Auricular point positioning method, auricular point positioning system and computer program product

Abstract

The invention provides an auricular point positioning method, an auricular point positioning system and a computer program product, and the method comprises the steps: inputting a to-be-recognized 3D human ear model into a trained auricular point partition recognition model, obtaining an auricular point partition result of the to-be-recognized 3D human ear model, and positioning the position of an auricular point in each auricular point partition according to the auricular point partition result. The auricular point partition recognition model can accurately recognize auricular point partitions, the recognized auricular point partitions are displayed to medical workers, the medical workers are assisted in auricular point partition, and the medical workers are helped to learn the ability of accurate auricular point partition. According to priori knowledge or rules, a calculation formula for calculating auricular point positions according to auricular point partitions is designed, so that the auricular point positions in each auricular point partition can be positioned according to auricular point partition results, in the process, the calculation formula for calculating the auricular point positions in the auricular point partitions is displayed to medical staff to assist the medical staff in auricular point positioning, and the auricular point positioning accuracy is improved. And medical personnel are helped to learn the ability of determining the position of the auricular point according to auricular point positioning.

Description

Technical Field

[0001] The present application relates to the field of ear acupoint locating technology, and in particular, to an ear acupoint locating method, an acupoint locating system and a computer program product. Background Art

[0002] With the advancement of science and technology, various testing methods have confirmed the remarkable effect of traditional Chinese medicine in the treatment of chronic diseases and daily health care. Therefore, major hospitals have set up traditional Chinese medicine clinics and chronic disease conditioning departments. In addition to traditional Chinese medicine treatment, there are also many traditional Chinese medicine treatment methods such as moxibustion, acupoint application, ear acupoint bean embedding, acupoint acupuncture, etc. Among these therapies, the difficulty of operating acupoints does not lie in the operation itself, but in accurately identifying acupoints. Especially in the relatively small area of ​​the ear, the acupoints are densely distributed, and even a slight deviation may cause misplacement of pressing. In addition, the ears of different patients have different shapes, which makes it difficult for medical staff who have just been interns or have been employed for a short time to accurately find the corresponding ear acupoints, thus affecting the promotion and popularization of ear acupoint health care and physical therapy.

[0003] In clinical practice, accurate auricular acupoint zoning is a prerequisite for auricular acupoint identification and is crucial for finding the right auricular acupoints. However, existing auricular acupoint locating methods often directly locate auricular acupoints, lack accurate auricular acupoint zoning steps, and are not explanatory enough, making it difficult for medical staff to master effective auricular acupoint locating methods. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an ear acupoint locating method, an acupoint locating system and a computer program product, so as to solve the problem that the existing ear acupoint locating methods often directly locate the ear acupoints, lack accurate ear acupoint zoning steps, and are not interpretive enough, making it difficult for medical staff to master effective ear acupoint locating methods.

[0005] An ear acupoint locating method provided in an embodiment of the present application includes:

[0006] According to the ear acupoint partition training set, the ear acupoint partition recognition model is trained to obtain the trained ear acupoint partition recognition model; the ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint partition labels, and the ear acupoint partition labels include: helix 1 to 12, scaphoid 1 to 6, antihelix 1 to 13, triangular fossa 1 to 5, tragus 1 to 4, antitragus 1 to 4, concha 1 to 18, earlobe 1 to 9, and back of ear 1 to 4;

[0007] Inputting the 3D human ear model to be identified into the trained ear acupoint partition recognition model to obtain the ear acupoint partition result of the 3D human ear model to be identified;

[0008] According to the ear point partitioning results, the positions of the ear points in each ear point partition are located.

[0009] Among them, the 3D human ear model input into the ear acupoint partition recognition model can be the three-dimensional point cloud data of the human ear. The helix is ​​divided into 12 small areas, each of which contains different ear acupoints. The scaphoid is divided into 6 small areas, corresponding to different parts of the upper limbs. The antihelix is ​​divided into 13 small areas, each of which corresponds to different parts of the trunk and lower limbs. The triangular fossa is divided into 5 small areas, each of which contains different ear acupoints. The tragus is divided into 4 small areas, corresponding to acupoints such as the nose, throat and internal nose. The antitragus is divided into 4 small areas, including acupoints such as the forehead, occipital, marginal and brain. The concha is divided into 18 small areas, corresponding to different internal organs and parts. The earlobe is divided into 9 small areas, corresponding to acupoints such as the eye, tongue, front lobe, inner ear and tonsil. The back of the ear is divided into 4 small areas, including acupoints such as the lower foot groove, ear root and upper and lower ear roots.

