[0084] In this embodiment, the adjacent region association rule describes the set of associations between the acupoints in the region to be refined and the surrounding acupoints / tissues. It includes three core elements: location association: the distance between adjacent acupoints is ≤3cm (based on the TCM acupoint location standard); functional association: the acupoints have complementary effects, such as relieving muscle soreness and unblocking meridians; and meridian association: they belong to the same meridian or are interconnected. The specific construction method is as follows: Establish a meridian-acupoint mapping database, which includes the location, efficacy, and spacing data of 14 major meridians and their corresponding acupoints; For each commonly used acupoint for cupping, surrounding acupoints that meet the conditions of location, function, and meridian association are selected from the database to form an initial rule set; Validated through 1000+ clinical cupping cases, invalid associations (such as combinations that did not improve the relief effect after association) were removed, and the rule set was optimized.

[0085] Taking Fengchi acupoint as an example: Locational correlation: 2cm distance from Fengfu acupoint and 3cm distance from Jianjing acupoint.

[0086] Functional association: It works synergistically with Fengfu acupoint to enhance the effect of clearing the meridians in the neck, and works synergistically with Jianjing acupoint to relieve stiffness in the shoulder muscles.

[0087] Meridian Connections: Fengchi (GB20) and Fengfu (GV16) belong to the Governing Vessel and are connected to each other. Fengchi (GB20) and Jianjing (GB21) belong to the Lesser Yang Meridian and are interconnected.

[0088] In this embodiment, features related to the location and function of the region to be refined are extracted from the semantic description and used as associated features. For example, the associated feature of the semantic description that the pain on the left side of the neck is aggravated when looking down is that the pain is aggravated when looking down. Combining this feature, it can be known that the region to be refined is the muscle attachment point region related to neck flexion and extension.

[0089] In this embodiment, a geometric bounding box algorithm is used to initially select the area to be refined based on the approximate location determined by the rules of association between adjacent regions and the range defined by the associated feature items, thereby obtaining a coarse positioning bounding box range. For example, for the left side of the neck to be refined, based on the associated feature item that aggravates soreness when looking down and the rules of association between adjacent regions (related to acupoints of cervical flexion and extension), the skin area corresponding to the C3 to C6 vertebrae on the left side of the neck is coarsely positioned and bounded, with the bounding box boundary being an approximate rectangular outline.

[0090] In this embodiment, contour refinement refers to finely adjusting the boundary of the coarse positioning selection according to the accuracy of the description, so that the selection range more accurately matches the region corresponding to the semantic description. For example, for the neck region to be refined with a description accuracy of 0.88, the boundary of the coarse positioning selection is relatively blurry. According to the refinement requirements corresponding to the high-precision description, the selection boundary is adjusted to accurately select the region of muscle attachment points on the left side of the C4 to C5 vertebrae.

[0091] In this embodiment, the area coverage refers to the precise area range corresponding to the semantic expression, determined after coarse positioning selection and contour refinement.

[0092] In this embodiment, the predefined items refer to the standard items that are pre-set for matching information features under each dimension. For example, the predefined items for the pain perception dimension include pain location, pain level, pain type, and duration.

[0093] In this implementation, the feature-confidence vector refers to the vector containing the matching confidence of each predefined item after matching each information feature under the same dimension with the predefined items of that dimension one by one. Each element of the vector corresponds to the confidence of a predefined item, and its value ranges from 0 to 1. For example, the predefined items of the pain perception dimension are: [pain location, pain intensity, pain type, duration]. If the matching confidence of a certain information feature with each predefined item is 0.95, 0.88, 0.92, and 0.90, respectively, then the feature-confidence vector is [pain location - 0.95, pain intensity - 0.88, pain type - 0.92, duration - 0.90].

[0094] In this embodiment, a separate initial layer refers to an initial graphical layer constructed based on each predefined item, combined with the region coverage and confidence level corresponding to that predefined item in the feature-confidence vector. Each predefined item corresponds to a separate initial layer. For example, in the pain perception dimension, the predefined item of pain location, combined with its region coverage (the left side of the C4 to C5 vertebral bodies) and confidence level of 0.95, constructs a separate initial layer containing the initial points of that region.

[0095] In this embodiment, the dimension attribute refers to the essential characteristics of each dimension. For example, the attribute of the pain perception dimension is symptom association, the attribute of the expected relief effect dimension is effect orientation, the attribute of the physical fitness dimension is basic state, and the attribute of the individual preference dimension is subjective preference.

