Visual image spine posture analysis method and system based on machine learning

Abstract

The invention relates to a visual image spine posture analysis method based on machine learning, and relates to the field of general image data processing or generation, and the method comprises the steps: carrying out multiple learning actions on a BP neural network to obtain a spine posture analysis model; and intelligently analyzing scoliosis severity typing of the to-be-analyzed person according to the maximum numerical value of the body width of the to-be-analyzed person, various sensing parameters of the visual sensing mechanism and visual data of continuous multi-frame visual pictures by adopting a spine posture analysis model. The invention further relates to a visual image spine posture analysis system based on machine learning. According to the invention, aiming at the technical problem that the analysis result of the severity typing of scoliosis is not reliable and stable enough, the customized structure design of the spine posture analysis model is completed by adopting a targeted machine learning mode, and the severity typing of scoliosis is directly analyzed by using the basic data including the visual image; therefore, the technical problem is solved.

Description

Technical Field

[0001] The present invention relates to the general field of image data processing or generation, and in particular to a visual image spine posture analysis method and system based on machine learning. Background Art

[0002] The spine of a normal person should be a straight line when viewed from behind, and the two sides of the trunk should be symmetrical. If the shoulders are not at the same height when viewed from behind, or the back is uneven from side to side, scoliosis should be suspected. If scoliosis occurs, it is recommended to see a doctor as soon as possible to determine the degree of the disease, and then follow the doctor's advice and actively cooperate with the treatment, which will help to improve as soon as possible. Scoliosis, also known as scoliosis, is a three-dimensional deformity of the spine, including abnormal sequences in the coronal, sagittal and axial planes.

[0003] Scoliosis of degree one, degree two and degree three refers to the severity classification of scoliosis, with degree one being the mildest and degree three being the most severe. Seek medical attention as soon as possible if necessary. Degree one: also known as mild scoliosis, patients generally have no symptoms, with obvious scoliosis when standing, which disappears when lying flat, bending forward or hanging on a horizontal bar; it can generally be gradually relieved by adjusting daily sitting and standing postures. Degree two: also known as moderate scoliosis, with a relatively fixed scoliosis that does not disappear when lying flat, bending forward or hanging on a horizontal bar. At the same time, the shoulders may be unequal in height and the pelvis may be tilted, and the condition is more serious at this time. Degree three: also known as severe scoliosis, with the back of the convex side of the spine bulging and the front chest wall concave. At the same time, rotational deformity occurs, leading to torso twisting, which reduces the volume of the chest and abdominal cavities, causes a decrease in lung capacity, and heart, lung and digestive disorders.

[0004] For example, a Chinese invention patent with publication number CN117274151A proposes an artificial intelligence-based visual image spinal posture analysis method, which uses a distortion-free visible light camera to capture images of the front and back of a user, number and store them, and uses a model trained with artificial intelligence to identify bone key points in the visual image; uses X-rays to take pictures of the user's spine, calculates the spinal Cobb angle, and numbers and stores them; uses bone key point data and spinal information obtained by X-rays to train an artificial intelligence model to obtain a bone key point and Cobb angle correlation model; uses the bone key point and Cobb angle correlation model to perform visual image recognition on the user to directly obtain spinal posture information, thereby helping doctors to conduct rapid screening and further diagnosis and treatment.

[0005] For example, a Chinese invention patent with publication number CN118576186A proposes a scoliosis screening system based on digital images, the system comprising: an acquisition module for acquiring a first image of a natural standing posture of an object to be identified and a second image of a human body in a forward bending posture; an image processing module for preprocessing the first image and the second image; a first feature extraction module for extracting a first feature from the preprocessed first image; a second feature extraction module for extracting a second feature from the preprocessed second image; and a comprehensive evaluation module for comprehensively evaluating the scoliosis status of the object to be evaluated based on the first feature and the second feature. The digital image-based scoliosis screening system of the present invention analyzes the standing posture and forward bending posture images through digital image processing and computer vision technology, and performs a multi-dimensional comprehensive evaluation of the degree of scoliosis, thereby realizing a convenient, effective, and safe scoliosis screening evaluation without manual intervention, thereby greatly reducing costs.

