Human body back acupuncture point positioning method, device and equipment

Abstract

The invention relates to the technical field of human body acupoint recognition, and discloses a human body back acupoint positioning method, device and equipment. The method comprises the steps of obtaining a back image of a target object; generating a back acupoint image based on the back image and a target model; each back acupoint of the target object is marked in the back acupoint image; and projecting the back acupuncture point image to the back of the target object through projection equipment so as to position the position of each back acupuncture point of the target object. The back acupuncture points of the target object are recognized and accurately projected and positioned to the back of the target object through the combination of the back image and the target model obtained through pre-training, and compared with existing manual acupuncture point positioning of medical staff, the subjective influence of the medical staff can be avoided, and then the accuracy of back acupuncture point recognition and positioning is improved.

Description

Technical Field

[0001] The present application relates to the technical field of human acupoint identification, and in particular to a method, device and equipment for locating acupoints on the back of the human body. Background Art

[0002] In Traditional Chinese Medicine (TCM), acupoints are specific points along the meridians, recognized as key nodes where qi and blood converge in the body. Acupoint-based treatments are often focused on the back. However, current acupoint location on the back relies heavily on the expertise and clinical experience of medical personnel. Manual acupoint location is highly subjective, and the expertise and clinical experience vary widely, resulting in inaccurate acupoint location on the back. Summary of the Invention

[0003] In view of this, the purpose of this application is to overcome the deficiencies in the prior art and to provide a method, device and equipment for locating acupuncture points on the back of the human body.

[0004] An embodiment of the present application provides a method for locating acupuncture points on the back of a human body, the method comprising:

[0005] Acquire a back image of the target object;

[0006] generating a back acupoint image based on the back image and the target model; wherein each back acupoint of the target object is marked in the back acupoint image;

[0007] The back acupuncture point image is projected onto the back of the target object by a projection device to locate the position of each back acupuncture point of the target object.

[0008] The embodiment of the present application further provides a device for locating acupoints on the back of a human body, the device comprising:

[0009] An acquisition module, used for acquiring a back image of a target object;

[0010] A generating module, configured to generate a back acupoint image based on the back image and the target model; the back acupoint image is marked with each back acupoint of the target object;

[0011] A positioning module is used to project the back acupuncture point image onto the back of the target object based on a projection device to locate the position of each back acupuncture point of the target object.

[0012] An embodiment of the present application further provides a computer device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the steps of a method for locating acupuncture points on the back of a human body.

[0013] An embodiment of the present application further provides a computer storage medium storing a computer program, which, when executed on a processor, implements the steps of the method for locating acupuncture points on the back of the human body.

[0014] The embodiments of the present application have the following beneficial effects:

[0015] The method for locating acupuncture points on the back of a human body according to an embodiment of the present application obtains a back image of a target object; generates a back acupuncture point image based on the back image and a target model; the back acupuncture point image is annotated with each back acupuncture point of the target object; and the back acupuncture point image is projected onto the back of the target object via a projection device to locate the position of each back acupuncture point of the target object. By combining the back image with a target model obtained by pre-training, the back acupuncture points of the target object are identified and accurately projected onto the back of the target object. Compared with the existing manual acupuncture point positioning method of medical personnel, the method can avoid the subjective influence of medical personnel, thereby improving the accuracy of back acupuncture point identification and positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the technical solution of this application, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of this application and should not be considered as limiting the scope of protection of this application. Those skilled in the art can also derive other relevant drawings based on these drawings without inventive effort.

[0017] Figure 1 A schematic diagram of the process of the first embodiment of the present application is shown;

[0018] Figure 2 A schematic diagram of the process of the second embodiment of the present application is shown;

[0019] Figure 3 A schematic diagram showing an image of basic acupuncture points on the back of the present application is shown;

[0020] Figure 4 A schematic diagram showing an image of acupuncture points on the back of the present application is shown;

[0021] Figure 5 A schematic diagram of the process of the third embodiment of the present application is shown;

[0022] Figure 6 Shows a schematic structural diagram of the acupoint recognition model of the present application;

[0023] Figure 7 The schematic diagram of the structure of the position fusion attention mechanism module of this application is shown;

[0024] Figure 8 A schematic diagram of the process of the fourth embodiment of the present application is shown;

[0025] Figure 9 A schematic diagram of the process of the fifth embodiment of the present application is shown;

[0026] Figure 10 A schematic diagram showing the process of extracting acromion position information of the present application is shown;

[0027] Figure 11 A flow chart of the sixth embodiment of the present application is shown. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0029] The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0030] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present application, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or the possibility of adding one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0031] Furthermore, the terms “first,” “second,” “third,” etc., are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0032] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as in the context of the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.

[0033] It is understandable that the method of the present application is applied to an acupoint positioning device, which may be a smart phone, a personal computer, a server, or a network device, etc., and is not limited here.

[0034] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0035] Please refer to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the present application. The method for locating acupuncture points on the back of the human body includes the following steps:

[0036] Step S101, obtaining a back image of a target object;

[0037] In this embodiment, the acupoint locating device can capture the back image of the target object based on its own camera module or projection module; the acupoint locating device can also directly receive the back image of the target object uploaded by the user. The target object is usually a human body.

[0038] Step S102: generating a back acupoint image based on the back image and the target model; each back acupoint of the target object is marked in the back acupoint image.

[0039] In this embodiment, after the acupoint locating device acquires a back image of the target object, it analyzes the back image based on the target model, and then labels each back acupoint of the target object in the back image, generating a back acupoint image. Optionally, the target model is a pre-trained acupoint recognition model. The acupoint locating device inputs the back image into the acupoint recognition model. The acupoint recognition model analyzes the back image to obtain the position of each back acupoint on the back image, and then labels each back acupoint of the target object in the back image, and outputs a back acupoint image. Optionally, the target model is a pre-trained image processing model. The acupoint locating device inputs the back image into the image processing model. The image processing model identifies the spine and acromion in the back image, and then infers the position of each back acupoint on the back image based on the position of the spine and acromion. Then, each back acupoint of the target object is labeled in the back image, and outputs a back acupoint image.

[0040] Step S103 : Projecting the back acupuncture point image onto the back of the target object by a projection device to locate the position of each back acupuncture point of the target object.

[0041] In this embodiment, the acupoint locating device includes a projection device, and the acupoint locating device projects a back acupoint image onto the back of a target subject via the projection device to locate the position of each back acupoint of the target subject on the back of the target subject. It is understood that the acupoint locating device can adjust the size and position of the back acupoint image based on the projection parameters of the projection device, the distance between the projection device and the back of the target subject, and the size information of the back of the target subject, so that the back acupoint projection image projected by the projection device onto the back of the target subject can overlap with the back of the target subject, thereby accurately locating the position of each back acupoint of the target subject. Furthermore, the projection device can adjust the size and position of the back acupoint image in response to a manual adjustment operation instruction from a user, so that the back acupoint projection image projected by the projection device onto the back of the target subject can overlap with the back of the target subject, thereby accurately locating the position of each back acupoint of the target subject.