[0010] According to the results of ear point division, the positions of ear points are located, for example: "Appendix" is at the junction of concha zones 6 and 7. "Ureter" is at the junction of concha zones 9 and 10. "Pancreas and Gallbladder" is at concha zone 11. "Cheek" is at the junction of earlobe zones 5 and 6.

[0011] In the above technical solution, the 3D human ear model to be identified is input into the trained ear acupoint partition recognition model to obtain the ear acupoint partition result of the 3D human ear model to be identified, and then the position of the ear acupoint in each ear acupoint partition is located according to the ear acupoint partition result. The ear acupoint partition recognition model can accurately identify the ear acupoint partitions, display the identified ear acupoint partitions to medical staff, assist medical staff in performing ear acupoint partitioning, and help medical staff learn the ability to accurately perform ear acupoint partitioning. According to prior knowledge or rules, a calculation formula for calculating the ear acupoint position according to the ear acupoint partition is designed, so that the position of the ear acupoint in each ear acupoint partition can be located according to the ear acupoint partition result. In this process, the calculation formula for calculating the ear acupoint position according to the ear acupoint partition is displayed to medical staff, assisting medical staff in performing ear acupoint positioning, and helping medical staff learn the ability to determine the ear acupoint position according to ear acupoint positioning.

[0012] In some optional embodiments, after locating the position of the ear points in each ear point partition according to the ear point partitioning result, it also includes: projecting the 3D human ear model to be identified and the target ear points onto the real human ear; wherein the target ear points are determined by TCM syndrome differentiation diagnosis information and the patient's symptoms.

[0013] In the above technical solution, the TCM syndrome differentiation diagnosis information and the target ear acupoints corresponding to the patient's symptoms can be obtained through the ear acupoint diagnosis database, which includes the TCM syndrome differentiation diagnosis information and the ear acupoints corresponding to the patient's symptoms.

[0014] In order to realize the projection of the ear acupoints on the 3D model onto the real human ear, augmented reality (AR) technology or projection mapping technology may be used.

[0015] Augmented reality technology: By wearing AR glasses or helmets, the 3D human ear model and ear acupoint information are presented to the user in the form of a virtual image. The user can observe the match between the model and the real human ear by adjusting the viewing angle and focal length, and manually or automatically project the ear acupoints onto the real human ear.

[0016] Projection mapping technology: Use professional projection equipment (such as a projector) to project the 3D human ear model and auricular acupoint information onto a real human ear. This requires precise calibration and positioning of the projection equipment in advance to ensure the accuracy and clarity of the projection.

[0017] In some optional embodiments, the method further comprises:

[0018] According to the ear acupoint location training set, the ear acupoint location recognition model is trained to obtain the trained ear acupoint location recognition model; the ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint location labels, and the ear acupoint location labels include 91 different ear acupoints;

[0019] The 3D human ear model to be identified is input into the trained ear acupoint location recognition model to obtain the ear acupoint location of the 3D human ear model to be identified.

[0020] The 3D human ear model input into the ear acupoint positioning and recognition model may be three-dimensional point cloud data of the human ear.

[0021] In some optional implementations, obtaining the 3D human ear model includes:

[0022] Using a depth camera to capture a depth image of a human ear;

[0023] Obtain three-dimensional point cloud data of the human ear according to the depth image;

[0024] Construct a triangular mesh or a quadrilateral mesh based on the 3D point cloud data of the human ear to form a preliminary 3D model of the human ear;

[0025] The preliminary 3D model is smoothed and details enhanced to obtain a 3D human ear model.

[0026] In the above technical solution, a point cloud is generated using depth image data. Each pixel in the depth image corresponds to a point in three-dimensional space. By calculating the coordinates of these points, a point cloud representing the shape of the human ear can be generated. A triangular mesh or a quadrilateral mesh is constructed based on the point cloud data to form a preliminary 3D model of the human ear. This step can be completed using a mesh generation algorithm in computer graphics. The generated mesh model is optimized in detail, including smoothing, detail enhancement and other steps to make the model more realistic and refined. In addition, texture can be added to the 3D model of the human ear by aligning the actual captured color image with the depth image and mapping the color image as a texture onto the 3D model. Appropriate material properties such as reflectivity, glossiness, etc. can also be set for the model to enhance the realism of the model.

[0027] In some optional implementations, obtaining the 3D human ear model includes:

[0028] Using 3D modeling software, the basic geometric bodies are stretched, squeezed, scaled, etc. to form a preliminary human ear model;

[0029] For the human ear model, add the details of the auricle and use Boolean operations, cutting or subdivision surfaces to further refine the shape and structure of the human ear model;

[0030] Set the adjustable parameters of the human ear model and the corresponding adjustment range;

[0031] The adjustable parameters are adjusted using a random algorithm to generate different 3D human ear models.