[0096] In this embodiment, a single rendering process refers to the initial graphic rendering of a single initial layer based on dimensional attributes. Rendering methods include color differentiation, transparency differentiation, etc. For example, the single initial layer of the pain perception dimension (symptom-related attributes) is rendered in red with a transparency of 0.6; the single initial layer of the physical fitness dimension (basic state attributes) is rendered in blue with a transparency of 0.6; the expected relief effect dimension is rendered in green; and the individual preference dimension is rendered in yellow.

[0097] In this embodiment, a confidence threshold (e.g., 0.8) is set, and points with confidence ≥ the threshold in the individual initial layer are selected and determined as additional points. For example, in the individual initial layer of the cupping force predefined item, the core area point corresponding to a confidence of 0.93 is an additional point.

[0098] In this embodiment, secondary rendering refers to the re-rendering of the additional points locked in the individual initial layer. The rendering methods include bolding, highlighting, and different color markings. For example, the additional points in the individual initial layer of the pain perception dimension are rendered with red bolding (line width 0.2mm); the additional points in the physical fitness dimension are rendered with blue highlighting (brightness increased by 50%).

[0099] In this embodiment, a separate rendering layer refers to the graphics layer obtained after one and two rendering processes. This layer clearly presents the area range and significant additional points corresponding to the predefined items, i.e., the additional point map. For example, the separate rendering layer corresponding to the predefined item of pain location in the pain perception dimension uses red to represent the area range, and the red bold dots are additional points. This layer is the additional point map corresponding to the pain location. The separate rendering layer of the predefined item of cupping intensity uses yellow to represent the area, and the blue highlighted dots are additional points. This serves as the additional point map for this predefined item.

[0100] The beneficial effects of the above technical solution are as follows: Through the collaborative work of multiple units, the accurate transformation from user interaction input to additional point maps is achieved: the accuracy of semantic expression is clarified by the expression determination, correlation and coefficient determination, and accuracy determination units; the accurate regional coverage is obtained by the preliminary framing and refinement units; and the additional point maps are obtained by the vector construction and point map acquisition units. The whole process fully explores the core features of the user's multi-dimensional information, effectively handles the accuracy impact caused by fuzzy expressions, and ensures a high degree of matching between the additional point maps and the user's individual information, providing reliable personalized data support for the subsequent fusion of the three-dimensional region contour and the additional point maps and accurate acupoint positioning.

[0101] This invention provides an automatic cupping robot, wherein the point map acquisition unit includes: The array sub-unit is used to determine the confidence level of each initial point in the individual initial layer, obtain the rendering coefficient and point type of each initial point according to the result of a rendering process, and obtain the value array of the corresponding initial point. When the initial point is located in the refined area of ​​the corresponding area coverage, the corresponding point type is regarded as a fine type; otherwise, the corresponding point type is regarded as a coarse type. The function construction subunit is used to perform cluster analysis on all value arrays in the single initial layer with the number of conventional cupping acupoints in the target cupping area as the number of clusters, and to construct a distance trend map of the corresponding cluster analysis results to obtain the concentrated distance density function based on the distance between each core feature in the corresponding dimension information and the cluster feature corresponding to the cluster center cluster of each cluster analysis result. The first point determines the sub-unit, which is used to extract matching features and the first point corresponding to the matching features from all core features based on the central distance density function of each cluster analysis result; The second point determines the sub-unit, which is used to analyze the maximum length step change, average length step change, and minimum length step change of adjacent lengths after being sorted in order of distance length under the lumped density function, and based on the first ratio of the average length step change to the maximum length step change and the second ratio of the minimum length step change to the average length step change. Based on the first ratio and the second ratio, the central point of the concentrated distance density function is corrected to obtain a concentrated representative point as the second point; Significantly additional subunits are used to make the first point and the second point explicit additional contour points in the separate initial layer.

[0102] In this embodiment, the confidence sum refers to the sum of the confidence scores of all relevant elements in the feature-confidence vectors corresponding to each initial point in a single initial layer. For example, if the confidence score of an initial point in a single initial layer is 0.95 with the predefined term of pain location and 0.90 with the predefined term of duration, the confidence sum of the initial point is 0.95 + 0.90 = 1.85.

