An acupuncture teaching assistance platform and method based on intelligent analysis

Abstract

The present invention discloses an acupuncture teaching assistance platform and method based on intelligent analysis. S1. Output a preprocessing action signal data set; S2. Obtain a delay parameter set and an embedding dimension value; S3. Form a phase space reconstruction trajectory set; S4. Generate an action trajectory segment object set; S5. Construct a bilateral preference sequence according to the dynamic similarity index and the stability difference index; S6. Execute an improved stable marriage algorithm based on the bilateral preference sequence to obtain a comprehensive stability metric value; S7. Map a preset scoring interval according to the comprehensive stability metric value to generate a stability scoring result, and push the stability scoring result and the corresponding action gap information to the teaching platform to complete the feedback. The present invention can accurately identify unstable paragraphs and control weak points existing in the operations of students, and the teaching intervention is more targeted.

Description

Technical Field

[0001] The present invention relates to the technical field of acupuncture teaching, and particularly to an acupuncture teaching assistance platform and method based on intelligent analysis. Background Art

[0002] With the advancement of medical education informatization and visualization technology, acupuncture, as a highly operational diagnosis and treatment technique in traditional Chinese medicine, the teaching quality directly affects the clinical efficacy and patient safety. Traditional acupuncture teaching usually relies on teachers' oral explanations combined with practical demonstrations for guidance. Students master the operation steps of needle insertion, retention, and needle withdrawal through imitation learning. However, this "mentor-apprentice" teaching method has significant limitations: on the one hand, the teaching process lacks an accurate and quantitative feedback mechanism, and it is difficult for students to understand the specific differences between their own actions and the standard actions in a timely manner during the operation process; on the other hand, the subjective evaluation of teachers accounts for a high proportion in teaching, and the evaluation criteria are not unified, which easily leads to large differences in teaching results.

[0003] In recent years, some acupuncture teaching institutions have tried to introduce sensor devices and computer vision technology to collect and analyze the actions during acupuncture, and assist teaching evaluation by recording hand trajectories and force conditions. However, existing methods mostly focus only on the data analysis of single modalities such as displacement or posture, ignoring the coupling relationship between the three-dimensional spatial posture changes and axial force application during acupuncture operation, resulting in a single evaluation dimension and poor result interpretability. In addition, existing technologies generally fail to achieve dynamic segmentation of the entire action process and accurate extraction of stage characteristics, making it difficult to perform fine-grained analysis on each link of "needle insertion - retention - needle withdrawal" during the acupuncture process, and lacking quantitative support for core indicators such as dynamic stability and action trajectory consistency.

[0004] More critically, most existing teaching evaluation tools compare the actions of students and experts in a way based on template matching or similarity scoring. They fail to conduct comprehensive modeling by combining the complexity of the dynamic evolution process, and also fail to integrate a bilateral preference matching mechanism to globally analyze the matching stability between "student - standard" action pairs. This leads to the feedback results being limited to static difference judgments, making it difficult to reflect the process stability and overall consistency of action execution, and restricting the teaching system's ability to accurately judge the skill level of students and generate personalized improvement suggestions.

[0005] In summary, there is an urgent need for a teaching assistance method that can simultaneously integrate multi-modal data, multi-stage process division, and dynamic stability evaluation mechanisms to improve teaching efficiency and evaluation scientificity. Summary of the Invention

[0006] An object of the present invention is to propose an acupuncture teaching assistance platform and method based on intelligent analysis. The present invention can accurately identify unstable paragraphs and weak control points in students' operations, and the teaching intervention is more targeted.

[0007] An acupuncture teaching auxiliary method based on intelligent analysis according to an embodiment of the present invention comprises the following steps:

[0008] S1. Collect multimodal action data sets for the whole process of acupuncture teaching, perform preprocessing operations, and output preprocessed action signal data sets;

[0009] S2. Calculate the delay time parameter and the embedding dimension parameter for the preprocessed action signal data set to obtain a delay parameter set and an embedding dimension value;

[0010] S3. Perform phase space reconstruction on the preprocessed motion signal data set according to the delay parameter set and the embedding dimension value to form a phase space reconstruction trajectory set;

[0011] S4. Based on the phase space reconstruction trajectory set, the action segments are divided according to the acupuncture insertion stage, the stay stage and the needle removal stage, and the local dynamic features are extracted to generate the action trajectory segment object set;

[0012] S5. construct a standard motion trajectory segment object set according to the teaching reference library, use the motion trajectory segment object set and the standard motion trajectory segment object set as the matching parties, and construct a bilateral preference sequence according to the dynamic similarity index and the stability difference index;

[0013] S6. Execute the improved stable marriage algorithm based on the bilateral preference sequence to obtain a comprehensive stability measurement index value;

[0014] S7. Generate a stability score result based on the preset score interval mapped by the comprehensive stability measurement index value, and push the stability score result and the corresponding action gap information to the teaching platform to complete the feedback.