[0032] In the above technical solution, existing 3D modeling software such as Blender, Maya, 3Ds Max, etc. can be used. Create a basic geometry such as a sphere or a cylinder as the starting point of the ear. Shape it into an ear-like shape by stretching, squeezing, scaling, etc. Add details of the auricle, such as the helix, tragus, antihelix, etc. Use Boolean operations, cutting or subdivision surface techniques to further refine the shape and structure of the ear. Under the premise of maintaining the basic structure of the ear, adjust the vertex position, shape parameters, etc. of the ear through a randomization algorithm. This can be achieved by writing a script, which generates a different random ear model each time it is run. Apply appropriate materials and textures to the ear model to make it look more realistic. You can use an existing material library or create materials and textures yourself. Render the ear model using a 3D rendering engine to view its visual effect. Adjust the material, lighting, and camera settings as needed to obtain the best rendering effect. Test the stability and compatibility of the model to ensure that it can be displayed normally in different 3D environments. When generating a random human ear 3D model, make sure that the structure and proportion of the model conform to the characteristics of a real human ear. Detail is crucial to the realism of the model, so every part of the ear should be carefully portrayed. When introducing randomness, the degree of randomness should be controlled appropriately to avoid generating an overly exaggerated or unnatural ear model.

[0033] In some optional embodiments, obtaining the ear acupoint partitioning training set and / or the ear acupoint positioning training set includes:

[0034] Obtain a 3D human ear model, and annotate each 3D human ear model;

[0035] The annotated 3D ear model is rotated and / or scaled and / or translated and / or bent and / or stretched and / or squeezed and deformed, and then saved as multiple 3D ear models, and a training set is constructed based on these 3D ear models and the corresponding labels.

[0036] In the above technical solution, a large number of 3D human ear models with different shapes can be generated by performing deformation operations such as rotation, scaling, translation, bending, stretching and squeezing on the 3D human ear model. These models, as training data, can significantly improve the recognition ability of the ear acupoint partition recognition model for human ears of different shapes and enhance the generalization performance of the model. Since the training set contains 3D human ear models that have undergone multiple deformation operations, the model will learn how to adapt to human ears of different shapes during the training process, so as to more accurately locate the ear acupoints in practical applications. This helps to improve the accuracy and effect of traditional Chinese medicine physiotherapy. Although it takes a certain amount of time and cost to annotate the original 3D human ear model, by deforming the annotated model to generate multiple training samples, the need for annotating new samples can be reduced to a certain extent, thereby reducing the overall annotation cost. In practical applications, the patient's ear shape may change due to disease, age, genetics and other factors. By including a 3D human ear model with multiple deformation operations in the training set, the model can better adapt to these complex scenarios and improve stability and reliability in practical applications.

[0037] In some optional embodiments, the auricular acupoint partition recognition model includes a deep cascade convolutional neural network;

[0038] The deep cascade convolutional neural network includes input layer, feature extraction layer, cascade layer, pooling layer, fully connected layer and output layer;

[0039] The input layer is used to receive the 3D human ear model; the feature extraction layer is used to extract the local and global features of the 3D human ear model through multi-layer convolution operations; the cascade layer is used to cascade multiple convolution layers; the pooling layer is used to downsample the feature map; the fully connected layer is used to map the extracted features to the corresponding labels; the output layer is used to output the ear acupoint partitioning results.

[0040] In the above technical solution, a deep cascade convolutional neural network is used as the ear acupoint partition recognition model, which can significantly improve the recognition accuracy, generalization performance and processing speed while maintaining high scalability and flexibility.

[0041] In some optional embodiments, the auricular acupoint location recognition model includes a residual neural network;

[0042] The residual neural network includes input layer, convolution layer, batch normalization layer, activation function, residual block, fully connected layer and output layer;

[0043] The input layer is used to receive the 3D human ear model; the convolution layer is used to extract the local features of the 3D human ear model; the batch normalization layer is used to normalize the output of the convolution layer; the activation function is used to increase the nonlinear characteristics of the network; the residual block includes one or more convolution layers, batch normalization layers, activation functions and shortcut connections; the fully connected layer is used to map the extracted features to the corresponding labels; the output layer is used for the ear point positioning results.

[0044] In the above technical solution, the residual neural network is used as the ear acupoint positioning and recognition model, which can significantly improve the performance of the model, accelerate the training process, enhance the generalization ability, and maintain high scalability and flexibility.