[0006] However, although the above-mentioned existing technologies all analyze the spinal posture data of the tested person based on visual images, the analysis model or evaluation mode adopted is relatively rough, and the basic data used is not comprehensive and sufficient, and it is impossible to directly give the severity classification of the scoliosis of the tested person. As a result, it is necessary to continue to use electronic or manual modes to perform further data analysis on the obtained analysis results or evaluation results in order to obtain the severity classification of the scoliosis of the tested person as second-hand data. Obviously, since it is second-hand data itself, its reliability and stability cannot be effectively guaranteed. Summary of the invention

[0007] In order to solve the technical problems in the prior art, the present invention provides a visual image spinal posture analysis method and system based on machine learning. By adopting a targeted machine learning model, a customized design of the structure of the spinal posture analysis model is completed. On the basis of targeted screening of comprehensive and sufficient basic data including visual images, a spinal posture analysis model with a customized structural design is used to directly analyze the severity of scoliosis of the tested person. The characteristics of first-hand data also ensure the reliability and stability of the analysis results.

[0008] According to one aspect of the present invention, a visual image spine posture analysis method based on machine learning is provided, the method comprising: Obtaining the maximum value of body width, the maximum value of body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; Acquiring various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data, and the visual sensing mechanism faces the back of the person to be analyzed to perform the collection of multiple frames of continuous visual sensing images at the preset frame rate; Acquire multiple sets of visual data corresponding to multiple consecutive frames of visual images, wherein a single set of visual data corresponding to each frame of visual image is a vertical coordinate value, a horizontal coordinate value, and a depth of field value of each pixel point of an image block occupied by a person to be analyzed in the frame of visual image; Performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed the multiple learning actions and outputting it as a spinal posture analysis model; A spinal posture analysis model is used to intelligently analyze the severity classification of scoliosis of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images; Among them, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis.

[0009] According to another aspect of the present invention, a visual image spine posture analysis system based on machine learning is provided, the system comprising: The first capturing mechanism is used to obtain the maximum value of the body width, the maximum value of the body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; The second capture mechanism is used to obtain various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data. The visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at the preset frame rate; The third capturing mechanism is used to obtain multiple sets of visual data corresponding to consecutive frames of visual images, wherein the single set of visual data corresponding to each frame of visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of visual image; A target construction mechanism is used to perform multiple learning actions on the BP neural network to obtain the BP neural network that has performed the multiple learning actions and output it as a spinal posture analysis model; A grade analysis mechanism, connected to the first capture mechanism, the second capture mechanism, the third capture mechanism and the target construction mechanism respectively, for using a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism and multiple visual data corresponding to consecutive multi-frame visual images; Among them, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis.

[0010] It can be seen that the present invention has at least the following outstanding substantive features: Substantive feature A: Targeted selection of multiple basic data for intelligent analysis of scoliosis severity classification of the person to be analyzed, including the number of consecutive multi-frame visual sensing images obtained by shooting the back of the person to be analyzed, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images. The sufficiency and comprehensiveness of the above multiple basic data ensure the stability and effectiveness of the intelligent analysis results, and are used for efficient and high-precision on-site analysis of human spinal posture based on visual images; Substantive feature B: Specifically, the multiple body-related contents of the person to be analyzed are the maximum value of the body width, the maximum value of the body thickness, age information, gender information, height data and weight data of the person to be analyzed; the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data; the visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at a preset frame rate; and the single visual data corresponding to each frame of the visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of the visual image; Substantive feature C: A customized spinal posture analysis model is constructed for intelligent analysis of the severity classification of scoliosis of the person to be analyzed. The spinal posture analysis model is a BP neural network that has performed multiple learning actions, and the number of learning actions is positively correlated with the number of pixel circuits of the visual sensing mechanism, thereby customizing different structures of spinal posture analysis models based on machine learning for different visual sensing mechanisms, further ensuring the stability and effectiveness of the intelligent analysis results; Substantive feature D: In each learning action performed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of consecutive multi-frame visual sensing images corresponding to the analyzed person, the maximum value of the body width of the analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images corresponding to the analyzed person are used as the input content of the BP neural network to complete this learning action of the BP neural network, thereby ensuring the learning effect of each learning action performed on the BP neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The embodiments of the present invention will be described below with reference to the accompanying drawings, wherein: Figure 1 The present invention is a technical flow chart of a visual image spine posture analysis method and system based on machine learning.

[0012] Figure 2 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to a first embodiment of the present invention.

[0013] Figure 3 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to a second embodiment of the present invention.

[0014] Figure 4 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the third embodiment of the present invention.

[0015] Figure 5 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the fourth embodiment of the present invention.

[0016] Figure 6 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the fifth embodiment of the present invention.