[0103] In this embodiment, the rendering coefficient refers to the rendering intensity parameter corresponding to each initial point after a rendering process. This parameter is related to the dimension attribute and confidence level, and its value ranges from 0 to 1. It is used to quantify the rendering effect. A rendering coefficient calculation model is established, and the dimension attribute weight, confidence level and so on are input. The rendering coefficient of each initial point is calculated by the algorithm.

[0104] In this embodiment, the array format of the value array is: [confidence score, rendering coefficient, point type code], where the fine type code is 1 and the coarse type code is 0.

[0105] In this embodiment, the number of acupoints for conventional cupping is shown in Table 3: Table 3. Number of acupoints for routine cupping In this embodiment, the number of clusters refers to the number of categories into which the value array is divided during cluster analysis. In this scheme, the number of clusters is set to the number of regular cupping acupoints in the target cupping area. For example, the number of regular cupping acupoints in the target cupping area of ​​the neck is 4, and the number of clusters is 4.

[0106] In this embodiment, the K-Means algorithm is used to divide the value array of all initial points in a single initial layer into multiple clusters according to the number of clusters. Each cluster contains initial points with similar value array characteristics.

[0107] In this embodiment, the information features of each dimension are sorted according to their confidence level, and the information feature with the highest confidence level is selected as the core feature.

[0108] The cluster center cluster refers to the value array corresponding to the center position of each cluster. This array is the mean of the value arrays of all initial points in the cluster, reflecting the overall characteristics of the cluster.

[0109] Cluster features refer to the common features of the value array of all initial points in each cluster. They are the core attributes of the cluster and reflect the commonalities of the initial points within the cluster. For example, if the initial points in a certain cluster are all related to pain on the left side of the neck, its cluster features are left side of the neck, high confidence, and fine type.

[0110] In this embodiment, the distance trend graph refers to a chart that reflects the trend of distance changes, with the distance between the core feature and the cluster feature of the cluster center cluster as the vertical axis and the core feature number as the horizontal axis. For example, the core feature of the pain perception dimension is left neck pain, and the distances between it and the cluster features of the four clusters are 0.1, 0.8, 0.9, and 0.7, respectively. The line graph drawn with the core feature as the horizontal axis and the distance as the vertical axis is the distance trend graph.

[0111] In this embodiment, a kernel density estimation algorithm is used to process the distance data of the distance path graph and construct a concentrated distance density function. For example, the concentrated distance density function corresponding to the distance path graph has the highest density near a distance of 0.1, indicating that the core feature has the highest matching degree with the corresponding cluster.

[0112] In this embodiment, the peak position of the concentrated distance density function is analyzed, and the core features corresponding to the peak are selected and determined as matching features.

[0113] In this embodiment, based on the area coverage and confidence level corresponding to the matching feature, the core point is located in a separate initial layer and determined as the first point. For example, the first point corresponding to the matching feature of pain on the left side of the neck is the core point of the pain area on the left side of the neck (coordinates x1, y1, z1).

[0114] In this embodiment, the adjacent length refers to the length difference between two adjacent distances after sorting the distances between all core features and cluster features involved in the centralized distance density function in ascending order. For example, if the sorted distances are 0.1, 0.3, 0.5, and 0.7, the adjacent lengths are 0.2 (0.3-0.1), 0.2 (0.5-0.3), and 0.2 (0.7-0.5), respectively. All adjacent lengths are compared, and the largest value is selected as the maximum length step change. The sum of all adjacent lengths is calculated and divided by the number of adjacent lengths to obtain the average length step change. All adjacent lengths are then compared, and the smallest value is selected as the minimum length step change.

[0115] In this embodiment, the center point of concentration refers to the center point where the distance distribution in the concentrated distance density function is most concentrated, and it is the initially determined core reference point. For example, if the concentrated distance density function is most concentrated near a distance of 0.1, the corresponding coordinate point is the center point of concentration.

[0116] In this embodiment, a correction model is established, and the first ratio and the second ratio are input to adjust the central point of concentration to obtain the representative point of concentration. For example, the distance between the central points of concentration is 0.1, the first ratio is 0.777, the second ratio is 0.429, and the distance between the representative points of concentration is 0.12 after correction.

[0117] In this embodiment, based on the distance to the representative point and the cluster characteristics, the precise point is located in the initial layer and determined as the second point.