[0045] In some optional implementations, the auricular point partition recognition model and the auricular point location recognition model are obtained by training according to a loss function; the loss function is a weighted sum of the first loss function, the second loss function and the third loss function;

[0046] The training of the ear acupoint partition recognition model and the ear acupoint location recognition model includes:

[0047] Inputting the first 3D human ear model into the ear acupoint partition recognition model to obtain a first ear acupoint partition result; locating the position of the ear acupoint in each ear acupoint partition according to the first ear acupoint partition result to obtain first ear acupoint positioning data;

[0048] At the same time, the first 3D human ear model is input into the ear acupoint location recognition model to obtain second ear acupoint location data;

[0049] The first loss function is the mean square error between the predicted first auricular point location data and the actual value;

[0050] The second loss function is the mean square error between the predicted second auricular point location data and the actual value;

[0051] The third loss function is the mean square error between the first ear acupoint location data and the second ear acupoint location data.

[0052] In the above technical solution, the input of the ear point partition recognition model and the ear point positioning recognition model are both 3D human ear models. The ear point partition recognition model and the ear point positioning recognition model are trained as a whole, and the weighted sum of the first loss function, the second loss function and the third loss function is used as the total loss function. This multi-loss function joint optimization method can comprehensively consider errors in different aspects, so that the model can learn the characteristics of ear point partition and positioning more comprehensively during the training process. Through the joint optimization of multiple loss functions, the model can better adapt to 3D human ear models of different shapes, sizes and postures, thereby enhancing the generalization ability of the model.

[0053] In addition, the auricular point partition recognition model uses a deep cascade convolutional neural network, which is good at extracting fine features from complex data. The auricular point location recognition model uses a residual neural network, which effectively alleviates the training difficulties of deep networks through its unique residual structure and improves the generalization performance of the model. The auricular point partition recognition model and the auricular point location recognition model are trained as a whole. By jointly optimizing the loss function, the performance of the two models can be improved simultaneously, reducing the performance bottlenecks that may occur when training separately, such as the reliability and robustness issues that may exist when training a deep cascade convolutional neural network alone.

[0054] An ear acupoint locating system provided in an embodiment of the present application includes: a processor, a network interface, a memory, and a computer program; the computer program is loaded into the memory; and the computer program is executed by the processor to perform any of the above methods when it is executed.

[0055] A computer program product provided in an embodiment of the present application includes a computer program / instruction, which implements the steps of any of the above methods when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 A flowchart of the ear acupoint positioning method provided in the first embodiment of the present application;

[0058] Figure 2 A flowchart of the ear acupoint positioning method provided in the second embodiment of the present application;

[0059] Figure 3 This is a flowchart of the steps of the ear acupoint locating method provided in the third embodiment of the present application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0061] Please refer to Figure 1 , Figure 1 The step flow chart of the ear acupoint locating method provided in the first embodiment of the present application specifically includes:

[0062] Step 110: training an ear acupoint partition recognition model according to an ear acupoint partition training set to obtain a trained ear acupoint partition recognition model; the ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint partition labels, and the ear acupoint partition labels include: helix 1 to 12, scaphoid 1 to 6, antihelix 1 to 13, triangular fossa 1 to 5, tragus 1 to 4, antitragus 1 to 4, concha 1 to 18, earlobe 1 to 9, and back of ear 1 to 4;

[0063] Step 120, inputting the 3D human ear model to be identified into the trained ear acupoint partition recognition model to obtain the ear acupoint partition result of the 3D human ear model to be identified;

[0064] Step 130: locate the position of the ear points in each ear point partition according to the ear point partition result.

[0065] Among them, the 3D human ear model input into the ear acupoint partition recognition model can be the three-dimensional point cloud data of the human ear. The helix is ​​divided into 12 small areas, each of which contains different ear acupoints. The scaphoid is divided into 6 small areas, corresponding to different parts of the upper limbs. The antihelix is ​​divided into 13 small areas, each of which corresponds to different parts of the trunk and lower limbs. The triangular fossa is divided into 5 small areas, each of which contains different ear acupoints. The tragus is divided into 4 small areas, corresponding to acupoints such as the nose, throat and internal nose. The antitragus is divided into 4 small areas, including acupoints such as the forehead, occipital, marginal and brain. The concha is divided into 18 small areas, corresponding to different internal organs and parts. The earlobe is divided into 9 small areas, corresponding to acupoints such as the eye, tongue, front lobe, inner ear and tonsil. The back of the ear is divided into 4 small areas, including acupoints such as the lower foot groove, ear root and upper and lower ear roots.