[0017] Figure 7 1 is an internal structure diagram of a visual image spine posture analysis system based on machine learning according to the sixth embodiment of the present invention. DETAILED DESCRIPTION

[0018] like Figure 1 As shown, a technical flow chart of a visual image spine posture analysis method and system based on machine learning according to the present invention is given.

[0019] like Figure 1 As shown, the specific technical process of the present invention is as follows: Technical process 1: Targeted screening of multiple basic data for intelligent analysis of the severity of scoliosis of the person to be analyzed, wherein the multiple basic data involve dynamic multi-frame video images, various sensing parameters of the visual sensing mechanism, and multiple body-related contents of the person to be analyzed; Specifically, Figure 1 As shown, the multiple basic data include the number of consecutive multiple-frame visual sensing images obtained by shooting the back of the person to be analyzed, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple sets of visual data corresponding to the consecutive multiple-frame visual images; More specifically, the multiple body-related contents of the person to be analyzed are the maximum value of the body width, the maximum value of the body thickness, age information, gender information, height data and weight data of the person to be analyzed, the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data, the visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at a preset frame rate, and the single visual data corresponding to each frame of the visual image is the vertical coordinate value, the horizontal coordinate value and the depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of the visual image; For example, among the multiple body-related contents of the person to be analyzed, the maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed can be the contents input by the user interface, or can be the contents directly measured by multiple sensors on site; In this way, the sufficiency and comprehensiveness of the above-mentioned multiple basic data ensure the stability and effectiveness of the intelligent analysis results, which are used to complete the efficient and high-precision on-site analysis of human spine posture based on visual images; Technical process 2: Building a customized spinal posture analysis model for intelligent analysis of the scoliosis severity classification of the person to be analyzed; For example, the structural customization of the spinal posture analysis model mainly manifests in the following three aspects: In the first aspect: the spinal posture analysis model is a BP neural network that has performed multiple learning actions; Second aspect: the number of learning actions performed by the BP neural network is positively correlated with the number of pixel circuits in the visual sensing mechanism; Third aspect: in each learning action performed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of continuous multi-frame visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multi-frame visual images corresponding to the certain analyzed person are used as the input content of the BP neural network to complete this learning action of the BP neural network, thereby ensuring the learning effect of each learning action performed on the BP neural network; In this way, the machine learning-based spinal posture analysis models with different structures are customized for different visual sensing mechanisms, which further ensures the stability and effectiveness of the intelligent analysis results; Technical process three: The spinal posture analysis model customized by technical process two is used to intelligently analyze the severity of scoliosis of the person to be analyzed based on multiple basic data selected in a targeted manner by technical process one; For example, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis; Technical process 4: wirelessly sending the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis in technical process 3 to the nearest medical monitoring terminal via a wireless communication link; In this way, real-time analysis and data distribution of the scoliosis severity classification of each person to be analyzed who is located in front of the visual sensing mechanism and has his back to the visual sensing mechanism is completed. The characteristics of the intelligent analysis results as first-hand data also improve the stability and reliability of the analysis results of the spinal health status of each person to be analyzed.

[0020] The key points of the present invention are: using a targeted machine learning model to complete the customized design of the structure of the spinal posture analysis model, targeted screening of comprehensive and sufficient basic data including visual images, and using the spinal posture analysis model with a customized structural design to directly analyze the severity classification of scoliosis of the tested person as first-hand data.

[0021] Below, a visual image spine posture analysis method and system based on machine learning of the present invention will be specifically described in the form of an embodiment.

[0022] Example 1 Figure 2 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to a first embodiment of the present invention.