[0015] Optionally, the S1 includes the following steps:

[0016] S11. Use a multi-modal acquisition device to synchronously collect signals during the entire acupuncture teaching process, obtain three-axis displacement information, three-axis posture information, and axial force information, and form a multi-modal action original data set ;

[0017] ;in, Indicates Single-frame multimodal action data at sampling time, Indicates the sampling timestamp, Represents the three-axis displacement vector, reflecting the position of the hand or needle in space. Indicates the three-axis attitude angle, corresponding to the pitch angle, yaw angle and roll angle, representing the attitude state of the needle body. Represents the instantaneous force in the needle axis direction, reflecting the mechanical response during the penetration process. Represents the total number of sampling frames;

[0018] S12. Perform wavelet threshold denoising on the original multi-modal action dataset, retain the main frequency characteristics of the signal, and perform linear interpolation resampling to align all modal data to a unified time resolution. , and through timestamp consistency alignment, form an action signal dataset after time alignment. The action signal dataset after time alignment is used as the preprocessed action signal dataset. .

[0019] Optionally, the S2 includes the following steps:

[0020] S21. For the three-axis displacement information, three-axis attitude information, and axial force information in the preprocessed action signal dataset, respectively use the average mutual information function to estimate the optimal delay time parameter, and find the time point where the signal first shows information independence during the time evolution by calculating the statistical correlation of the signal at different time intervals . Based on the time point where the signal first shows information independence, determine the delay time parameter of the three-axis displacement information , the delay time parameter of the three-axis attitude information , and the delay time parameter of the axial force information respectively. The three together form a set of delay time parameters ;

[0021] ;

[0022] Among them, represents the joint probability distribution of different time values of the signal at time intervals , , , are their respective marginal distributions, is the candidate delay time parameter; so that the average mutual information function reaches the local minimum for the first time corresponding to as the optimal delay time parameter of the current signal;

[0023] S22. After obtaining the set of delay time parameters , for each type of signal component, use the pseudo-nearest neighbor ratio function Determine the optimal embedding dimension. By comparing the changes in the relative distances between adjacent trajectory points in the reconstructed phase space under different embedding dimensions, identify the proportion of false neighbors. When the false neighbor ratio at a certain embedding dimension drops below the threshold, it is considered that the original nonlinear structure has been sufficiently unfolded at this embedding dimension, and the embedding dimensions of the three-axis displacement information are obtained respectively. The embedding dimension of the three-axis attitude information And the embedding dimension of the axial force information Together, they form an embedding dimension set : ;

[0024] Among them, Is the reconstructed vector with dimension , Is its nearest neighbor, Is the delay time parameter, Indicates in the -dimensional case, whether the -th vector has false neighbors; when the false neighbor ratio function value is less than the threshold , the current is selected as the embedding dimension.

[0025] Optionally, the S3 includes the following steps:

[0026] S31. Based on the delay time parameter set and the embedding dimension set, for the preprocessed action signal data set, construct a multi-modal collaborative reconstructed phase space that incorporates the dynamic stability constraints of acupuncture actions. On the basis of single-signal independent reconstruction, introduce the dynamic interaction between the three-axis displacement information, the three-axis attitude information, and the axial force information as the collaborative reconstruction constraint conditions;

[0027] S32. For the data at the -th moment in the preprocessed action signal data set, construct a collaborative phase space state vector that incorporates the multi-dimensional information coupling of acupuncture actions :

[0028] ;

[0029] Among them, Represents the value of the three-axis displacement information at the -th moment, where Is the delay time parameter of the three-axis displacement information, , used to construct the time-delay embedding sequence of the three-axis displacement information, Is the embedding dimension of the three-axis displacement information, indicating how many delay embedding components are used to construct the state space trajectory of this signal, Represents the three-axis attitude information at the The value at a certain moment is the delay time parameter of the three-axis attitude information and is used to construct the time-delay embedding sequence of the attitude signal is the embedding dimension of the three-axis attitude information, indicating the dimension used when constructing the phase space of this signal represents the value of the axial force information at the th moment is the delay time parameter of the axial force and is used to construct the time-delay embedding sequence of the force signal is the embedding dimension of the axial force information