[0066] According to the results of ear point division, the positions of ear points are located, for example: "Appendix" is at the junction of concha zones 6 and 7. "Ureter" is at the junction of concha zones 9 and 10. "Pancreas and Gallbladder" is at concha zone 11. "Cheek" is at the junction of earlobe zones 5 and 6.

[0067] In an embodiment of the present application, the 3D human ear model to be identified is input into the trained ear acupoint partition recognition model to obtain the ear acupoint partition result of the 3D human ear model to be identified, and then the position of the ear acupoint in each ear acupoint partition is located based on the ear acupoint partition result. The ear acupoint partition recognition model can accurately identify the ear acupoint partitions, display the identified ear acupoint partitions to medical staff, assist medical staff in performing ear acupoint partitioning, and help medical staff learn the ability to accurately perform ear acupoint partitioning. Based on prior knowledge or rules, a calculation formula for calculating the ear acupoint position based on the ear acupoint partition is designed, so that the position of the ear acupoint in each ear acupoint partition can be located based on the ear acupoint partition result. In this process, the calculation formula for calculating the ear acupoint position based on the ear acupoint partition is displayed to medical staff, assisting medical staff in performing ear acupoint positioning, and helping medical staff learn the ability to determine the ear acupoint position based on ear acupoint positioning.

[0068] In some optional implementations, after locating the position of the ear points in each ear point partition according to the ear point partition result, the method further includes:

[0069] Step 140: projecting the 3D human ear model to be identified and the target ear acupoints onto a real human ear; wherein the target ear acupoints are determined by TCM syndrome differentiation diagnosis information and the patient's symptoms.

[0070] In the above technical solution, the TCM syndrome differentiation diagnosis information and the target ear acupoints corresponding to the patient's symptoms can be obtained through the ear acupoint diagnosis database, which includes the TCM syndrome differentiation diagnosis information and the ear acupoints corresponding to the patient's symptoms.

[0071] In order to realize the projection of the ear acupoints on the 3D model onto the real human ear, augmented reality (AR) technology or projection mapping technology may be used.

[0072] Augmented reality technology: By wearing AR glasses or helmets, the 3D human ear model and ear acupoint information are presented to the user in the form of a virtual image. The user can observe the match between the model and the real human ear by adjusting the viewing angle and focal length, and manually or automatically project the ear acupoints onto the real human ear.

[0073] Projection mapping technology: Use professional projection equipment (such as a projector) to project the 3D human ear model and auricular acupoint information onto a real human ear. This requires precise calibration and positioning of the projection equipment in advance to ensure the accuracy and clarity of the projection.

[0074] Please refer to Figure 2 , Figure 2 The step flow chart of the ear acupoint positioning method provided in the second embodiment of the present application specifically includes:

[0075] Step 210: training an ear acupoint location recognition model according to the ear acupoint location training set to obtain a trained ear acupoint location recognition model; the ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint location labels, and the ear acupoint location labels include 91 different ear acupoints;

[0076] Step 220: input the 3D human ear model to be identified into the trained ear acupoint location recognition model to obtain the ear acupoint location of the 3D human ear model to be identified.

[0077] The 3D human ear model input into the ear acupoint positioning and recognition model may be three-dimensional point cloud data of the human ear.

[0078] Step 230: project the 3D human ear model to be identified and the target ear acupoints onto the real human ear; wherein the target ear acupoints are determined by TCM syndrome differentiation diagnosis information and the patient's symptoms.

[0079] Please refer to Figure 3 , Figure 3The step flow chart of the ear acupoint locating method provided for the third embodiment of the present application specifically includes: obtaining a 3D human ear model, and marking the 3D human ear model with ear acupoint partition labels and ear acupoint locating labels; after marking, performing operations such as rotating, scaling, translating, bending, stretching, and squeezing the 3D human ear model, and saving one or more deformed 3D human ear models for each 3D human ear model, and the deformed 3D human ear model still has ear acupoint partition labels and ear acupoint locating labels to achieve sample augmentation; after sample augmentation, according to the 3D human ear model and the ear acupoint partition labels and ear acupoint locating labels, the ear acupoint partition recognition model and the ear acupoint locating recognition model are trained at the same time. Specifically, in one training process, the first 3D human ear model is simultaneously input into the ear acupoint partition recognition model and the ear acupoint locating recognition model. The following is a detailed description of each process:

[0080] The acquisition of the 3D human ear model includes: using a depth camera to capture a depth image of the human ear; obtaining three-dimensional point cloud data of the human ear based on the depth image; constructing a triangular mesh or a quadrilateral mesh based on the three-dimensional point cloud data of the human ear to form a preliminary 3D model of the human ear; and smoothing and enhancing the details of the preliminary 3D model to obtain a 3D human ear model.