[0118] The beneficial effects of the above technical solution are as follows: the core value of the initial point is quantified by array sub-units, the function construction sub-unit establishes a concentrated distance density function to reflect the feature matching relationship, the first point determination sub-unit accurately extracts the core point position corresponding to the matching feature, the second point determination sub-unit obtains auxiliary point positions through multi-dimensional ratio correction, and finally the explicit additional contour points are obtained by integration by significant additional sub-units. This significantly improves the accuracy and representativeness of the contour points in the additional point map, ensuring that the core information of each dimension can be accurately reflected by the explicit additional contour points, and providing high-quality core point position support for the subsequent fusion of the three-dimensional region contour and the additional point map.

[0119] This invention provides an automatic cupping robot, wherein the fusion module includes: Point marking unit, used to mark the explicit additional contour points in the additional point map of each dimension in the three-dimensional region contour to obtain the additional region contour; The comparison unit is used to compare the outline of the additional area with the outline of the standard human acupoints one by one to obtain the first positioning outline.

[0120] In this embodiment, the additional region contour refers to the integrated contour obtained by accurately marking the explicit additional contour points in the additional point map of all dimensions onto the three-dimensional region contour.

[0121] The beneficial effects of the above technical solution are: it achieves a precise match between standard acupoints and individual physical characteristics, effectively eliminates the deviation of acupoint positions caused by individual differences, and significantly improves the accuracy of acupoint positioning.

[0122] This invention provides an automatic cupping robot, the control module comprising: The effective point determination unit is used to determine effective acupoints based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception. The parameter set determination unit is used to determine the cupping parameter set for each effective acupoint based on the distribution of effective acupoints and the theoretical interconnection effect of acupoints. The scheme correction unit is used to obtain a preliminary scheme based on the cupping parameter set, and to correct it according to the cupping parameter anomaly set of adjacent effective acupoints in the preliminary scheme to obtain the cupping scheme.

[0123] In this embodiment, specifically: Obtain the global acupoints corresponding to the first localization contour, and calculate the comprehensive adaptation weight of each global acupoint i. : ; in, The standard fat thickness threshold corresponding to acupoints in the cupping area of ​​the human body; The estimated fat thickness for global acupoint i; A baseline coefficient is used to adjust the fat thickness, with a value ranging from 0.6 to 0.8. A correction index for fat thickness is used, with a value ranging from 1.2 to 1.5. The degree of fit between the contour curve of the global acupoint i and the local skin contour of the target cupping area, with a value range of 0 to 1; This is the contour fit weighting coefficient, with a value ranging from 0.7 to 0.9; Let be the priority coefficient of the key acupoint corresponding to global acupoint i, and the key acupoint corresponding to is... Non-key acupoints correspond ; This is the priority weighting coefficient for key acupoints, with a value ranging from 0.8 to 1.0. The value ranges from 0 to 1, representing the strong correlation between global acupoint i and the pain perception of the target cupping patient. This is the weighting coefficient for the strong correlation between pain and illness, with a value ranging from 0.9 to 1.1. Filter out the comprehensive adaptation weight The effective acupoints were obtained from the global acupoints, among which, , Preset adaptation weight threshold; Configure a corresponding set of cupping parameters for each effective acupoint j: Calculate the negative pressure value of cupping ,in, This is the baseline negative pressure value for cupping. The minimum fat thickness threshold for acupoints in the cupping area of ​​the human body; The maximum fat thickness threshold for acupoints in the cupping area of ​​the human body; The value is the adjustment coefficient for the effect of fat thickness on negative pressure, ranging from 0.3 to 0.5. The comprehensive adaptation weight for effective acupoint j; For all The maximum value in; Calculate the duration of tank retention ,in, The standard cupping duration is set at 5 to 10 minutes. This represents the minimum value of strong correlation between pain perception and other symptoms. This represents the maximum value of the strong correlation between pain perception and other symptoms. The moderating factor for the effect of the strong correlation between pain and the duration of cupping is 0.4 to 0.6. The degree of fit between the contour curve of the effective acupoint j and the local skin contour of the target cupping area; For all The maximum value in; Determine the type of cupping procedure: Obtain the rate of curvature change of the contour curve of the effective acupoint j. ,like Then select "can moving" as the operation type; if and If the condition is met, then select "keep in container" as the operation type; otherwise, select "flash container" as the operation type. The cupping parameter set corresponding to each effective acupoint is integrated according to the distribution order of acupoints in the target cupping area to obtain the initial scheme; Calculate the difference degree of the cupping parameter anomaly set of adjacent effective acupoints in the initial scheme. If the difference degree is ≥ preset difference threshold, the corresponding cupping parameters are adaptively adjusted based on the weighted average of the comprehensive adaptation weight of adjacent acupoints to obtain the cupping scheme. The cupping parameter anomaly set includes negative pressure difference rate and duration difference rate.