[0030] The collaborative phase space state vector is formed by splicing the sub-vectors obtained by embedding the three modal signals under their respective optimal delay times and embedding dimensions, and is used to capture the collaborative dynamic characteristics of the acupuncture action in terms of spatial movement, attitude change, and mechanical feedback

[0031] S33. Calculate the trajectory neighborhood density distribution of the multi-dimensional information of the acupuncture action in the collaborative phase space based on the collaborative phase space state vector , and the trajectory neighborhood density distribution reflects the stability of the trainee's acupuncture action in the dynamic state space ; ;

[0032] Among them, is the volume of the neighborhood sphere centered on the current collaborative phase space state vector , is the Heaviside step function , is the scale of the set of collaborative phase space state vectors; in the topic of analyzing the stability of acupuncture actions in acupuncture teaching

[0033] S34. Calculate the local Lyapunov exponent of the collaborative phase space trajectory according to the trajectory neighborhood density distribution and analyze the local stability of the dynamic evolution process of the acupuncture action ; ;

[0034] Among them, is the sampling interval is the frame number increment of the trajectory advancing along the time axis represents the average of all point pairs in the neighborhood of the current point

[0035] In the topic of analyzing the stability of acupuncture actions in acupuncture teaching, the local Lyapunov exponent is used to measure the local dynamic stability of the trainee's acupuncture action in real time

[0036] S35. Combine the collaborative phase space state vector , the trajectory neighborhood density distribution and the local Lyapunov exponent to jointly form a set of phase space reconstruction trajectories with the constraint of the dynamic stability of acupuncture actions .

[0037] Optionally, the S4 includes the following steps:

[0038] S41. According to the set of phase space reconstruction trajectories and the first derivative of the three-axis displacement information in the axial direction, identify three key stages in the whole process of acupuncture, including the insertion stage, the retention stage, and the withdrawal stage. By analyzing the change trend of the displacement in the time series, judge the interval boundaries where the speed continuously rises, stabilizes at zero, or continuously decreases, dynamically divide the time period indexes corresponding to the stages, and classify the time period indexes into the insertion stage index set, the retention stage index set, and the withdrawal stage index set respectively. Extract the corresponding collaborative phase space state vector sequences from the set of phase space reconstruction trajectories according to the three index sets, and form the insertion stage trajectory sequence, the retention stage trajectory sequence, and the withdrawal stage trajectory sequence respectively, constituting a set of stage trajectory sequences;

[0039] S42. For each stage trajectory sequence in the set of stage trajectory sequences, subtract the adjacent two collaborative phase space state vectors in chronological order to obtain the state transition direction vector sequence within the current stage. The state transition direction vector sequence reflects the evolution direction trend of the acupuncture action in the collaborative phase space within this stage;

[0040] S43. Within each stage, use the included angle information between adjacent state transition direction vectors to calculate the trajectory curvature sequence of the trajectory in this stage. The larger the trajectory curvature value, the more obvious the bending change of the trajectory in this stage, reflecting the instability degree of the acupuncture action trajectory, and used to measure the local nonlinear degree of the evolution trajectory of the action in the collaborative phase space;

[0041] S44. Combine the collaborative phase space state vector sequence, the state transition direction vector sequence, and the trajectory curvature sequence corresponding to each stage to form the insertion stage action trajectory segment object. The insertion stage action trajectory segment object includes the insertion stage trajectory sequence, the state transition direction vector sequence, and the insertion stage curvature sequence. The retention stage action trajectory segment object includes the retention stage trajectory, the retention stage direction vector, and the retention stage curvature sequence. The withdrawal stage action trajectory segment object includes the withdrawal stage trajectory, the withdrawal stage direction vector, and the withdrawal stage curvature sequence;

[0042] S45. Combine the action trajectory segment objects corresponding to the insertion stage, the retention stage, and the withdrawal stage to form a set of action trajectory segment objects.