[0023] like Figure 2As shown, the visual image spine posture analysis method based on machine learning comprises the following steps: Step S201: obtaining the maximum value of body width, the maximum value of body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; For example, among the multiple body-related contents of the person to be analyzed, the maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed can be the contents input by the user interface, or can be the contents directly measured by multiple sensors on site; Step S202: Acquire various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, shooting focal length, preset frame rate, exposure time and aperture data, and the visual sensing mechanism faces the back of the person to be analyzed to perform the collection of multiple frames of continuous visual sensing images at a preset frame rate; For example, the number of pixel circuits, shooting focal length, preset frame rate, exposure time and aperture data of the visual sensing mechanism can be obtained by accessing the API interface mode of the visual sensing mechanism; Step S203: Acquire multiple sets of visual data corresponding to multiple consecutive frames of visual images, wherein a single set of visual data corresponding to each frame of visual image is a vertical coordinate value, a horizontal coordinate value, and a depth of field value of each pixel point of an image block occupied by the person to be analyzed in the frame of visual image; Specifically, multiple time stamps corresponding to the consecutive multiple frames of visual images are evenly spaced on the time axis; Step S204: performing multiple learning actions on the BP neural network to obtain a BP neural network that has performed multiple learning actions and outputting the BP neural network as a spinal posture analysis model; For example, performing multiple learning actions on the BP neural network to obtain the BP neural network after performing multiple learning actions and outputting the BP neural network as a spinal posture analysis model includes: optionally using a MATLAB toolbox to complete the testing and simulation of the data processing process of performing multiple learning actions on the BP neural network to obtain the BP neural network after performing multiple learning actions and outputting the spinal posture analysis model; Step S205: using a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple sets of visual data corresponding to the consecutive multi-frame visual images; The scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis; Specifically, first degree, second degree, and third degree scoliosis refer to the severity classification of scoliosis, with first degree being the mildest and third degree being the most severe. Seek medical treatment as soon as possible if necessary.

[0024] To be more specific: First degree: also known as mild scoliosis, patients generally do not have any symptoms, the scoliosis is obvious when standing, and disappears when lying flat, bending forward or hanging on a horizontal bar; it can usually be gradually relieved by adjusting daily sitting and standing postures. Second degree: also known as moderate scoliosis, the scoliosis is relatively fixed, and it does not disappear when lying flat, bending forward or hanging on a horizontal bar. At the same time, the shoulders may be unequal in height and the pelvis may be tilted. At this time, the condition is more serious. Third degree: also known as severe scoliosis, the back of the convex side of the spine bulges, the front chest wall is concave, and there is a rotational deformity, which leads to torsion of the trunk, a reduction in the volume of the chest and abdominal cavities, a decrease in lung capacity, and disorders of the heart, lungs and digestive functions; The maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed are obtained as the output of multiple body-related contents of the person to be analyzed, including: the maximum value of the body width of the person to be analyzed is the maximum value of each length corresponding to each horizontal section from top to bottom of the person to be analyzed, and the maximum value of the body thickness of the person to be analyzed is the maximum value of each width corresponding to each horizontal section from top to bottom of the person to be analyzed; The method of performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and outputting the BP neural network as a spinal posture analysis model includes: the number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism; For example, the number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism, including: the number of pixel circuits of the visual sensing mechanism corresponds to the picture resolution of standard definition, high definition, ultra-high definition, 2K, 4K, and 8K; Specifically, 720P: Its effective display format is 1280×720, and the corresponding number of pixel circuits is also 1280×720, using progressive scanning; 1080P: Its effective display format is 1920×1080, and it also uses progressive scanning.

[0025] 1080i: Compared with 1080p, 1080i uses interlaced scanning and has a resolution of 1920×1080. This format may provide a higher frame rate during transmission, but the image quality may be reduced compared to 1080p; 2K: The horizontal resolution reaches about 2000 pixels. The most standard 2K resolution is 2048×1024, and the common 2K resolution is 2560×1440 (16:9).

[0026] (UHDTV1) 4K: UHDTV1 is a VHD ultra-high definition system defined by the SMPTE ST 2036-1 standard with a resolution of 3840×2160 (16:9). (UHDTV2) 8K: UHDTV2 is a VHD ultra-high-definition system also defined by the SMPTE ST 2036-1 standard, with a resolution of 7680×4320; Among them, performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and output it as a spinal posture analysis model also includes: in each learning action performed on the BP neural network, using the known scoliosis severity classification of a certain analyzed person as the output content of the BP neural network, using the number of continuous multiple frames of visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images corresponding to the certain analyzed person as the input content of the BP neural network, to complete this learning action of the BP neural network; And wherein, in each learning action performed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of continuous multi-frame visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multi-frame visual images corresponding to the certain analyzed person are used as the input content of the BP neural network, and completing this learning action of the BP neural network includes: the continuous multi-frame visual sensing images corresponding to the certain analyzed person are the continuous multi-frame visual sensing images collected by the visual sensing mechanism facing the back of the certain analyzed person.

[0027] Example 2 Figure 3 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to a second embodiment of the present invention.