[0042] The acupoint locating device of this embodiment analyzes a back image using a target model to improve the accuracy of locating acupoints on the back of a person. Projecting the back acupoint image onto the back of a target subject using a projection device can improve the accuracy of locating each acupoint on the target subject's back. This eliminates the need for medical personnel performing acupoint treatment to manually locate acupoints, allowing them to quickly and accurately locate acupoints on the target subject's back.

[0043] Please refer to Figure 2 , Figure 2 This is a flow chart of the second embodiment of the present application. The difference between the second embodiment and the first embodiment is that the target model is an acupoint recognition model. The step of generating a back acupoint image based on the back image and the target model includes:

[0044] Step S201: input the back image into the acupoint recognition model, and output a back basic acupoint image.

[0045] In this embodiment, the target model is a pre-trained acupoint recognition model. The acupoint positioning device inputs a back image into the acupoint recognition model. The acupoint recognition model analyzes the back image to determine the location of each basic back acupoint on the back image. The model then labels each basic back acupoint of the target object in the back image and outputs a basic back acupoint image. It should be noted that the basic back acupoints are those marked on the sample back acupoint images used by the acupoint recognition model during training. Since there are 141 acupoints on the human back, if all acupoints were marked on the sample back acupoint images, the excessive amount of data would result in a decrease in model recognition performance. Therefore, setting a certain number of basic back acupoints for recognition by the acupoint recognition model can improve the model's recognition performance.

[0046] For example, Figure 3 As shown, Figure 3 : This is a schematic diagram of the basic acupoints on the back in this embodiment. The number of basic acupoints on the back is set to 49, namely: right acromion, left acromion, right axilla, left axilla, right blood pressure point, left blood pressure point, right shoulder well, left shoulder well, right Tianzong, left Tianzong, right Jiaji, left Jiaji, right Tianliao, left Tianliao, right Fufen, left Fufen, Dazhui, right Tianzong, left Tianzong, right Chengfeng, left Chengfeng, right Jingmen, left Jingmen, right shoulder outer shu, left shoulder outer shu, right waist eye, left waist eye, right shoulder zhen, left shoulder zhen, right Jianshu, left Jianshu, right Quyuan, left Quyuan, right Pigen, left Pigen, right shoulder back, left shoulder back, Taodao, Shenzhu, Shendao, Lingtai, Zhiyang, Jinsuo, Zhongshu, Zhongzhong, Xuanshu, Mingmen, and Yangguan.

[0047] Step S202: generating a back acupoint image based on the back basic acupoint image and a preset acupoint distribution rule.

[0048] In this embodiment, after obtaining the basic acupoint image of the back, the acupoint positioning device generates a back acupoint image based on the basic acupoint image of the back and the preset acupoint distribution rules. Specifically, according to the bone measurement method in traditional Chinese medicine for acupoint search, the distance between the left and right acromion is 16 inches according to the human body proportions; the acupoint positioning device determines the left acromion coordinates and the right acromion coordinates based on the basic acupoint image of the back, and obtains the difference between the pixel horizontal coordinates of the left acromion coordinates and the pixel horizontal coordinates of the right acromion coordinates. The difference is divided by 16 to obtain the pixel distance corresponding to "one inch" in the basic acupoint image of the back; the acupoint positioning device uses this pixel distance as a basis, combined with the preset acupoint distribution rules, to obtain the position of each back inference acupoint, and then marks each back inference acupoint on the basic acupoint image of the back to obtain a back acupoint image. Exemplarily, the back acupoint image is as follows Figure 4 shown.

[0049] For example, taking the inference of "Dingchuan Point" based on the basic acupuncture point "Dazhui Point" as an example, the pixel coordinates of "Dazhui Point" in the basic acupuncture point image of the back are (X, Y). The acupuncture point positioning device combines the pixel distance S corresponding to "one inch" in the basic acupuncture point image of the back with the preset acupuncture point distribution rule that "Dazhui Point is located on the posterior midline of the human body, in the depression under the spinous process of the seventh vertebra; and Dingchuan Point is located under the spinous process of the seventh cervical vertebra, 0.5 inches to the left and right of the posterior midline". It can be inferred that the pixel coordinates of "Left Dingchuan Point" in the basic acupuncture point image of the back are (X-0.5S, Y), and the pixel coordinates of "Right Dingchuan Point" in the basic acupuncture point image of the back are (X+0.5S, Y). Taking the inference of "Jian Zhongshu" based on the basic acupuncture point "Dazhui" as an example, the preset acupuncture point distribution rule records that "the right Jian Zhongshu is 2 inches to the right of Dazhui, and the left Jian Zhongshu is 2 inches to the left of Dazhui". It can be inferred that the pixel coordinates of "Left Jian Zhongshu" in the basic acupuncture point image on the back are (X+2S, Y), and the pixel coordinates of "Right Jian Zhongshu" in the basic acupuncture point image on the back are (X-2S, Y).

[0050] The acupoint locating device in this embodiment only identifies a certain number of basic acupoints during the acupoint recognition model phase. It then uses pre-defined acupoint distribution rules to infer and locate the remaining back acupoints. This layered processing strategy ensures accurate acupoint detection while addressing the performance degradation of the degree recognition model when the number of acupoints increases dramatically.

[0051] Please refer to Figure 5 , Figure 5 This is a flow chart of the third embodiment of the present application. The difference between the third embodiment and the first to second embodiments is that the step of inputting the back image into the acupoint recognition model and outputting the back basic acupoint image includes:

[0052] Step S301: input the back image into the acupoint recognition model, process the back image through the backbone network in the acupoint recognition model, and obtain a pooled feature map corresponding to the back image.

[0053] In this embodiment, it should be noted that Figure 6 As shown in the figure, the structure of the acupoint recognition model is constructed by adding the Non-Local attention mechanism to the backbone network of the network based on YOLOv8-Pose; among them, the main structure of the acupoint recognition model consists of three parts: Backbone, Neck, and Head.

[0054] Backbone is the backbone network of the model, which pre-trains and extracts features from the input image through the CBS, C2F, and SPPF modules. The CBS module is a combination of convolution, batch normalization, and activation functions. The CBS module first extracts features through convolution, then performs batch normalization to accelerate the training process of the neural network. Finally, the SILU activation function is applied to improve the network's nonlinear expression capabilities. The C2F module first performs preliminary processing on the input feature map through labeled convolution. The feature map is then split into two parts. One part directly retains the original features, and the other part is processed step by step through N Botteleneck modules. Each Botteleneck module contains two CBS modules and a Concat module. After the Bottleneck modules process, the feature flow is decomposed into two paths. One path is responsible for transforming the features and passing them to the next Bottleneck, while the other path directly retains the current features for subsequent feature splicing. After N Botteleneck processing, the features in all paths are merged. The SPPF module enhances the network's receptive field by using multi-scale pooling operations on the input feature map, enabling the model to better handle different targets in the detection image without increasing the computational cost; the specific operation steps are as follows: after entering the SPPF module, the input features are first processed by a CBS module into two paths, one path retains the direct features and enters the Concat splicing operation; the other path downsamples the processed features through three maximum pooling layers in sequence, and the features after each maximum pooling are fed into the Concat module for fusion. The fully spliced features are operated by a CBS module to generate the final output result.