[0043] Optionally, S5 includes the following steps:

[0044] S51. Call the standard acupuncture action data from the teaching reference library. The standard acupuncture action data consists of multimodal information during the operation of acupuncture experts, including three-axis displacement information, three-axis posture information and axial force information. It is consistent with the format of the student's pre-processed action signal data set and is recorded as the standard action original data set. ;

[0045] S52. Standard action original data set Execute the processing steps consistent with the student data processing flow, and finally generate a set of standard action trajectory fragment objects , including standard motion trajectory fragment objects of the needle insertion stage, the stay stage and the needle removal stage, which are respectively composed of a standard collaborative state vector sequence, a standard state transfer direction vector sequence and a standard trajectory curvature sequence;

[0046] S53. Collect the student's motion trajectory segment objects The student-side preference sequence is constructed by taking the set of standard action trajectory fragments as the matching parties. The preference degree of the student's action trajectory fragments to the standard action trajectory fragments is jointly determined by two dimensions: the first is the structural similarity between the collaborative state vector sequences, which is used to measure the consistency of the action trajectory evolution path; the second is the difference between the average trajectory curvatures, which is used to reflect the deviation of the action trajectory in stability; the two are weightedly combined to form the student-side preference score result, and the student-side preference sequence is generated by arranging the preference scores from high to low;

[0047] S54. The standard side preference sequence is constructed based on the same method. When generating preferences, the standard action trajectory fragments also form a preference score based on the structural similarity of the collaborative state vector sequence and the difference in the average trajectory curvature, and the standard side preference sequence is generated by sorting according to the preference score. The student side preference sequence and the standard side preference sequence together constitute a bilateral preference sequence.

[0048] Optionally, the S6 includes the following steps:

[0049] S61. Based on the bilateral preference sequence, aiming at the specific needs of acupuncture action stability analysis in acupuncture teaching, an improved stable marriage algorithm is constructed that incorporates the dynamic feature constraints of acupuncture actions and the adaptive adjustment of the weight of staged action stability. The improved stable marriage algorithm introduces a dynamic update mechanism of dynamic preferences between acupuncture action segments, and updates the preference list in real time during each round of matching, so as to meet the requirements for action stability evaluation in the process of acupuncture teaching;

[0050] S62. During the round matching process, for each student's action trajectory segment object , dynamically update the preference value according to its historical matching status with each standard action trajectory segment object ; :

[0051] ;

[0052] Among them, , are respectively the collaborative state vector sequences of the th trainee action trajectory segment object and the th standard action trajectory segment object in the matching edge, reflecting the structural similarity degree between acupuncture action trajectories. represents the cosine similarity of the trajectory sequence, , respectively represent the average trajectory curvature of the corresponding trajectory segments, is the fixed weight of the trajectory structure similarity, is the dynamic weight, which is adaptively adjusted with the increase of the matching round number. represents the cumulative historical times of the trainee action trajectory segment object rejecting or accepting the standard action trajectory segment object in the previous round of matching, is the adjustment factor for the influence of the historical rejection or acceptance times on the preference value update;

[0053] S63. During each round of stable matching process in the th round of matching, calculate and record the local stability values of each matching edge in real time: ;

[0054] Among them, is the local Lyapunov exponent, representing the mean value of the local Lyapunov exponents of the trainee action trajectory segment object with respect to the standard action trajectory segment object , which is used to characterize the local dynamic stability difference between the two trajectory segments in the collaborative phase space. The introduction of the exponential function is used to emphasize the decisive role of the local stability characteristics of acupuncture action dynamics in the stable marriage matching process;

[0055] S64. When the absolute value of the difference between the local stability values corresponding to the matching edge between adjacent rounds of matching is lower than the preset convergence threshold , the improved stable marriage algorithm stops iterating, determines the final stable matching relationship set , and outputs the corresponding global one-to-many stable matching relationship;

[0056] S65. For the final stable matching relationship set The local stability values of all matching edges are weighted and aggregated to obtain a comprehensive stability metric value for the stability analysis of acupuncture teaching actions. : ;

[0057] Among them, is the number of matching edges in the set of final stable matching relationships.