[0081] In an embodiment of the present application, a point cloud is generated using depth image data. Each pixel in the depth image corresponds to a point in a three-dimensional space, and by calculating the coordinates of these points, a point cloud representing the shape of the human ear can be generated. A triangular mesh or a quadrilateral mesh is constructed based on the point cloud data to form a preliminary 3D model of the human ear. This step can be completed using a mesh generation algorithm in computer graphics. The generated mesh model is optimized in detail, including smoothing, detail enhancement and other steps to make the model more realistic and refined. In addition, texture can be added to the 3D model of the human ear by aligning the actual captured color image with the depth image and mapping the color image as a texture onto the 3D model. Appropriate material properties such as reflectivity, glossiness, etc. can also be set for the model to enhance the realism of the model.

[0082] Acquisition of a 3D human ear model also includes: using 3D modeling software to form a preliminary human ear model by stretching, squeezing, scaling and other operations on basic geometric bodies; adding auricle details to the human ear model, and using Boolean operations, cutting or subdividing surfaces to further refine the shape and structure of the human ear model; setting adjustable parameters of the human ear model and the corresponding adjustment ranges; and using a random algorithm to adjust the adjustable parameters to generate different 3D human ear models.

[0083] In the embodiment of the present application, existing 3D modeling software such as Blender, Maya, 3Ds Max, etc. can be used. Create a basic geometric body, such as a sphere or a cylinder, as the starting point of the ear. Shape it into an ear-like shape by stretching, squeezing, scaling, etc. Add details of the auricle, such as the helix, tragus, antihelix, etc. Use Boolean operations, cutting or subdivision surface techniques to further refine the shape and structure of the ear. Under the premise of maintaining the basic structure of the ear, adjust the vertex position, shape parameters, etc. of the ear through a randomization algorithm. This can be achieved by writing a script, and the script will generate a different random ear model each time it runs. Apply appropriate materials and textures to the ear model to make it look more realistic. You can use an existing material library or create materials and textures yourself. Use a 3D rendering engine to render the ear model and view its visual effects. Adjust the material, lighting, and camera settings as needed to obtain the best rendering effect. Test the stability and compatibility of the model to ensure that it can be displayed normally in different 3D environments. When generating a random human ear 3D model, make sure that the structure and proportion of the model conform to the characteristics of a real human ear. Detail is crucial to the realism of the model, so every part of the ear should be carefully portrayed. When introducing randomness, the degree of randomness should be controlled appropriately to avoid generating an overly exaggerated or unnatural ear model.

[0084] Acquisition of ear acupoint partitioning training sets and / or ear acupoint positioning training sets includes: acquiring 3D human ear models and labeling each 3D human ear model; rotating and / or scaling and / or translating and / or bending and / or stretching and / or squeezing the labeled 3D human ear models and saving them as multiple 3D human ear models, and constructing a training set based on these 3D human ear models and corresponding labels.

[0085] In the embodiment of the present application, a large number of 3D human ear models with different shapes can be generated by performing deformation operations such as rotation, scaling, translation, bending, stretching and squeezing on the 3D human ear model. These models, as training data, can significantly improve the recognition ability of the ear acupoint partition recognition model for human ears of different shapes and enhance the generalization performance of the model. Since the training set contains 3D human ear models that have undergone multiple deformation operations, the model will learn how to adapt to human ears of different shapes during the training process, so as to more accurately locate the ear acupoints in practical applications. This helps to improve the accuracy and effect of Chinese medicine physiotherapy. Although it takes a certain amount of time and cost to mark the original 3D human ear model, by performing deformation operations on the marked model to generate multiple training samples, the demand for marking new samples can be reduced to a certain extent, thereby reducing the overall marking cost. In practical applications, the ear shape of the patient may change due to factors such as disease, age, and genetics. By including a 3D human ear model with multiple deformation operations in the training set, the model can better adapt to these complex scenes and improve stability and reliability in practical applications.

[0086] In some optional embodiments, the ear acupoint partition recognition model includes a deep cascaded convolutional neural network; the deep cascaded convolutional neural network includes an input layer, a feature extraction layer, a cascade layer, a pooling layer, a fully connected layer and an output layer; the input layer is used to receive a 3D human ear model; the feature extraction layer is used to extract local and global features of the 3D human ear model through multi-layer convolution operations; the cascade layer is used to cascade multiple convolution layers; the pooling layer is used to downsample the feature map; the fully connected layer is used to map the extracted features to corresponding labels; the output layer is used to output the ear acupoint partition results.