[0055] During the detection of acupoints on the human back, there are relative dependencies between them. To better extract the correlations between different acupoints, the acupoint recognition model incorporates a Non-local module, a position fusion attention mechanism, into the Backbone network. This module is placed after the spatial pyramid pooling (SPPF). The Non-local module captures global features by calculating the similarity between any two locations in the input feature map. This mechanism dynamically adjusts the feature representation of each location, allowing the model to extract local features while incorporating global contextual information.

[0056] Neck is the model's neck network. It fuses and further processes features extracted from the Backbone using the Concatenation to Fractionation (C2F), Concat, and Upsample modules. Concat combines features from different layers, while Upsample restores image resolution, helping the model generate clearer predictions.

[0057] Head is the head network of the model. Two CBS modules process the fused features, and finally the data is convolved by the Conv2D layer and the results are output.

[0058] Specifically, the acupoint positioning device inputs the back image into the acupoint recognition model, performs feature extraction processing on the back image through the CBS, C2F, and SPPF modules in the backbone network of the acupoint recognition model, and obtains a pooled feature map corresponding to the back image.

[0059] Step S302: Process the pooled feature map through the position fusion attention mechanism module in the acupoint recognition model to obtain a target feature map.

[0060] In this embodiment, after the acupoint positioning device outputs the pooled feature map corresponding to the back image through the SPPF module, the pooled feature map is input into the position fusion attention mechanism module, and the pooled feature map is processed based on the position fusion attention mechanism module to obtain the target feature map.

[0061] It should be noted that the structure of the position fusion attention mechanism module is as follows Figure 7 As shown, when the pooling feature Figure X Input to the Non-local attention mechanism for linear mapping, where T is the number of time steps, H and W are the height and width of the feature, and 1024 is the number of channels. Pooling features Figure X The channel is compressed by 1×1×1 convolution to generate three transition features θ, φ and g, with a channel number of 512. The two features θ and φ are then matrix-dot-producted to obtain a similarity matrix THW×THW to calculate the autocorrelation between the two features. The similarity matrix is then normalized using Softmax to convert it into a weight matrix. Each element in the matrix represents the similarity weight of a position relative to other positions, with a value between 0 and 1. The weight matrix is multiplied by the feature g to obtain a transition feature map y of T×H×W×512. i Finally, a 1×1 convolution is used to restore the number of channels to 1024, which is consistent with the original input pooling feature. Figure X Perform residual operation and output a target feature map Z that integrates the long-distance dependencies of the image.

[0062] Specifically, step S302 includes:

[0063] Step S3021: Perform compression channel processing on the pooled feature map through the position fusion attention mechanism module in the acupoint recognition model to obtain query features, key features, and value features corresponding to the pooled feature map.

[0064] In this embodiment, the acupoint positioning device uses the position fusion attention mechanism module in the acupoint recognition model to perform compression channel processing on the pooled feature map through 1×1×1 convolution to generate three features θ, φ and g, where θ is the query feature, φ is the key feature, and g is the value feature.

[0065] Step S3022 : Calculate a similarity matrix between the query feature and the key feature, and determine a transition feature based on the similarity matrix and the value feature, where the transition feature includes global context information and long-range dependency information.

[0066] In this embodiment, the acupoint positioning device performs matrix dot product operation on the query features and key features through the position fusion attention mechanism module in the acupoint recognition model to obtain a similarity matrix. The specific matrix dot product operation formula is:

[0067] f(x i ,x j )=θ(x i ) T φ(x j )

[0068] Among them, x i and y i Not pooled features Figure X In the i-th position and the j-th position, θ(x_i) and φ(x_j) are the representations of the query feature and the key feature after being mapped to the new feature space by 1×1 convolution, and T is the number of time steps.

[0069] After obtaining the similarity matrix, the position fusion attention mechanism module in the acupoint recognition model uses Softmax to normalize the similarity matrix and convert it into a weight matrix. Each element in the matrix represents the similarity weight of a position relative to other positions, and the value is between 0 and 1. The weight matrix is multiplied by the value feature to obtain the transition feature y of T×H×W×512. i , the transition features contain global context information and long-distance dependency information; global context information refers to the overall features and environmental information of the entire back image. It not only focuses on the local area in the image, but also grasps the content of the image as a whole, including the overall shape and posture of the back, the overall texture characteristics of the skin, and the relative position relationship between the acupoints and the surrounding tissues and organs; long-distance dependency information refers to the long-term dependency or long-distance correlation between different positions in the back image. In the back image, this means that the model can capture the association information between acupoints that are far apart. For example, there may be physiological connections between some acupoints, or although they are far apart in space in the image, they have certain correlations in function or structure. Among them, x i Transition characteristic y of position iDefined as:

[0070]

[0071] Among them, y i is the transition feature obtained by performing dot product operation on the weight matrix and the value feature, The weighted sum of all positions j that can be associated with the input feature map is calculated, C(x) is the normalization factor, g(x j ) is the result of linear transformation of the feature at position j.

[0072] Step S3023: Obtain a target feature map based on the transition feature and the pooled feature map.

[0073] In this embodiment, after obtaining the transition features, the acupoint locating device uses the position fusion attention mechanism module in the acupoint recognition model to first restore the number of channels of the transition features to 1024 using a 1×1 convolution, and then obtains the target feature map based on the transition features with restored channel numbers and the pooled feature map. The specific formula is:

[0074] Z=W z y i +X

[0075] Among them, Z is the feature map that integrates long-distance dependencies. z is a linear transformation used to project the transition features to the same dimension as the original input features, and X is the pooled feature map.

[0076] Step S303: Process the target feature map through the neck network and the head network in the acupoint recognition model to output a basic acupoint image on the back.

[0077] In this embodiment, after obtaining the target feature map, the acupoint localization device processes the target feature map through the neck network and head network in the acupoint recognition model, outputting a basic acupoint image on the back. Specifically, the acupoint recognition model inputs the output features of the second and third C2F modules in the backbone network, as well as the target feature map, into the neck network for processing. The head network then processes the output of the neck network to ultimately output a basic acupoint image on the back.

[0078] The acupoint positioning device of this embodiment processes the input back image based on the acupoint recognition model that has added a position fusion attention mechanism module. The position fusion attention mechanism module can calculate the similarity between any two positions in the input feature map to achieve global feature capture. This mechanism can dynamically adjust the feature representation of each position, so that the model can combine global contextual information while extracting local features, thereby improving the accuracy of back acupoint recognition.