[0028] like Figure 3 As shown, compared with Figure 2, before obtaining various sensing parameters of the visual sensing mechanism, the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data, the visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at a preset frame rate, that is, before step S202, the visual image spinal posture analysis method based on machine learning also includes: Step S301: using the number of pixel circuits of the visual sensing mechanism, shooting focal length, preset frame rate, exposure time and aperture data to complete the on-site configuration of various sensing parameters of the visual sensing mechanism; Specifically, the IIC serial configuration interface can be used to send the number of pixel circuits, shooting focal length, preset frame rate, exposure time and aperture data of the visual sensing mechanism to the visual sensing mechanism to achieve on-site configuration of the visual sensing mechanism.

[0029] Example 3 Figure 4 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the third embodiment of the present invention.

[0030] like Figure 4 As shown, compared with Figure 2 After the spinal posture analysis model is used to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, that is, after step S205, the visual image spinal posture analysis method based on machine learning also includes: Step S401: wirelessly sending the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis to the nearest medical monitoring terminal via a wireless communication link; Specifically, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is wirelessly sent to the nearest medical monitoring terminal via a wireless communication link, including: the wireless communication link is a frequency division duplex communication link or a time division duplex communication link.

[0031] Example 4 Figure 5 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the fourth embodiment of the present invention.

[0032] like Figure 5 As shown, compared with Figure 2After the spinal posture analysis model is used to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, that is, after step S205, the visual image spinal posture analysis method based on machine learning also includes: Step S501: receiving the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis and displaying the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis in real time; For example, receiving the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis and displaying the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis in real time includes: using an LED display array or an LCD display array to receive the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis and displaying the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis in real time.

[0033] Example 5 Figure 6 The present invention is a flowchart showing a method for visual image spine posture analysis based on machine learning according to the fifth embodiment of the present invention.

[0034] like Figure 6 As shown, compared with Figure 5 After performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and outputting it as a spinal posture analysis model, that is, after step S204, the visual image spinal posture analysis method based on machine learning also includes: Step S601: receiving various model parameters of the spinal posture analysis model, and completing model storage of the spinal posture analysis model through various model parameters of the spinal posture analysis model; For example, receiving various model parameters of a spinal posture analysis model, and completing model storage of the spinal posture analysis model through the various model parameters of the spinal posture analysis model includes: using an MMC storage chip or a TF storage chip to receive various model parameters of the spinal posture analysis model, and completing model storage of the spinal posture analysis model through the various model parameters of the spinal posture analysis model.

[0035] Next, various embodiments of the present invention will be further described.

[0036] In each of the above embodiments, optionally, in the visual image spine posture analysis method based on machine learning: Acquire multiple sets of visual data corresponding to multiple consecutive frames of visual images, wherein the single set of visual data corresponding to each frame of visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of visual image, including: in each frame of visual image, take the pixel point in the lower left corner of the frame of visual image as the origin of the two-dimensional coordinate system, take the bottommost pixel row in the frame of visual image as the positive horizontal axis of the two-dimensional coordinate system, and take the leftmost pixel column in the frame of visual image as the positive vertical axis of the two-dimensional coordinate system, and establish the two-dimensional coordinate system of the frame of visual image; For example, in each frame of visual picture, taking the pixel point at the lower left corner of the frame of visual picture as the origin of the two-dimensional coordinate system, taking the bottommost pixel row in the frame of visual picture as the positive horizontal axis of the two-dimensional coordinate system, and taking the leftmost pixel column in the frame of visual picture as the positive vertical axis of the two-dimensional coordinate system, establishing the two-dimensional coordinate system of the frame of visual picture includes: the two-dimensional coordinate system of the frame of visual picture may also be executed by other establishment methods; What is important is that the two-dimensional coordinate system in each visual frame should be established in the same way; Among them, the spine posture analysis model is used to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, including: the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images are respectively converted into binary values ​​and then synchronously input into the spine posture analysis model; And wherein, using a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images also includes: running the spinal posture analysis model to obtain the scoliosis severity classification of the person to be analyzed represented by a binary value output by the spinal posture analysis model.

[0037] And in each of the above embodiments, optionally, in the visual image spine posture analysis method based on machine learning: The number of images of the continuous multiple frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images are converted into binary values ​​and then synchronously input into the spinal posture analysis model, including: using a numerical conversion component to perform binary value conversion processing on the number of images of the continuous multiple frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images; And wherein, the number of images of the continuous multiple frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multiple frames of visual images are converted into binary values ​​and then synchronously input into the spinal posture analysis model, which also includes: using a synchronous control interface connected to the numerical conversion component to control the synchronous input of the number of images of the continuous multiple frames of visual sensing images after binary value conversion, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multiple frames of visual images.