[0087] In the embodiments of the present application, a deep cascade convolutional neural network is used as an ear acupoint partition recognition model, which can significantly improve the recognition accuracy, generalization performance and processing speed while maintaining a high degree of scalability and flexibility.

[0088] In some optional embodiments, the ear acupoint positioning and recognition model includes a residual neural network; the residual neural network includes an input layer, a convolution layer, a batch normalization layer, an activation function, a residual block, a fully connected layer and an output layer; the input layer is used to receive a 3D human ear model; the convolution layer is used to extract local features of the 3D human ear model; the batch normalization layer is used to normalize the output of the convolution layer; the activation function is used to increase the nonlinear characteristics of the network; the residual block includes one or more convolution layers, batch normalization layers, activation functions and shortcut connections; the fully connected layer is used to map the extracted features to corresponding labels; the output layer is used for the ear acupoint positioning results.

[0089] In the embodiments of the present application, a residual neural network is used as an ear acupoint location recognition model, which can significantly improve the performance of the model, accelerate the training process, enhance the generalization ability, and maintain a high degree of scalability and flexibility.

[0090] In some optional implementations, the auricular point partition recognition model and the auricular point location recognition model are obtained by training according to a loss function; the loss function is a weighted sum of the first loss function, the second loss function and the third loss function;

[0091] The training of the ear acupoint partition recognition model and the ear acupoint location recognition model includes: inputting the first 3D human ear model into the ear acupoint partition recognition model to obtain a first ear acupoint partition result; locating the position of the ear acupoint in each ear acupoint partition according to the first ear acupoint partition result to obtain first ear acupoint location data; at the same time, inputting the first 3D human ear model into the ear acupoint location recognition model to obtain second ear acupoint location data;

[0092] Among them, the first loss function is the mean square error between the predicted first ear acupoint positioning data and the actual value; the second loss function is the mean square error between the predicted second ear acupoint positioning data and the actual value; the third loss function is the mean square error between the first ear acupoint positioning data and the second ear acupoint positioning data.

[0093] In the embodiment of the present application, the input of the ear point partition recognition model and the ear point positioning recognition model are both 3D human ear models, and the ear point partition recognition model and the ear point positioning recognition model are trained as a whole, and the weighted sum of the first loss function, the second loss function and the third loss function is used as the total loss function. This multi-loss function joint optimization method can comprehensively consider errors in different aspects, so that the model can learn the characteristics of ear point partition and positioning more comprehensively during the training process. Through the joint optimization of multiple loss functions, the model can better adapt to 3D human ear models of different shapes, sizes and postures, thereby enhancing the generalization ability of the model.

[0094] After obtaining the trained ear acupoint partition recognition model and ear acupoint location recognition model, the 3D human ear model to be recognized is input into the trained ear acupoint location recognition model or the ear acupoint location recognition model to obtain the ear acupoint location of the 3D human ear model to be recognized. The 3D human ear model input into the ear acupoint location recognition model may be three-dimensional point cloud data of a human ear.

[0095] Finally, the 3D human ear model and target ear acupoints to be identified are projected onto the real human ear; the target ear acupoints are determined by TCM syndrome differentiation diagnosis information and the patient's symptoms.

[0096] In addition, the auricular point partition recognition model uses a deep cascade convolutional neural network, which is good at extracting fine features from complex data. The auricular point location recognition model uses a residual neural network, which effectively alleviates the training difficulties of deep networks through its unique residual structure and improves the generalization performance of the model. The auricular point partition recognition model and the auricular point location recognition model are trained as a whole. By jointly optimizing the loss function, the performance of the two models can be improved simultaneously, reducing the performance bottlenecks that may occur when training separately, such as the reliability and robustness issues that may exist when training a deep cascade convolutional neural network alone.

[0097] An ear acupoint locating system provided in an embodiment of the present application includes: a processor, a network interface, a memory, and a computer program; the computer program is loaded into the memory; and the computer program is executed by the processor to perform any of the above methods when it is executed.

[0098] A computer program product provided in an embodiment of the present application includes a computer program / instruction, which implements the steps of any of the above methods when executed by a processor.

[0099] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0100] In addition, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0101] Furthermore, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0102] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0103] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for locating ear points, characterized in that: include: According to the ear acupoint partition training set, the ear acupoint partition recognition model is trained to obtain a trained ear acupoint partition recognition model; The ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint partition labels, wherein the ear acupoint partition labels include: helix 1 to 12, scaphoid 1 to 6, antihelix 1 to 13, triangular fossa 1 to 5, tragus 1 to 4, antitragus 1 to 4, concha 1 to 18, earlobe 1 to 9, and back of ear 1 to 4; Inputting the 3D human ear model to be identified into the trained ear acupoint partition recognition model to obtain the ear acupoint partition result of the 3D human ear model to be identified; According to the ear point partitioning results, the positions of the ear points in each ear point partition are located.