[0079] Please refer to Figure 8 , Figure 8 This is a flow chart of the fourth embodiment of the present application. The difference between the fourth embodiment and the first to third embodiments is that, before the step of inputting the back image into the acupoint recognition model and outputting the back basic acupoint image, the process includes:

[0080] Step S401: training a target network based on a back acupoint sample image set to obtain a reference acupoint recognition model.

[0081] In this embodiment, the target network is iteratively trained a preset number of times based on a set of back acupoint sample images to obtain a reference acupoint recognition model; wherein, the real positions of 49 basic back acupoints are marked in the back acupoint sample images, and the target network is a YOLOv8-Pose network with a position fusion attention mechanism module added.

[0082] Optionally, in order to construct a high-quality set of back acupoint sample images, back areas of different individuals were selected for collection, covering a variety of body shapes, ages, genders and skin colors, and the back acupoints were labeled based on Labelme labeling software; Labelme is an open source image labeling tool that supports multiple labeling formats. The labeling process strictly follows the national standards for acupoint positioning, and the accuracy of each labeled back acupoint is ensured under the guidance of relevant traditional Chinese medicine practitioners; during the labeling process, all back acupoints are stored in the form of key points, and the two-dimensional pixel coordinates of each back acupoint in the image are recorded. These coordinate information can accurately reflect the position of each back acupoint in the image and is used for supervised learning in subsequent model training.

[0083] Optionally, in order to improve the efficiency of constructing the back acupoint sample image set, a pre-trained image processing model can be used to perform image analysis processing on the back images obtained by collecting the back areas of individuals of different body shapes, ages, genders and skin colors, and infer the spinal curve and acromion point of the back image, and then all the back acupoints can be inferred based on the spinal curve and acromion point to form a back acupoint sample image set.

[0084] Step S402: input the back acupoint verification image into the reference acupoint recognition model to obtain the back acupoint prediction image.

[0085] In this embodiment, after obtaining the reference acupoint recognition model, the back acupoint verification image is input into the reference acupoint recognition model to obtain a back acupoint prediction image; wherein, the back acupoint verification image is generated before the training model, and the real positions of all back acupoints are also marked in the back acupoint verification image; the back acupoint prediction image output by the reference acupoint recognition model is marked with the predicted positions of all back acupoints.

[0086] Step S403 : determining the Euclidean distance loss of the reference acupoint recognition model based on each annotated back acupoint in the back acupoint verification image and each predicted back acupoint in the back acupoint prediction image.

[0087] In this embodiment, after obtaining the back acupoint prediction image, the Euclidean distance loss of the reference acupoint recognition model is determined based on each labeled back acupoint in the back acupoint verification image and each predicted back acupoint in the back acupoint prediction image.

[0088] Specifically, the calculation formula of Euclidean distance loss is:

[0089] L OKS =σ1·d(p1,g1)+σ2·d(p2,g2)+σ3·d(p3,g3)+…+σ n ·d(p n ,g n )

[0090] Among them, p is the location information of the predicted back acupuncture points, g is the location information of the marked back acupuncture points, d(p n ,g n ) is the Euclidean distance between the nth group of predicted back acupoints and the annotated back acupoints. For example, p1 is the predicted location information of Dazhui acupoint, and g1 is the annotated location information of Dazhui acupoint. σ is the weight coefficient corresponding to the location information of each group of back acupoints, and the weight coefficient is determined by the number of back acupoint groups. L OKS The Euclidean distance loss of the reference acupoint recognition model is determined by accumulating the product of the Euclidean distance between the location information of each group of back acupoints and the weight coefficient.

[0091] Step S404 : determining a bounding box loss of the reference acupoint recognition model based on a real bounding box corresponding to each labeled back acupoint in the back acupoint verification image and a predicted bounding box corresponding to each predicted back acupoint in the back acupoint prediction image.

[0092] In this embodiment, to increase the focus on the relative spatial position of the annotated and predicted back acupoints, the bounding box loss of the reference acupoint recognition model is determined based on the ground-truth bounding box corresponding to each annotated back acupoint in the back acupoint verification image and the predicted bounding box corresponding to each predicted back acupoint in the back acupoint prediction image. The upper left corner of the back acupoint verification image is considered the common construction point for the annotated and predicted back acupoints, thereby establishing the ground-truth bounding box and the predicted bounding box, respectively.

[0093] Specifically, the bounding box loss is calculated as:

[0094]

[0095] Where IOU is the intersection-over-union ratio between the predicted bounding box and the true bounding box. 2 (b,b gt ) is the predicted bounding box center point b and the true bounding box center point b gt The square of the Euclidean distance between the predicted bounding box and the true bounding box. c is the diagonal length of the minimum enclosing area of the predicted bounding box and the true bounding box. w and h are the width and height of the predicted bounding box respectively. w gt and h gt are the width and height of the ground-truth bounding box, C w and C h is the maximum value of the width and height of the predicted bounding box and the true bounding box.

[0096] Step S405 : determining a target loss of the reference acupoint recognition model based on the Euclidean distance loss and the bounding box loss.

[0097] In this embodiment, after obtaining the Euclidean distance loss and bounding box loss of the reference acupoint recognition model, the target loss of the reference acupoint recognition model is determined based on the Euclidean distance loss and bounding box loss. Specifically, the target loss is the regression loss, and the specific calculation formula of the regression loss is:

[0098] L EOKS =α·L OKS +β·L EIOU

[0099] Among them, L EOKS is the regression loss, L EIOU is the bounding box loss, L OKS is the Euclidean distance loss, α and β are the weight coefficients of the loss, which are used to balance the contribution of the two losses to the total loss. For example, α and β are set to 0.5.

[0100] It should be noted that the loss function combining bounding box loss and Euclidean distance loss showed a lower loss value throughout the training process, indicating that the loss function has improved the accuracy of acupoint detection and better generalization performance after optimization.

[0101] Step S406: obtaining the acupoint recognition model based on the target loss and the preset expected loss.

[0102] In this embodiment, the target loss is compared with the preset expected loss. If the target loss is less than or equal to the preset expected loss, the reference acupoint recognition model obtained in this round of iteration is determined as the final acupoint recognition model; if the target loss is greater than the preset expected loss, the reference acupoint recognition model is iteratively trained again until the target loss is less than or equal to the preset expected loss, and the final acupoint recognition model is obtained.

[0103] In the process of training the acupoint recognition model, this embodiment optimizes the loss function. Combining the loss functions of bounding box loss and Euclidean distance loss, not only can the Euclidean distance loss of the reference acupoint recognition model be determined, but also the relative spatial relationship between the labeled back acupoints and the predicted back acupoints can be avoided. The acupoint recognition model obtained by training with the optimized loss function performs better in accurately predicting the position of the back acupoints and accurately predicting each back acupoint.