[0124] In this embodiment, an orthogonal experiment was conducted to select 30 subjects with different body constitutions (cold / hot / neutral) and different fat thicknesses (0.5-2.0cm). An L9 orthogonal array was designed, and the accuracy of acupoint location and the pain relief rate were used as evaluation indicators to optimize the range of coefficient values. The value is 0.8. The value is 0.9. The value is 1.0. When the value is 0.65, the acupoint location accuracy reaches ±0.15cm, and the pain relief rate increases by an average of 28%, which is significantly better than the random coefficient value (location accuracy ±0.3cm, relief rate increase of 12%). It should be noted that... , Choose different parameter values ​​according to the scenario, such as cupping for the shoulders and neck: The value is 0.65. The value is 1.4; cupping on the waist: The value is 0.7. The value is 1.3.

[0125] In this embodiment, the distribution of acupoints refers to the spatial distribution of effective acupoints within the target cupping area, including the distance and relative positional relationship between acupoints. For example, the distribution of effective acupoints on the neck, namely Fengchi (GB20), Jianjing (GB21), and Dazhui (GV14), is as follows: Fengchi is located on the upper left side of the neck, Jianjing is located at the junction of the shoulder and neck, and Dazhui is located in the lower center of the neck. The distances between the three acupoints are 3.5cm (Fengchi-Jianjing), 4.0cm (Jianjing-Dazhui), and 5.0cm (Fengchi-Dazhui), respectively.

[0126] In this embodiment, the difference rate = .

[0127] In this embodiment, deviation correction refers to the adaptive adjustment of cupping parameters based on a weighted average of the comprehensive adaptation weights of adjacent acupoints when the difference rate of cupping parameters between adjacent effective acupoints is greater than or equal to a preset threshold (e.g., 15%). For example, if the negative pressure difference rate between a pair of adjacent acupoints is 20% (≥15%), the comprehensive adaptation weight of acupoint A is 0.8, the weight of acupoint B is 0.7, and the weighted average weight is calculated as (0.8 × parameter A + 0.7 × parameter B) / (0.8 + 0.7). Based on this, parameters A and B are adjusted to reduce the difference rate below the threshold.

[0128] In this embodiment, the control module has a built-in acupoint interconnection matrix. After inputting the set of effective acupoints, it automatically queries the quantification value of the synergistic effect between acupoints in the matrix. It prioritizes the selection of acupoint combinations with a quantification value ≥ 0.6 to generate a cupping scheme. The quantification rules are: the synergistic effect quantification value of acupoints on the same meridian is ≥ 0.8; the synergistic effect quantification value of acupoints with interconnected meridians is 0.6-0.8; and the synergistic effect quantification value of acupoints with no directly related acupoints is < 0.6. At this time, the theoretical acupoint interconnection effect can be retrieved from the acupoint interconnection matrix.

[0129] In this embodiment, the cupping parameters are abnormally concentrated, with a negative pressure difference rate threshold of 15% and a duration difference rate threshold of 20%. Let the parameters of adjacent effective acupoints F1 and F2 be as follows: , The overall adaptation weights are respectively , The corrected parameters are: ; .

[0130] The beneficial effects of the above technical solution are as follows: the effective point determination unit selects the most suitable effective acupoints for the user based on multi-dimensional parameters; the parameter set determination unit accurately configures cupping parameters by combining individual characteristics and meridian theory; and the scheme correction unit ensures that the parameters of adjacent acupoints are coordinated and reasonable. Overall, the personalization, precision and coordination of the cupping scheme are realized, which effectively improves the cupping effect on relieving the user's pain, while adapting to the user's individual physical characteristics and reducing the risk of discomfort caused by improper parameters.