[0058] Optionally, S7 includes the following steps:

[0059] S71. Based on the comprehensive stability metric value , combined with the teaching evaluation requirements at different stability levels in the acupuncture teaching scenario, set a stability scoring interval and establish a hierarchical scoring rule to convert the numerical stability metric result into a discrete level feedback, forming a stability scoring result. ;

[0060] S72. For the stability scoring result , combined with the matching pairs in the set of final stable matching relationships , calculate the set of difference vectors between each pair of trainee trajectories and the standard trajectory in the co-phase space , and extract the following feedback information based on the set of difference vectors:

[0061] Structure deviation description: According to the average Euclidean distance value between co-state vectors, judge the overall deviation degree of the trajectory;

[0062] Stability defect location: Based on the difference in local Lyapunov exponents, locate the key time periods of unstable fluctuations;

[0063] Force control anomaly detection: Combine the direction of the vector with the largest difference in the axial force embedding dimension to identify the physical channel where instability occurs;

[0064] S73. Integrate the stability scoring result , the structure deviation description, the stability defect location, and the force control anomaly detection result to generate a structured evaluation report for teaching feedback, and push it to the trainee side and the teacher side through the interaction interface of the teaching platform to achieve a feedback closed-loop.

[0065] Optionally, the hierarchical scoring rule divides the interval according to the value range of the comprehensive stability metric value , and correspondingly sets five levels: A, B, C, D, E. The specific rules are as follows:

[0066] Level A: , the structure of the acupuncture movement trajectory is higher than the first coincidence threshold, the dynamic stability is extremely high, and the action execution is highly consistent with the standard action;

[0067] Level B: , the structure of the acupuncture motion trajectory is higher than the second anastomosis threshold, and there are small local dynamic deviations;

[0068] Level C: , the structure of the acupuncture motion trajectory is higher than the third anastomosis threshold, and it is necessary to locally correct the motion path or the force control rhythm;

[0069] Level D: , the structure of the acupuncture motion trajectory is lower than the third anastomosis threshold, and there are multi-stage trajectory offsets or severe unstable points;

[0070] Level E: , the structure of the acupuncture motion trajectory is lower than the fourth anastomosis threshold, and there are significant teaching risks or operation misunderstandings.

[0071] An intelligent analysis-based acupuncture teaching assistance platform, which is applied to the intelligent analysis-based acupuncture teaching assistance method, includes the following modules:

[0072] The multi-modal motion acquisition module is used to collect the three-axis displacement information, three-axis attitude information and axial force information of the students during the acupuncture teaching process, and output a multi-modal motion original data set;

[0073] The signal preprocessing and synchronization module is used to perform wavelet threshold filtering, linear interpolation resampling and timestamp alignment operations on the multi-modal motion original data set to generate a preprocessed motion signal data set;

[0074] The delay parameter and embedding dimension estimation module is used to calculate the delay time parameter set and embedding dimension set for each modal information in the preprocessed motion signal data set based on the average mutual information function and the pseudo-nearest neighbor ratio function respectively;

[0075] The collaborative phase space reconstruction module is used to perform multi-dimensional embedding reconstruction on each modal signal according to the delay time parameter set and embedding dimension set, construct a collaborative phase space state vector sequence integrating three-axis displacement, three-axis attitude and axial force, and output a phase space reconstruction trajectory set;

[0076] The motion trajectory segmentation and feature extraction module is used to divide the acupuncture insertion, staying, and withdrawal stages according to the derivative change trend of the three-axis displacement information, and extract the collaborative state vector sequence, state transition direction vector sequence and trajectory curvature sequence within each stage to construct a set of student motion trajectory segment objects;

[0077] The standard motion processing module is used to extract standard acupuncture motion data from the teaching reference library and perform the same processing steps as the student data to generate a set of standard motion trajectory segment objects, and construct a bilateral preference sequence based on the structure similarity and stability index;

[0078] Improve the stable marriage matching module, which is used to execute an improved stable marriage algorithm that introduces local Lyapunov exponents and historical matching feedback regulation. Under the multi-round preference iteration and dynamic weight update mechanism, generate the final set of stable matching relationships, calculate the local stability values of each pair of matching edges, and summarize them to form the comprehensive stability metric value;

[0079] The teaching feedback generation module is used to generate a stability score result by mapping the comprehensive stability metric value to a preset scoring range, and combine the difference vector set, structural deviation, stability defect, and force control anomaly to generate a structured evaluation report, which is pushed to the student side and the teacher side through the teaching platform interface to achieve a feedback closed-loop.