[0104] Please refer to Figure 9 , Figure 9 This is a flow chart of the fifth embodiment of the present application. The difference between the fifth embodiment and the first to fourth embodiments is that the target model is an image processing model, and the step of generating a back acupuncture point image based on the back image and the target model includes:

[0105] Step S501 : performing a first processing on the back image based on the image processing model to determine a spinal curve of the back image.

[0106] In this embodiment, after acquiring the back image, the acupoint locating device performs a first processing on the back image based on the image processing model to determine the spinal curve of the back image; wherein the image processing model is trained in advance; the first processing includes feature enhancement processing, feature point extraction processing, fitting processing, etc.

[0107] In one embodiment, the step of performing a first processing on the back image based on the image processing model to determine the spinal curve of the back image includes:

[0108] Step a: performing feature enhancement processing on the back image based on the image processing model to obtain a region of interest image corresponding to the back image.

[0109] In this embodiment, the acupoint locating device performs feature enhancement processing on the back image based on the image processing model to obtain a region of interest image corresponding to the back image. Specifically, the feature enhancement processing includes image channel conversion, adaptive histogram equalization, and region of interest extraction.

[0110] In back images, due to the low contrast of the skin surface and the unclear outline of the spinal curve, the features of acupoint positioning are difficult to extract accurately; to solve this problem, the image processing model first converts the back image into a blue channel back image.

[0111] The brightness changes of the back skin area in the blue channel image are relatively gentle, and the overall contrast is low, making it difficult to clearly identify important anatomical structures. Therefore, the image processing model performs adaptive histogram equalization on the blue channel back image. Adaptive histogram equalization can dynamically optimize the brightness distribution of different areas by adjusting the contrast of local areas of the image. It is particularly suitable for images with uneven contrast or blurred details. It can highlight details in local areas, so that the spinal curve can be better presented under different contrast conditions. Specifically, when performing adaptive histogram equalization, in order to avoid excessive amplification of noise, the contrast limit is adjusted to 2. During the local histogram equalization process, the amplitude of the grayscale distribution change within each pixel neighborhood is controlled within an appropriate range, thereby enhancing the image contrast while suppressing excessive amplification of noise.

[0112] To improve processing speed and reduce image size, we performed region of interest (ROI) extraction on the back image after adaptive histogram equalization. This extraction process involved using the cv2.selectROI function to select a rectangular box from the image and save the coordinates of the four vertices within the box to obtain the corresponding ROI image of the back image.

[0113] Step b: extracting spine feature points in the image of the region of interest, and performing curve fitting based on the spine feature points to determine the spine curve of the back image.

[0114] In this embodiment, after the acupoint locating device obtains the image of the region of interest corresponding to the back image through the image processing model, it extracts the spinal feature points in the image of the region of interest through the image processing model, and performs curve fitting based on the spinal feature points to determine the spinal curve of the back image.

[0115] Analysis of the region of interest image shows that the position of the pixel point of the human spine curve is the minimum brightness point of the row, so the minimum brightness point of each row in the region of interest image is extracted as the spine feature point.

[0116] Spline interpolation fitting, polynomial fitting and Bessel function fitting can be used to fit the spine feature points to determine the spine curve of the back image. Preferably, the curve fitted by the Bessel function can restore the human spine curve to the greatest extent and ensure smoothness and continuity at each node. Therefore, the image processing model preferably uses the Bessel function fitting to fit the spine feature points. In the fitting process, the starting point P0 and the ending point P n Determine the starting point of the spinal curve iControls the shape and curvature of the curve. The value of i ranges from 1 to n-1. By adjusting the number of characteristic control points, the Bezier function can restore the shape of the spine to the greatest extent. The definition of the Bezier curve is:

[0117]

[0118] Among them, B(t) is the parameterized form of the Bezier curve, t is a variable between 0 and 1, which defines the position of the point on the Bezier curve. As t changes from 0 to 1, the Bezier curve moves from the starting control point P0 to the ending control point P n .P i When fitting the spinal curve, the control points are selected from the characteristic points of the human spine. Each control point P i There is a corresponding Bessel basis function B i,n (t). The basis function is defined as:

[0119]

[0120] Where B i,n (t) is the Bezier curve basis function, which defines the influence of each control point on the curve. The number of combinations n, with i as the weight coefficient, reflects the combination method used when selecting i control points from n control points. Experimental results show that setting the number of control points to 100 during fitting yields the best results.

[0121] In one embodiment, the step of extracting spine feature points in the region of interest image includes:

[0122] Step b1: traverse the brightness values of each row of pixels in the image of the region of interest, and determine the pixel with the smallest brightness value in each row of pixels as a reference feature point.

[0123] In this embodiment, the image processing model pre-builds an all-zero image of the same size as the image of the region of interest, traverses the brightness values of each row of pixels in the image of the region of interest, and determines the pixel with the smallest brightness value in each row of pixels as the reference feature point, and sets the brightness of the pixel at the position corresponding to the reference feature point in the all-zero image to 1.

[0124] Step b2: verifying each of the reference feature points to determine the spine feature points.

[0125] In this embodiment, since there may be multiple pixels with the smallest brightness value in each row of pixels, there will be multiple pixels with brightness set to 1 in the same row of pixels in the all-zero image, which will interfere with the subsequent spinal curve fitting. Therefore, each reference feature point needs to be verified to determine the spinal feature point.

[0126] In one embodiment, the step of verifying each of the reference feature points to determine the spine feature points includes:

[0127] Step b21: for each reference feature point, counting the number of adjacent reference feature points of the reference feature point.

[0128] In this embodiment, the image processing model counts the number of adjacent reference feature points of each reference feature point, where adjacent refers to eight directions: upper left, upper, upper right, left, right, lower left, lower, and lower right, and counts the number of points with a brightness value of 1 for each pixel in these eight directions.

[0129] Step b22: If the number of adjacent reference feature points is less than a preset threshold, determine that the reference feature point is a noise point.

[0130] In this embodiment, if the number of adjacent reference feature points is determined to be less than a preset threshold, the reference feature point is determined to be a noise point, and the pixel brightness value of the reference feature point determined to be a noise point is set to 0. For example, the preset threshold is set to 3.

[0131] Step b23: If the number of adjacent reference feature points is not less than a preset threshold, the reference feature point is determined to be a spine feature point.

[0132] In this embodiment, if it is determined that the number of adjacent reference feature points is not less than a preset threshold, the reference feature point is determined to be a spine feature point. Exemplarily, the preset threshold is set to 3.

[0133] After setting the pixel brightness values of all reference feature points identified as noise points to 0 to avoid missing smaller noise points, a morphological opening operation is performed on the image. This operation consists of erosion followed by dilation. Erosion removes small noise points, while dilation restores the main shape lost during the erosion process. The opening operation uses a 2×2 kernel with element values of 1, eliminating any remaining interference points while preserving the curve's essential shape.

[0134] Step S502 : performing a second process on the back image based on the image processing model to determine the acromion position information of the back image.