[0131] This invention provides an automatic cupping robot, wherein the prevention module includes: The skin vector construction unit is used to extract features from the skin image to obtain the proportion of redness area, skin wrinkling degree and redness texture density of the cupping area, and construct the first input vector; The feedback vector construction unit is used to analyze the semantic negativity and emotional anxiety of cupping feedback information and construct the second input vector. The model analysis unit is used to input the first input vector and the second input vector into the dual-vector analysis model to obtain safety protection measures, and then send them to the corresponding cupping robotic arm to perform control operations.

[0132] In this embodiment, the percentage of reddened area refers to the ratio of the area of ​​reddened skin in the cupping area to the area of ​​the entire shooting area in the skin image, which is used to quantify the degree of skin redness.

[0133] Skin wrinkle degree refers to the density of wrinkles in the cupping area of ​​the skin in the skin image caused by negative pressure adsorption, with a value range of 0-1 (0 for no wrinkles, 1 for extremely dense wrinkles).

[0134] Redness and swelling texture density refers to the density of texture lines in the red and swollen areas of the skin in the cupping area in the skin image. The value ranges from 0 to 1 (0 is no redness and swelling texture, and 1 is extremely dense redness and swelling texture), reflecting the severity of redness and swelling.

[0135] The first input vector refers to the vector formed by combining the proportion of redness area, skin wrinkling degree, and redness texture density in a fixed order. For example, the first input vector after cupping the Fengchi acupoint for 3 minutes is [0.30, 0.2, 0.1].

[0136] In this embodiment, the specific process of semantic negativity parsing is as follows: A negative dictionary for cupping therapy was constructed, containing 100 core negative terms such as "pain, stinging, numbness, too tight, redness and swelling, discomfort", with each term assigned a negative weight (0.1-1.0).

[0137] The user feedback text is segmented, stop words are removed, and it is matched against a negative dictionary to calculate the total weight of negative words.

[0138] Semantic negativity = Sum of negative word weights / Total number of words in the text, with a value range of 0-1.

[0139] In this embodiment, a pre-trained BERT sentiment analysis model is used, which is then fine-tuned to identify the anxiety level of cupping feedback information. The input is the word vector after segmentation of the feedback text; the output is the anxiety quantification value from 0 to 1. For example, if the feedback text is “It hurts so much, I can’t stand it”, the anxiety level is 0.85.

[0140] In this embodiment, the second input vector = {semantic negativity, emotional anxiety}.

[0141] In this embodiment, the construction information of the bi-vector analysis model is as follows: Model structure: It adopts a hybrid architecture of LSTM + fully connected layer, which includes a first input branch (skin state vector), a second input branch (feedback information vector), a fusion layer, two LSTM layers, one fully connected layer and an output layer.

[0142] First input branch: Receives a 3-dimensional first input vector, which is mapped to a 16-dimensional feature vector through a fully connected layer (32 nodes).

[0143] The second input branch receives a 2-dimensional second input vector, which is then mapped to an 8-dimensional feature vector through a fully connected layer (16 nodes).

[0144] Fusion layer: The feature vectors of the two branches are merged by splicing to output a 24-dimensional fusion vector.

[0145] LSTM layer: 2 LSTM layers, 64 nodes per layer, dropout rate = 0.3.

[0146] Output layer: Fully connected layer (8 nodes) + Softmax activation, outputting the probability distribution of 8 security protection measures.

[0147] Training data: Skin condition data: 10,000 skin images during cupping, labeled with the percentage of redness area, skin wrinkling degree, and density of redness and swelling texture, covering different skin types (sensitive / tolerant / neutral).

[0148] Feedback data: 5,000 sets of user voice feedback text, labeled with semantic negativity (0-1) and emotional anxiety (0-1), including three categories of feedback: no discomfort, mild discomfort, and severe discomfort.

[0149] The ratio of training set:validation set:test set is 8:1:1. The annotation was completed by two TCM doctors and one sentiment analysis engineer.

[0150] Training process: Loss function: Cross-entropy loss function.

[0151] Optimizer: SGD optimizer, learning rate = 0.005, momentum = 0.9, weight decay = 0.0001.

[0152] Number of iterations: 200 rounds. Training stops when the accuracy on the validation set is ≥95%.

[0153] Table 4 shows a portion of the mapping table between input vector combinations and protective measures: Table 4. Input Vector Combination and Protection Measures Mapping Table In this embodiment, safety protection measures refer to the operational instructions output by the bi-vector analysis model, used to avoid skin damage or alleviate discomfort for the user. These include adjusting the negative pressure value, shortening / extending the duration, pausing / stopping the operation, and changing the cupping mode. For example, reducing the negative pressure by 0.003 MPa, shortening the cupping time by 2 minutes, pausing the cupping operation for 30 seconds, observing the skin, and changing the cupping mode to flash cupping mode.