2. The method according to claim 1, characterized in that After locating the position of the ear points in each ear point partition according to the ear point partition result, the method further includes: The 3D human ear model to be identified and the target ear acupoints are projected onto a real human ear; wherein the target ear acupoints are determined by TCM syndrome differentiation diagnosis information and the patient's symptoms.

3. The method according to claim 1, characterized in that Also includes: According to the ear acupoint location training set, an ear acupoint location recognition model is trained to obtain a trained ear acupoint location recognition model; the ear acupoint partition training set includes a 3D human ear model and corresponding ear acupoint location labels, and the ear acupoint location labels include 91 different ear acupoints; The 3D human ear model to be identified is input into the trained ear acupoint location recognition model to obtain the ear acupoint location of the 3D human ear model to be identified.

4. The method according to claim 3, characterized in that The acquisition of the 3D human ear model includes: Using a depth camera to capture a depth image of a human ear; Obtain three-dimensional point cloud data of the human ear according to the depth image; Construct a triangular mesh or a quadrilateral mesh based on the 3D point cloud data of the human ear to form a preliminary 3D model of the human ear; The preliminary 3D model is smoothed and detail enhanced to obtain a 3D human ear model.

5. The method according to claim 3, characterized in that The acquisition of the 3D human ear model includes: Using 3D modeling software, the basic geometric bodies are stretched, squeezed, and scaled to form a preliminary human ear model; For the human ear model, add the details of the auricle and use Boolean operations, cutting or subdivision surfaces to further refine the shape and structure of the human ear model; Set the adjustable parameters of the human ear model and the corresponding adjustment range; The adjustable parameters are adjusted using a random algorithm to generate different 3D human ear models.

6. The method according to claim 3, characterized in that The acquisition of the ear acupoint partitioning training set and / or the ear acupoint positioning training set includes: Obtain a 3D human ear model, and annotate each 3D human ear model; The annotated 3D ear model is rotated and / or scaled and / or translated and / or bent and / or stretched and / or squeezed and deformed, and then saved as multiple 3D ear models, and a training set is constructed based on these 3D ear models and the corresponding labels.

7. The method according to claim 6, characterized in that The auricular acupoint partition recognition model includes a deep cascade convolutional neural network; The deep cascade convolutional neural network includes an input layer, a feature extraction layer, a cascade layer, a pooling layer, a fully connected layer and an output layer; The input layer is used to receive the 3D human ear model; the feature extraction layer is used to extract local and global features of the 3D human ear model through multi-layer convolution operations; the cascade layer is used to cascade multiple convolution layers; the pooling layer is used to downsample the feature map; the fully connected layer is used to map the extracted features to corresponding labels; and the output layer is used to output the ear acupoint partitioning results.

8. The method according to claim 6, characterized in that The ear point location recognition model includes a residual neural network; The residual neural network includes an input layer, a convolutional layer, a batch normalization layer, an activation function, a residual block, a fully connected layer and an output layer; The input layer is used to receive the 3D human ear model; the convolution layer is used to extract local features of the 3D human ear model; the batch normalization layer is used to normalize the output of the convolution layer; the activation function is used to increase the nonlinear characteristics of the network; the residual block includes one or more convolution layers, batch normalization layers, activation functions and shortcut connections; the fully connected layer is used to map the extracted features to corresponding labels; the output layer is used for the ear point positioning results.

9. The method according to claim 6, characterized in that The ear point partition recognition model and the ear point location recognition model are obtained by training according to a loss function; the loss function is a weighted sum of a first loss function, a second loss function and a third loss function; The training of the ear acupoint partition recognition model and the ear acupoint location recognition model includes: Inputting the first 3D human ear model into the ear acupoint partition recognition model to obtain a first ear acupoint partition result; According to the first ear point partitioning result, the position of the ear point in each ear point partition is located to obtain the first ear point positioning data; At the same time, the first 3D human ear model is input into the ear acupoint location recognition model to obtain second ear acupoint location data; The first loss function is the mean square error between the predicted first auricular point location data and the actual value; The second loss function is the mean square error between the predicted second auricular point location data and the actual value; The third loss function is the mean square error between the first ear acupoint location data and the second ear acupoint location data.

10. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 9.

Images