[0135] In this embodiment, the acupoint locating device performs secondary processing on the back image based on the image processing model to determine the acromion position information of the back image. This secondary processing includes grayscale processing, back contour extraction, and calculation of tangent slopes at contour points. The acromion position information includes left and right acromion position information.

[0136] In one embodiment, the step of performing a second processing on the back image based on the image processing model to determine the acromion position information of the back image includes:

[0137] Step c: performing grayscale processing on the back image based on the image processing model to obtain a grayscale back image.

[0138] In this embodiment, the image processing model first grayscales the back image. This converts the color back image into a single intensity value image, thereby reducing the amount of data processing. In the specific implementation steps, the color space conversion function in OpenCV is called to convert the color back image into a grayscale back image by passing in parameters.

[0139] Step d: performing contour extraction processing on the grayscale back image based on the image processing model to obtain a back contour image.

[0140] In this embodiment, the image processing model performs contour extraction processing on the grayscale back image to obtain a back contour image. Among them, the image processing model can use three operators: Sobel, Prewitt, and Canny to perform contour extraction processing; because the Sobel operator is more sensitive to noise, some small noise points contained in the back image are mistakenly identified as edges. The Prewitt operator is a convolution-based edge detection algorithm. Its working principle is to calculate the gradient of the image in the horizontal and vertical directions. The detection result is relatively discrete, so the jagged edge of the human body contour drawn is more serious. The Canny operator can remove noise while ensuring that the contour is sufficiently smooth through non-maximum suppression and dual threshold detection; therefore, preferably, the image processing model uses the Canny operator for contour extraction processing.

[0141] Step e: Calculate the tangent slope of each contour point on the left and right sides of the back contour image respectively, determine the contour points with the smallest tangent slope on the left and right sides as acromion points and determine acromion position information.

[0142] In this embodiment, if Figure 10 As shown, Figure 10 Figure 1 is a schematic diagram of the acromion location information extraction process. After obtaining the back contour image, the image processing model fits a circumscribed rectangle to the entire back contour and further calculates the vertical midline of the circumscribed rectangle, which is located at the mid-axis of the back contour. Based on the vertical midline, the back contour is divided into left and right sides. The tangent slope of each contour point on the left and right sides of the back contour is calculated. The contour point with the smallest tangent slope on each side is identified as the acromion point, and the acromion location information is determined.

[0143] Specifically, for the contour point P i (X i ,Y i ), the slope of its tangent line passes through the adjacent point Pi-1 (X i-1 ,Y i-1 ) and P i+1 (X i+1 ,Y i+1 ) is calculated by the difference between them. The forward difference calculation formula is:

[0144]

[0145] The backward difference calculation formula is:

[0146]

[0147] Among them, slope i-1 and slope i+1 Represent the slope between the point and the previous point and the next point respectively. To get point P i The tangent slope of is calculated by taking the average of the forward difference and the backward difference, and the calculation formula is:

[0148]

[0149] Where tangent_slope i Point P i For the shoulder point, select the point with the smallest average tangent slope.

[0150] Step S503: generating a back acupuncture point image based on the back image, the spinal curve and the acromion position information.

[0151] In this embodiment, the acupoint locating device generates a back acupoint image based on an image processing model in combination with the back image, spinal curve and acromion position information.

[0152] In one embodiment, the step of generating a back acupuncture point image based on the back image, the spinal curve, and the acromion position information includes:

[0153] Step f: determining a target pixel distance based on the shoulder peak position information and a preset shoulder peak distance.

[0154] In this embodiment, the acupoint locating device determines the target pixel distance based on the acromion position information and the preset acromion distance through an image processing model. The specific target pixel distance d is calculated as d=L / 16, and the pixel distance between the two acromions is L.

[0155] Step g: determining the back acupoint position information based on the spinal curve, the target pixel distance and the preset acupoint distribution rule.

[0156] In this embodiment, the image processing model determines the location of acupoints on the back based on the spinal curve, target pixel distance, and preset acupoint distribution rules. Specifically, during the acupoint inference process, the 141 acupoints detected on the back are divided into four regions: the Foot-Taiyang Bladder Meridian acupoint group, the Hand-Taiyang Small Intestine Meridian acupoint group, the Extraordinary Meridian acupoint group, and the Hua Tuo Jiaji acupoint group. The location of each acupoint in these acupoint groups can be inferred from the position of Dazhui acupoint. Therefore, the location of Dazhui acupoint can be determined based on the spinal curve, target pixel distance, and preset acupoint distribution rules. The locations of other acupoints can then be inferred based on the location of Dazhui acupoint and the preset acupoint distribution rules.

[0157] For example, taking the inference of "Dingchuan Point" based on the basic acupuncture point "Dazhui Point" as an example, the pixel coordinates of "Dazhui Point" in the basic acupuncture point image of the back are (X, Y). The acupuncture point positioning device combines the pixel distance d corresponding to "one inch" in the basic acupuncture point image of the back, and combines the preset acupuncture point distribution rule that "Dazhui Point is located on the posterior midline of the human body, in the depression under the spinous process of the seventh vertebra; and Dingchuan Point is located under the spinous process of the seventh cervical vertebra, 0.5 inches to the left and right of the posterior midline", it can be inferred that the pixel coordinates of "Left Dingchuan Point" in the basic acupuncture point image of the back are (X-0.5d, Y), and the pixel coordinates of "Right Dingchuan Point" in the basic acupuncture point image of the back are (X+0.5d, Y). Taking the inference of "Jian Zhongshu" based on the basic acupuncture point "Dazhui" as an example, the preset acupuncture point distribution rule records that "the right Jian Zhongshu is 2 inches to the right of Dazhui, and the left Jian Zhongshu is 2 inches to the left of Dazhui". It can be inferred that the pixel coordinates of "Left Jian Zhongshu" in the basic acupuncture point image on the back are (X+2d, Y), and the pixel coordinates of "Right Jian Zhongshu" in the basic acupuncture point image on the back are (X-2d, Y).

[0158] Step h: generating a back acupoint image based on the back image and the back acupoint position information.

[0159] In this embodiment, after obtaining the position information of all back acupuncture points, the image processing model performs corresponding annotations on the back image based on the position information of each back acupuncture point to generate a back acupuncture point image.

[0160] This embodiment designs an automated extraction process for the spinal curve by analyzing the distribution characteristics of acupuncture points on the back. The process is based on a variety of image processing technologies such as channel conversion, histogram equalization, region of interest extraction, eight-neighborhood pixel culling, morphological processing, etc., and uses Bezier curve fitting to achieve accurate extraction of the spinal curve, thereby improving the accuracy of spinal curve fitting. Subsequently, by detecting the position information of the acromion point, the target pixel distance corresponding to one inch in the image is calculated, providing a reliable basis for the subsequent precise positioning of the acupuncture points. The position information of all back acupuncture points is then inferred by combining the spinal curve, acromion position information and target pixel distance, thereby improving the accuracy of back acupuncture point recognition.