[0154] The beneficial effects of the above technical solution are as follows: The skin vector construction unit quantifies the skin state in real time, the feedback vector construction unit accurately analyzes the user's subjective feelings, and the model analysis unit implements targeted safety protection measures based on dual-vector input and output, achieving real-time safety monitoring and dynamic protection during the cupping process. It can promptly capture skin abnormalities and user discomfort, quickly adjust cupping parameters or pause the operation, effectively reducing safety risks such as excessive skin redness and swelling, severe bruising, and increased pain, thus ensuring the safety and user comfort of the cupping process.

[0155] This invention provides an automatic cupping robot, which further includes: The display module is used to display the operation information for each cupping acupoint, including: negative pressure value, cupping duration, and cupping mode.

[0156] The beneficial effects of the above technical solution are: by displaying the core operation information of each cupping acupoint in real time and clearly, the user can understand the key parameters and progress of the cupping process, thereby enhancing the user's right to know and sense of participation.

[0157] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An automatic cupping robot, characterized in that, include: The first acquisition module is used to scan the target cupping area of ​​the target cupping patient to obtain point cloud data, and to reconstruct the three-dimensional region contour from the point cloud data. The second data acquisition module is used to receive the interactive input of the target cupping subject to the cupping questionnaire survey, extract multi-dimensional information from the interactive input, and perform transformation analysis on each dimension of information to obtain an additional point map. The interactive input includes: the target cupping subject's pain perception and expected relief effect on the target cupping area, the target cupping subject's physical condition and individual preferences, and the additional point map is related to the explicit additional contour points of the corresponding dimension information. The fusion module is used to fuse the three-dimensional region contour with the additional point map in each dimension to obtain the additional region contour, and compare and analyze it with the standard human acupoint contour to obtain the first positioning contour. The control module is used to automatically generate a cupping plan based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception, and to control the cupping robotic arm to perform corresponding cupping operations on each cupping acupoint. The cupping operations include: residual cupping, moving cupping and flash cupping. The prevention module is used to control the third acquisition module to acquire skin images of the corresponding cupping acupoints when the cupping robotic arm performs the cupping operation, and to activate the voice interaction device to collect the cupping feedback information of the target cupping user during the cupping process, so as to carry out safety protection.

2. The automatic cupping robot according to claim 1, characterized in that, The first acquisition module includes: An initial construction unit is used to select the cross-sectional direction from the structure-section lookup table according to the skeleton structure of the target cupping area, and select the first data from the point cloud data to construct the initial framework based on the azimuth angle between the cross-sectional direction and the center line of the skeleton structure. The first intermediate unit is used to sequentially supplement the remaining point cloud data into the initial frame according to the scanning path to obtain the first intermediate contour. The second intermediate unit is used to input the point cloud data into a position registration model that matches the skeleton structure of the target cupping area, to obtain a view based on the direction of each section, and to arrange all views in directional order to obtain the second intermediate contour. The line deviation unit is used to obtain the relative path of the center point of the scanning spot based on the center point of the target cupping area during the movement of the center point of the scanning spot along the scanning path, and to obtain the relative deviation line with the scanning path, wherein the thickness of each position point in the relative deviation line is consistent with the deviation magnitude of the corresponding position point. The contour deviation processing unit is used to perform position deviation processing on the second intermediate contour to obtain a three-dimensional region contour based on the first deformation vector of adjacent point cloud data in the same intermediate contour and the second deformation vector of the first intermediate contour and the second intermediate contour based on the same contour position, and in combination with the undulation and jumping process and relative deviation line of the target cupping area.