[0038] Example 6 Figure 7 1 is an internal structure diagram of a visual image spine posture analysis system based on machine learning according to the sixth embodiment of the present invention.

[0039] like Figure 7 As shown, the visual image spine posture analysis system based on machine learning includes the following components: The first capturing mechanism is used to obtain the maximum value of the body width, the maximum value of the body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; For example, among the multiple body-related contents of the person to be analyzed, the maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed can be the contents input by the user interface, or can be the contents directly measured by multiple sensors on site; The second capture mechanism is used to obtain various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data. The visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at the preset frame rate; For example, the number of pixel circuits, shooting focal length, preset frame rate, exposure time and aperture data of the visual sensing mechanism can be obtained by accessing the API interface mode of the visual sensing mechanism; The third capturing mechanism is used to obtain multiple sets of visual data corresponding to consecutive frames of visual images, wherein the single set of visual data corresponding to each frame of visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of visual image; Specifically, multiple time stamps corresponding to the consecutive multiple frames of visual images are evenly spaced on the time axis; A target construction mechanism is used to perform multiple learning actions on the BP neural network to obtain the BP neural network that has performed the multiple learning actions and output it as a spinal posture analysis model; For example, performing multiple learning actions on the BP neural network to obtain the BP neural network after performing multiple learning actions and outputting the BP neural network as a spinal posture analysis model includes: optionally using a MATLAB toolbox to complete the testing and simulation of the data processing process of performing multiple learning actions on the BP neural network to obtain the BP neural network after performing multiple learning actions and outputting the spinal posture analysis model; A grade analysis mechanism, connected to the first capture mechanism, the second capture mechanism, the third capture mechanism and the target construction mechanism respectively, for using a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism and multiple visual data corresponding to consecutive multi-frame visual images; The scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis; Specifically, first degree, second degree, and third degree scoliosis refer to the severity classification of scoliosis, with first degree being the mildest and third degree being the most severe. Seek medical treatment as soon as possible if necessary.

[0040] To be more specific: First degree: also known as mild scoliosis, patients generally do not have any symptoms, the scoliosis is obvious when standing, and disappears when lying flat, bending forward or hanging on a horizontal bar; it can usually be gradually relieved by adjusting daily sitting and standing postures. Second degree: also known as moderate scoliosis, the scoliosis is relatively fixed, and it does not disappear when lying flat, bending forward or hanging on a horizontal bar. At the same time, the shoulders may be unequal in height and the pelvis may be tilted. At this time, the condition is more serious. Third degree: also known as severe scoliosis, the back of the convex side of the spine bulges, the front chest wall is concave, and there is a rotational deformity, which leads to torsion of the trunk, a reduction in the volume of the chest and abdominal cavities, a decrease in lung capacity, and disorders of the heart, lungs and digestive functions; The maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed are obtained as the output of multiple body-related contents of the person to be analyzed, including: the maximum value of the body width of the person to be analyzed is the maximum value of each length corresponding to each horizontal section from top to bottom of the person to be analyzed, and the maximum value of the body thickness of the person to be analyzed is the maximum value of each width corresponding to each horizontal section from top to bottom of the person to be analyzed; The method of performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and outputting the BP neural network as a spinal posture analysis model includes: the number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism; For example, the number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism, including: the number of pixel circuits of the visual sensing mechanism corresponds to the picture resolution of standard definition, high definition, ultra-high definition, 2K, 4K, and 8K; Specifically, 720P: Its effective display format is 1280×720, and the corresponding number of pixel circuits is also 1280×720, using progressive scanning; 1080P: Its effective display format is 1920×1080, and it also uses progressive scanning.

[0041] 1080i: Compared with 1080p, 1080i uses interlaced scanning and has a resolution of 1920×1080. This format may provide a higher frame rate during transmission, but the image quality may be reduced compared to 1080p; 2K: The horizontal resolution reaches about 2000 pixels. The most standard 2K resolution is 2048×1024, and the common 2K resolution is 2560×1440 (16:9).