[0161] Please refer to Figure 11 , Figure 11 This is a flow chart of the sixth embodiment of the present application. The difference between the sixth embodiment and the first to fifth embodiments is that the step of projecting the back acupuncture point image onto the back of the target object by a projection device to locate the position of each back acupuncture point of the target object includes:

[0162] Step S601 , determining the spatial size information of the back of the target object based on the spatial distance between the back image and the target object's back when the camera module in the projection device captures the back image, the pixel size information of the back image, and the focal length of the camera module.

[0163] In this embodiment, after obtaining the back acupoint image, the acupoint locating device determines the spatial size information of the back of the target object based on the spatial distance between the camera module and the back of the target object when shooting the back image, the pixel size information of the back image and the focal length of the camera module; wherein, the spatial distance between the camera module and the back of the target object when shooting the back image is determined during shooting, and the pixel size information of the back image can be determined after the camera module shoots the back image, and the focal length of the camera module is calibrated and corrected in advance.

[0164] Optionally, a specific calculation formula for calculating the spatial dimension information of the back of the target object is:

[0165]

[0166] Among them, X real and Y real The width and height of the object on the back of the target object in the real world, in millimeters, which is the spatial size information of the back of the target object. pixel and Y pixel is the pixel width and height of the back image. Z is the spatial distance between the camera module and the back of the target object when capturing the back image, in millimeters. x and f y is the focal length of the camera, which represents the field of view angle corresponding to each pixel in the image, in pixels.

[0167] Optionally, the Zhang checkerboard calibration method can be used to calibrate the camera module. The selected checkerboard grid size is 6×9-30mm, with 54 corner points and a grid size of 30mm. First, define the actual physical distance z between the camera module and the checkerboard as 50cm. After the checkerboard is aligned, capture the image from different angles and directions, ensuring that the checkerboard covers the entire image. The camera module resolution during capture is 640×480. After capturing the checkerboard image, use OpenCV to perform specific calibration of the camera module. First, define the number of intersections between the rows and columns of the checkerboard grid and the termination criteria for corner detection in the program. To improve calibration accuracy, set the number of iterations for corner detection to 30, and the accuracy threshold to 0.001. Specifically, the program stops detection when the number of detection iterations reaches the maximum value of 30 or when the movement of a corner point is less than the specified accuracy of 0.001. Then read the captured chessboard image, convert it into a grayscale image, and use the findChessboardCorners function in OpenCV to detect the corners of the chessboard. For the successfully detected corner points, call the cornerSubPix function to perform sub-pixel precision processing to improve the calibration accuracy. By traversing all images, enough corner points are collected, and the 3D points of the corner points and the 2D points in the image are saved. Finally, use the calibrateCamera calibration function in OpenCV to calculate the camera's intrinsic parameter matrix, distortion coefficient, rotation vector, and translation vector. The intrinsic parameter matrix K of the camera module contains the focal length and principal point position of the camera. The intrinsic parameter matrix obtained after solving is:

[0168]

[0169] In the matrix, fx and fy are the focal lengths of the camera in the x and y directions, with values of 548.7715962 and 536.3619627, respectively. cx and cy are the coordinates of the principal point in the camera image in the x and y directions, with values of 342.4261934 and 279.67487593, respectively.

[0170] Step S602 : determining a target scale factor between the back acupuncture point image and the back of the target object based on the pixel size information of the back acupuncture point image and the spatial size information of the back image.

[0171] In this embodiment, the acupoint locating device determines the target scale factor between the back acupoint image and the back of the target object based on the pixel size information of the back acupoint image and the spatial size information of the back image. Specifically, the projection angle, projection distance and focal length of the projection device are first adjusted to fixed values, L Px is the pixel distance of the back acupoint image in the x direction, L Py L is the pixel distance of the back acupoint image in the y direction, in pixels.x is the actual physical distance of the back image in the x direction, L y is the actual physical distance of the back image in the y direction, in mm. Then, the width scale factor in the target scale factor is the ratio between the pixel distance of the back acupoint image in the x direction and the actual physical distance of the back image in the x direction. The height scale factor in the target scale factor is the ratio between the pixel distance of the back acupoint image in the y direction and the actual physical distance of the back image in the y direction.

[0172] Exemplarily, when the projection distance of the projection device is 250 cm and the projection angle is 1.24°, the width scaling factor is 0.84 and the height scaling factor is 0.85.

[0173] Step S603 : Projecting the back acupoint image according to the target scale factor by the projection device to obtain a back acupoint projection image.

[0174] In this embodiment, after the target scale factor is determined and the projection distance and projection angle of the projection device are also set, the acupoint positioning device uses the projection device to restore the back acupoint image to its actual size in the real world according to the target scale factor to obtain a back acupoint projection image.

[0175] Furthermore, to ensure the quality and clarity of the projected image, other projector parameters such as contrast, resolution, and brightness must be further determined. These parameters not only determine the visual quality of the projected image but also directly impact the presentation of image detail and the accuracy of color reproduction. For example, the contrast is 50, the tracking parameter is 2172, the color mode is dynamic, the input signal is HDMI, the saturation is 64, the brightness is 52, the hue is 50, and the resolution is 1024×768.

[0176] Step S604: Adjust the back acupuncture point projection image to overlap with the back of the target object, so as to locate the position of each back acupuncture point of the target object.

[0177] In this embodiment, after the acupoint locating device projects the back acupoint projection image via the projection device, it adjusts the back acupoint projection image to overlap with the back of the target object, thereby locating the position of each back acupoint of the target object. It is understood that when the projection device projects the back acupoint projection image, the most important thing is to restore the back acupoint image to its actual size in the real world. However, the projection position of the back acupoint projection image may not be the same as the position on the back of the target object, or the size of the back acupoint projection image may differ slightly from the size of the back of the target object. In order to accurately locate the position of each back acupoint of the target object, the projection position and size of the back acupoint projection image need to be adjusted.

[0178] The step of adjusting the back acupuncture point projection image to overlap with the back of the target object to locate the position of each back acupuncture point of the target object includes:

[0179] Step S6041: Adjust the boundary of the back acupuncture point projection image to coincide with the boundary of the back of the target object, so as to locate the position of each back acupuncture point of the target object.

[0180] In this embodiment, after the projection device projects the back acupoint projection image, the camera module is used to identify the back acupoint projection image and the back of the target object, and the size of the back acupoint projection image is scaled or the projection position of the back acupoint projection image is translated and rotated according to the difference between the two. The overlap of the boundary of the back acupoint projection image and the boundary of the back of the target object is determined in real time until the boundary of the back acupoint projection image is adjusted to overlap with the boundary of the back of the target object, thereby locating the position of each back acupoint of the target object.

[0181] Step S6042: In response to the adjustment operation, the back acupuncture point projection image is adjusted in position until the target point of the back acupuncture point projection image coincides with the target point on the back of the target object, so as to locate the position of each back acupuncture point of the target object.