[0080] The beneficial effects of the present invention are:

[0081] Based on the traditional action signal reconstruction method, the present invention introduces the coupling relationship of triaxial displacement information, triaxial attitude information, and axial force information. By constructing a collaborative phase space state vector and calculating the trajectory neighborhood density and local Lyapunov exponents, it comprehensively reflects the temporal evolution law and stability characteristics during the action process, and significantly enhances the quantitative modeling ability of the core indicators of action stability while maintaining data integrity.

[0082] The present invention proposes to construct a bilateral preference sequence based on structural similarity + average trajectory curvature difference, and on this basis, design a stable marriage matching algorithm that dynamically adjusts weights and historical feedback factors, which can achieve an adaptive preference update mechanism and multi-round matching iteration, effectively improving the global stability of action segment matching, and significantly superior to the traditional static scoring-based matching method.

[0083] The present invention structurally models the core indicators of local Lyapunov difference and collaborative state vector difference obtained in the matching results, generates a feedback report including multiple dimensions such as "structural deviation description", "stability defect location", and "force control anomaly detection", and sets a five-level scoring system based on the stability metric Stotal, realizing a closed-loop output from quantitative indicators to teaching suggestions, which can accurately identify the unstable paragraphs and weak control points in the students' operations, and the teaching intervention is more targeted. Description of the Drawings

[0084] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0085] Figure 1 It is a flowchart of an acupuncture teaching assistance platform and method based on intelligent analysis proposed by the present invention;

[0086] Figure 2Flow chart for constructing the multi-modal collaborative phase space state vector and calculating the local Lyapunov exponent of an acupuncture teaching assistance platform and method based on intelligent analysis proposed by the present invention;

[0087] Figure 3 Flow chart for matching by an improved stable marriage algorithm introducing a dynamic weight and local stability regulation mechanism for an acupuncture teaching assistance platform and method based on intelligent analysis proposed by the present invention. Detailed implementation manners

[0088] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0089] Refer to Figure 1 , an acupuncture teaching assistance method based on intelligent analysis, comprising the following steps:

[0090] S1. Collect a multi-modal action data set for the whole process of acupuncture teaching, and perform preprocessing operations to output a preprocessed action signal data set;

[0091] S2. Calculate the delay time parameter and the embedding dimension parameter for the preprocessed action signal data set to obtain a delay parameter set and an embedding dimension value;

[0092] S3. Perform phase space reconstruction on the preprocessed action signal data set according to the delay parameter set and the embedding dimension value to form a phase space reconstruction trajectory set;

[0093] S4. Based on the phase space reconstruction trajectory set, segment the action segments according to the acupuncture needle insertion stage, the staying stage, and the needle withdrawal stage, extract local dynamic characteristics, and generate an action trajectory segment object set;

[0094] S5. Construct a standard action trajectory segment object set according to the teaching reference library, use the action trajectory segment object set and the standard action trajectory segment object set as the two sides of the matching, and construct a bilateral preference sequence according to the dynamic similarity index and the stability difference index;

[0095] S6. Perform an improved stable marriage algorithm on the bilateral preference sequence to obtain a comprehensive stability metric value;

[0096] S7. Map the comprehensive stability metric value to a preset scoring interval to generate a stability scoring result, and push the stability scoring result and the corresponding action gap information to the teaching platform to complete the feedback.

[0097] In this embodiment, S1 includes the following steps:

[0098] S11. Use a multi-modal acquisition device to synchronously collect signals throughout the acupuncture teaching process, obtain three-axis displacement information, three-axis attitude information, and axial force information, and form a multi-modal action original data set ;

[0099] ;

[0100] Among them, represents the single-frame multi-modal action data at the th sampling moment, represents the sampling timestamp, represents the three-axis displacement vector, reflecting the position of the hand or the needle body in space, represents the three-axis attitude angles, corresponding to the pitch angle, yaw angle, and roll angle respectively, characterizing the attitude state of the needle body, represents the instantaneous force along the needle axis direction, reflecting the mechanical response during the insertion process, represents the total number of sampling frames;

[0101] S12. Perform wavelet threshold denoising on the multi-modal action original data set, retain the main frequency characteristics of the signal, and perform linear interpolation resampling to align all modal data to a unified time resolution , and through timestamp consistency alignment, form an action signal data set after time alignment, and use the action signal data set after time alignment as the preprocessed action signal data set .