[0042] (UHDTV1) 4K: UHDTV1 is a VHD ultra-high definition system defined by the SMPTE ST 2036-1 standard with a resolution of 3840×2160 (16:9). (UHDTV2) 8K: UHDTV2 is a VHD ultra-high-definition system also defined by the SMPTE ST 2036-1 standard, with a resolution of 7680×4320; Among them, performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and output it as a spinal posture analysis model also includes: in each learning action performed on the BP neural network, using the known scoliosis severity classification of a certain analyzed person as the output content of the BP neural network, using the number of continuous multiple frames of visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images corresponding to the certain analyzed person as the input content of the BP neural network, to complete this learning action of the BP neural network; And wherein, in each learning action performed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of continuous multi-frame visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multi-frame visual images corresponding to the certain analyzed person are used as the input content of the BP neural network, and completing this learning action of the BP neural network includes: the continuous multi-frame visual sensing images corresponding to the certain analyzed person are the continuous multi-frame visual sensing images collected by the visual sensing mechanism facing the back of the certain analyzed person.

[0043] In addition, in a visual image spine posture analysis method and system based on machine learning according to the present invention: The number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism, including: using a numerical mapping function to represent a numerical mapping relationship that positively correlates the number of learning actions performed on the BP neural network with the number of pixel circuits of the visual sensing mechanism; For example, the numerical mapping relationship of the number of learning actions performed on the BP neural network and the number of pixel circuits of the visual sensing mechanism that is positively correlated with each other using a numerical mapping function includes: in the numerical mapping function, the number of pixel circuits of the visual sensing mechanism is an input value of the numerical mapping function, and the number of learning actions performed on the BP neural network corresponding to the number of pixel circuits of the visual sensing mechanism is an output value of the numerical mapping function; And wherein, in each learning action executed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of consecutive multi-frame visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images corresponding to the certain analyzed person are used as the input content of the BP neural network. Completing this learning action of the BP neural network also includes: multiple learning actions executed on the BP neural network correspond to multiple different analyzed persons respectively.

[0044] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A visual image spine posture analysis method based on machine learning, characterized in that: The method comprises: Obtaining the maximum value of body width, the maximum value of body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; Acquiring various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data, and the visual sensing mechanism faces the back of the person to be analyzed to perform the collection of multiple frames of continuous visual sensing images at the preset frame rate; Acquire multiple sets of visual data corresponding to multiple consecutive frames of visual images, wherein a single set of visual data corresponding to each frame of visual image is a vertical coordinate value, a horizontal coordinate value, and a depth of field value of each pixel point of an image block occupied by a person to be analyzed in the frame of visual image; Performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed the multiple learning actions and outputting it as a spinal posture analysis model; A spinal posture analysis model is used to intelligently analyze the severity classification of scoliosis of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images; Among them, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis.

2. A visual image spine posture analysis method based on machine learning as claimed in claim 1, characterized in that: The maximum value of the body width, the maximum value of the body thickness, the age information, the gender information, the height data and the weight data of the person to be analyzed are obtained, so as to output multiple body-related contents of the person to be analyzed, including: the maximum value of the body width of the person to be analyzed is the maximum value of each length corresponding to each horizontal section from top to bottom of the person to be analyzed, and the maximum value of the body thickness of the person to be analyzed is the maximum value of each width corresponding to each horizontal section from top to bottom of the person to be analyzed; Among them, performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and outputting it as a spinal posture analysis model includes: the number of learning actions performed on the BP neural network is positively correlated with the number of pixel circuits of the visual sensing mechanism.

3. A visual image spine posture analysis method based on machine learning as claimed in claim 2, characterized in that: Executing multiple learning actions on the BP neural network to obtain the BP neural network that has completed the multiple learning actions and outputting it as a spinal posture analysis model also includes: in each learning action executed on the BP neural network, using the known scoliosis severity classification of a certain analyzed person as the output content of the BP neural network, using the number of consecutive multiple frames of visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the consecutive multiple frames of visual images corresponding to the certain analyzed person as the input content of the BP neural network, to complete this learning action of the BP neural network; Among them, in each learning action performed on the BP neural network, the known scoliosis severity classification of a certain analyzed person is used as the output content of the BP neural network, and the number of continuous multi-frame visual sensing images corresponding to the certain analyzed person, the maximum value of the body width of the certain analyzed person, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the continuous multi-frame visual images corresponding to the certain analyzed person are used as the input content of the BP neural network. Completing this learning action of the BP neural network includes: the continuous multi-frame visual sensing images corresponding to the certain analyzed person are the continuous multi-frame visual sensing images collected by the visual sensing mechanism facing the back of the certain analyzed person.