[0182] In this embodiment, a manual adjustment module is provided in the projection device, and the user can manually scale the size of the back acupoint projection image or translate and rotate the projection position of the back acupoint projection image according to the difference between the back acupoint projection image observed by the user and the back of the target object, until the boundary of the back acupoint projection image is adjusted to coincide with the boundary of the back of the target object, thereby locating the position of each back acupoint of the target object.

[0183] This embodiment projects a back acupuncture point projection image through a projection device, and scales the size of the back acupuncture point projection image or translates and rotates the projection position of the back acupuncture point projection image according to the difference between the back acupuncture point projection image and the back of the target object, so as to locate the position of each back acupuncture point of the target object, thereby avoiding manual positioning of the acupuncture points by medical personnel, avoiding the subjective influence of medical personnel, and thus improving the accuracy of back acupuncture point identification and positioning, and at the same time improving the efficiency of acupuncture point positioning.

[0184] It can be understood that the present application also provides a device for locating acupuncture points on the back of the human body. The device of this embodiment corresponds to the method for locating acupuncture points on the back of the human body in the above embodiment. The optional items in the above embodiment are also applicable to this embodiment, so they will not be repeated here.

[0185] The present application also provides a computer device. Exemplarily, the computer device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program to enable the computer device to execute the functions of each module in the above-mentioned human back acupuncture point locating method or the above-mentioned human back acupuncture point locating device.

[0186] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU) and a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or at least one of other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application.

[0187] The memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The memory is used to store a computer program, and the processor may execute the computer program accordingly after receiving an execution instruction.

[0188] The present application also provides a computer storage medium for storing the computer program used in the above-mentioned computer device. The computer storage medium may be a readable storage medium, a non-volatile storage medium, or a volatile storage medium. For example, the computer storage medium may include, but is not limited to, various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0189] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or flowchart, and the combination of boxes in the structure diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0190] In addition, the functional modules or units in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0191] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a smart phone, personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0192] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for locating acupuncture points on the back of a human body, characterized in that: The method comprises: Acquire a back image of the target object; generating a back acupoint image based on the back image and the target model; wherein each back acupoint of the target object is marked in the back acupoint image; The back acupuncture point image is projected onto the back of the target object by a projection device to locate the position of each back acupuncture point of the target object.

2. The method for locating acupuncture points on the back of a human body according to claim 1, wherein: The target model is an acupoint recognition model, and the step of generating a back acupoint image based on the back image and the target model includes: Inputting the back image into the acupoint recognition model and outputting a back basic acupoint image; A back acupoint image is generated based on the back basic acupoint image and preset acupoint distribution rules.

3. The method for locating acupuncture points on the back of a human body according to claim 2, wherein: The step of inputting the back image into the acupoint recognition model and outputting a back basic acupoint image comprises: Inputting the back image into the acupoint recognition model, processing the back image through the backbone network in the acupoint recognition model, and obtaining a pooled feature map corresponding to the back image; The pooled feature map is processed by the position fusion attention mechanism module in the acupoint recognition model to obtain a target feature map; The target feature map is processed by the neck network and the head network in the acupoint recognition model to output a basic acupoint image on the back.

4. The method for locating acupuncture points on the back of a human body according to claim 3, wherein: The step of processing the pooled feature map through the position fusion attention mechanism module in the acupoint recognition model to obtain the target feature map includes: Performing compression channel processing on the pooled feature map through the position fusion attention mechanism module in the acupoint recognition model to obtain query features, key features, and value features corresponding to the pooled feature map; Calculating a similarity matrix between the query feature and the key feature, and determining a transition feature based on the similarity matrix and the value feature, wherein the transition feature includes global context information and long-range dependency information; A target feature map is obtained based on the transition feature and the pooled feature map.

5. The method for locating acupuncture points on the back of a human body according to any one of claims 2 to 4, characterized in that: Before the step of inputting the back image into the acupoint recognition model and outputting the back basic acupoint image, the method includes: The target network is trained based on the back acupoint sample image set to obtain the reference acupoint recognition model; Inputting the back acupoint verification image into the reference acupoint recognition model to obtain the back acupoint prediction image; determining a Euclidean distance loss of the reference acupoint recognition model based on each annotated back acupoint in the back acupoint verification image and each predicted back acupoint in the back acupoint prediction image; determining a bounding box loss of the reference acupoint recognition model based on a true bounding box corresponding to each annotated back acupoint in the back acupoint verification image and a predicted bounding box corresponding to each predicted back acupoint in the back acupoint prediction image; determining a target loss of the reference acupoint recognition model based on the Euclidean distance loss and the bounding box loss; The acupoint recognition model is obtained based on the target loss and the preset expected loss.

6. The method for locating acupuncture points on the back of a human body according to claim 1, characterized in that: The target model is an image processing model, and the step of generating a back acupuncture point image based on the back image and the target model includes: performing a first processing on the back image based on the image processing model to determine a spinal curve of the back image; performing a second processing on the back image based on the image processing model to determine the acromion position information of the back image; A back acupuncture point image is generated based on the back image, the spinal curve and the acromion position information.

7. The method for locating acupuncture points on the back of a human body according to claim 1, characterized in that: The step of projecting the back acupuncture point image onto the back of the target object by a projection device to locate the position of each back acupuncture point of the target object includes: determining spatial size information of the back of the target object based on the spatial distance between the back of the target object and the camera module in the projection device when capturing the back image, pixel size information of the back image, and the focal length of the camera module; determining a target scale factor between the back acupuncture point image and the back of the target object based on pixel size information of the back acupuncture point image and spatial size information of the back image; Projecting the back acupuncture point image according to a target scale factor by the projection device to obtain a back acupuncture point projection image; The back acupuncture point projection image is adjusted to overlap with the back of the target object, so as to locate the position of each back acupuncture point of the target object.

8. The method for locating acupuncture points on the back of a human body according to claim 7, characterized in that: The step of adjusting the back acupuncture point projection image to overlap with the back of the target object to locate the position of each back acupuncture point of the target object includes: Adjusting the boundary of the back acupuncture point projection image to coincide with the boundary of the back of the target object to locate the position of each back acupuncture point of the target object; or In response to the adjustment operation, the back acupuncture point projection image is adjusted in position until the target point of the back acupuncture point projection image coincides with the target point on the back of the target object, so as to locate the position of each back acupuncture point of the target object.

9. A device for locating acupuncture points on the back of a human body, characterized in that: The human back acupuncture point locating device comprises: An acquisition module, used for acquiring a back image of a target object; A generating module, configured to generate a back acupoint image based on the back image and the target model; the back acupoint image is marked with each back acupoint of the target object; A positioning module is used to project the back acupuncture point image onto the back of the target object based on a projection device to locate the position of each back acupuncture point of the target object.

10. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the method for locating acupuncture points on the back of the human body according to any one of claims 1 to 7.

Images