3. The automatic cupping robot according to claim 1, characterized in that, The second acquisition module includes: The representation determination unit is used to extract all information features of each dimension and obtain the semantic representation of each information feature. The correlation degree and coefficient determination unit is used to match the semantic expression with the physiological region description items in the physiological semantic association library, determine the physiological dimension correlation degree corresponding to the semantic expression, and identify the fuzzy expression items in the semantic expression, and determine the dimension deviation coefficient of the semantic expression based on the fuzziness level of the fuzzy expression items. The accuracy determination unit is used to determine the accuracy of the semantic representation according to the physiological dimension correlation degree and the dimension deviation coefficient. The preliminary delineation unit is used to initially delineate the initial range corresponding to the semantic expression based on the physiological region description item corresponding to the semantic expression and the skeletal structure features of the cupping area of ​​the target cupping patient. At the same time, it marks the area to be refined in the initial range that corresponds to the expression precision. The refinement unit is used to extract the adjacent region association rules of the cupping acupoints corresponding to the region to be refined, and combine the association feature items of the semantic expression to perform coarse positioning selection of the region to be refined, and refine the outline of the coarse selection according to the expression accuracy to obtain the corresponding region coverage. The vector construction unit is used to match each information feature in the same dimension with the predefined terms of the corresponding dimension one by one to obtain the feature-confidence vector of each predefined term; The bitmap acquisition unit is used to construct a separate initial layer based on each predefined item according to the region coverage of each element in each feature-confidence vector and in combination with the corresponding confidence, and to perform a first rendering process on the separate initial layer according to the dimension attribute of the corresponding dimension and a second rendering process on the additional points in the separate initial layer to obtain a separate rendering layer as an additional bitmap.

4. The automatic cupping robot according to claim 3, characterized in that, The point map acquisition unit includes: The array sub-unit is used to determine the confidence level of each initial point in the individual initial layer, obtain the rendering coefficient and point type of each initial point according to the result of a rendering process, and obtain the value array of the corresponding initial point. When the initial point is located in the refined area of ​​the corresponding area coverage, the corresponding point type is regarded as a fine type; otherwise, the corresponding point type is regarded as a coarse type. The function construction subunit is used to perform cluster analysis on all value arrays in the single initial layer with the number of conventional cupping acupoints in the target cupping area as the number of clusters, and to construct a distance trend map of the corresponding cluster analysis results to obtain the concentrated distance density function based on the distance between each core feature in the corresponding dimension information and the cluster feature corresponding to the cluster center cluster of each cluster analysis result. The first point determines the sub-unit, which is used to extract matching features and the first point corresponding to the matching features from all core features based on the central distance density function of each cluster analysis result; The second point determines the sub-unit, which is used to analyze the maximum length step change, average length step change, and minimum length step change of adjacent lengths after being sorted in order of distance length under the lumped density function, and based on the first ratio of the average length step change to the maximum length step change and the second ratio of the minimum length step change to the average length step change. Based on the first ratio and the second ratio, the central point of the concentrated distance density function is corrected to obtain a concentrated representative point as the second point; Significantly additional subunits are used to make the first point and the second point explicit additional contour points in the separate initial layer.

5. The automatic cupping robot according to claim 1, characterized in that, The fusion module includes: Point marking unit, used to mark the explicit additional contour points in the additional point map of each dimension in the three-dimensional region contour to obtain the additional region contour; The comparison unit is used to compare the outline of the additional area with the outline of the standard human acupoints one by one to obtain the first positioning outline.

6. The automatic cupping robot according to claim 1, characterized in that, The control module includes: The effective point determination unit is used to determine effective acupoints based on the global acupoints of the first positioning contour, the estimated fat thickness of each acupoint and the acupoint contour curve, the key acupoints of the target cupping area, and the acupoints that are strongly correlated with pain perception. The parameter set determination unit is used to determine the cupping parameter set for each effective acupoint based on the distribution of effective acupoints and the theoretical interconnection effect of acupoints. The scheme correction unit is used to obtain a preliminary scheme based on the cupping parameter set, and to correct it according to the cupping parameter anomaly set of adjacent effective acupoints in the preliminary scheme to obtain the cupping scheme.

7. The automatic cupping robot according to claim 1, characterized in that, The prevention module includes: The skin vector construction unit is used to extract features from the skin image to obtain the proportion of redness area, skin wrinkling degree and redness texture density of the cupping area, and construct the first input vector; The feedback vector construction unit is used to analyze the semantic negativity and emotional anxiety of cupping feedback information and construct the second input vector. The model analysis unit is used to input the first input vector and the second input vector into the dual-vector analysis model to obtain safety protection measures, and then send them to the corresponding cupping robotic arm to perform control operations.

8. The automatic cupping robot according to claim 1, characterized in that, Also includes: The display module is used to display the operation information for each cupping acupoint, including: negative pressure value, cupping duration, and cupping mode.

Images