4. A visual image spine posture analysis method based on machine learning as claimed in claim 3, characterized in that: Before obtaining various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data, and the visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at the preset frame rate, the method further includes: The number of pixel circuits of the visual sensing mechanism, shooting focal length, preset frame rate, exposure time and aperture data are used to complete the on-site configuration of various sensing parameters of the visual sensing mechanism.

5. A visual image spine posture analysis method based on machine learning as claimed in claim 3, characterized in that: After using the spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, the method further includes: The scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is wirelessly sent to the nearest medical monitoring terminal via a wireless communication link.

6. A visual image spine posture analysis method based on machine learning as claimed in claim 3, characterized in that: After using the spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, the method further includes: Receive the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis and display the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis in real time.

7. The method for analyzing spinal posture using visual images based on machine learning as claimed in claim 3, characterized in that: After performing multiple learning actions on the BP neural network to obtain the BP neural network that has performed multiple learning actions and outputting it as a spinal posture analysis model, the method further includes: Receive various model parameters of the spinal posture analysis model, and complete model storage of the spinal posture analysis model through the various model parameters of the spinal posture analysis model.

8. The method for analyzing spinal posture using visual images based on machine learning as claimed in claim 7, characterized in that: Acquire multiple sets of visual data corresponding to multiple consecutive frames of visual images, wherein the single set of visual data corresponding to each frame of visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of visual image, including: in each frame of visual image, take the pixel point in the lower left corner of the frame of visual image as the origin of the two-dimensional coordinate system, take the bottommost pixel row in the frame of visual image as the positive horizontal axis of the two-dimensional coordinate system, and take the leftmost pixel column in the frame of visual image as the positive vertical axis of the two-dimensional coordinate system, and establish the two-dimensional coordinate system of the frame of visual image; Among them, the spine posture analysis model is used to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images, including: the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images are respectively converted into binary values ​​and then synchronously input into the spine posture analysis model; Among them, the use of a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed based on the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive multi-frame visual images also includes: running the spinal posture analysis model to obtain the scoliosis severity classification of the person to be analyzed represented by a binary value output by the spinal posture analysis model.

9. The method for analyzing spinal posture using visual images based on machine learning as claimed in claim 7, characterized in that: The number of images of the continuous multiple frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images are converted into binary values ​​and then synchronously input into the spinal posture analysis model, including: using a numerical conversion component to perform binary value conversion processing on the number of images of the continuous multiple frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple copies of visual data corresponding to the continuous multiple frames of visual images; Among them, the number of consecutive frames of visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive frames of visual images are converted into binary values ​​and then synchronously input into the spinal posture analysis model. It also includes: using a synchronous control interface connected to the numerical conversion component to control the synchronous input of the number of consecutive frames of visual sensing images that have been converted into binary values, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism, and multiple visual data corresponding to the consecutive frames of visual images.

10. A visual image spine posture analysis system based on machine learning, characterized in that: The system comprises: The first capturing mechanism is used to obtain the maximum value of the body width, the maximum value of the body thickness, age information, gender information, height data and weight data of the person to be analyzed, so as to output multiple body-related contents of the person to be analyzed; The second capture mechanism is used to obtain various sensing parameters of the visual sensing mechanism, wherein the various sensing parameters of the visual sensing mechanism are the number of pixel circuits of the visual sensing mechanism, the shooting focal length, the preset frame rate, the exposure time and the aperture data. The visual sensing mechanism faces the back of the person to be analyzed to perform the acquisition of multiple frames of continuous visual sensing images at the preset frame rate; The third capturing mechanism is used to obtain multiple sets of visual data corresponding to consecutive frames of visual images, wherein the single set of visual data corresponding to each frame of visual image is the vertical coordinate value, horizontal coordinate value and depth of field value of each pixel point of the image block occupied by the person to be analyzed in the frame of visual image; A target construction mechanism is used to perform multiple learning actions on the BP neural network to obtain the BP neural network that has performed the multiple learning actions and output it as a spinal posture analysis model; A grade analysis mechanism, connected to the first capture mechanism, the second capture mechanism, the third capture mechanism and the target construction mechanism respectively, for using a spinal posture analysis model to intelligently analyze the scoliosis severity classification of the person to be analyzed according to the number of consecutive multi-frame visual sensing images, multiple body-related contents of the person to be analyzed, various sensing parameters of the visual sensing mechanism and multiple visual data corresponding to consecutive multi-frame visual images; Among them, the scoliosis severity classification of the person to be analyzed obtained by intelligent analysis is one of first degree scoliosis, second degree scoliosis and third degree scoliosis.

Images