[0102] In this embodiment, S2 includes the following steps:

[0103] S21. For the three-axis displacement information, three-axis attitude information, and axial force information in the preprocessed action signal data set, respectively use the mean mutual information function to estimate the optimal delay time parameter, and find the time point of the first appearance of information independence in the signal during the time shift by calculating the statistical correlation of the signal at different time intervals . Based on the time point of the first appearance of information independence, determine the delay time parameter of the three-axis displacement information, the delay time parameter of the three-axis attitude information, and the delay time parameter of the axial force information respectively. The three together form a delay time parameter set ;

[0104] ;

[0105] Among them, represents the values of the signal at different moments at the time interval ​ , The joint probability distribution of , are their respective marginal distributions, is the candidate delay time parameter; so that the average mutual information function reaches the local minimum for the first time corresponding to is used as the optimal delay time parameter of the current signal;

[0106] S22. After obtaining the set of delay time parameters , for each type of signal component, the pseudo-nearest neighbor ratio function is used to determine the optimal embedding dimension. By comparing the relative distance changes between adjacent trajectory points in the reconstructed phase space under different embedding dimensions, the proportion of pseudo-nearest neighbors is identified. When the pseudo-nearest neighbor ratio at a certain embedding dimension drops below the threshold, it is considered that the original nonlinear structure has been sufficiently unfolded at this embedding dimension. The embedding dimensions of the three-axis displacement information , the embedding dimension of the three-axis attitude information and the embedding dimension of the axial force information are obtained respectively, and the three together form the embedding dimension set :

[0107] ;

[0108] where is the reconstructed vector of dimension , is its nearest neighbor, is the delay time parameter, represents whether there is a pseudo-nearest neighbor for the th vector in the th dimension; when the value of the pseudo-nearest neighbor ratio function is less than the threshold , the current is selected as the embedding dimension.

[0109] In this embodiment, S3 includes the following steps:

[0110] S31. Based on the set of delay time parameters and the set of embedding dimensions, for the preprocessed action signal dataset, a multi-modal collaborative reconstructed phase space that incorporates the dynamic stability constraints of acupuncture actions is constructed. The dynamic interaction between the three-axis displacement information, the three-axis attitude information, and the axial force information is introduced as the collaborative reconstruction constraint condition on the basis of the independent reconstruction of a single signal;

[0111] S32. For the data at the th moment in the preprocessed action signal dataset, a collaborative phase space state vector that incorporates the multi-dimensional information coupling of acupuncture actions is constructed :

[0112] ;

[0113] Among them, represents the value of the three-axis displacement information at the th moment, where is the delay time parameter of the three-axis displacement information, , which is used to construct the time-delay embedding sequence of the three-axis displacement information, is the embedding dimension of the three-axis displacement information, indicating how many delayed embedding components are used to construct the state space trajectory of the signal, represents the value of the three-axis attitude information at the th moment, is the delay time parameter of the three-axis attitude information, , which is used to construct the time-delay embedding sequence of the attitude signal, is the embedding dimension of the three-axis attitude information, indicating the dimension used when constructing the phase space of the signal, represents the value of the axial force information at the th moment, is the delay time parameter of the axial force, , which is used to construct the time-delay embedding sequence of the force signal, is the embedding dimension of the axial force information;

[0114] The collaborative phase space state vector is formed by splicing the sub-vectors obtained by embedding the three modal signals under their respective optimal delay times and embedding dimensions, and is used to capture the collaborative dynamic characteristics of the acupuncture action in terms of spatial movement, attitude change, and mechanical feedback;

[0115] S33. Calculate the trajectory neighborhood density distribution of the multi-dimensional information of the acupuncture action in the collaborative phase space based on the collaborative phase space state vector , and the trajectory neighborhood density distribution reflects the stability degree of the trainee's acupuncture action in the dynamic state space: ; ;

[0116] Among them, is the volume of the neighborhood sphere centered on the current collaborative phase space state vector , is the Heaviside step function, is the scale of the collaborative phase space state vector set; in the topic of analyzing the stability of acupuncture actions in acupuncture teaching;

[0117] S34. According to the trajectory neighborhood density distribution ​Computing local Lyapunov exponents of cooperative phase space trajectories , analyze the local stability of the dynamic evolution process of acupuncture action: ;

[0118] in, is the sampling interval, is the frame increment of the track along the time axis, Represents the average of all point pairs in the neighborhood of the current point;

[0119] In the topic of acupuncture movement stability analysis in acupuncture teaching, local Lyapunov exponent Used to measure the local dynamic stability of students' acupuncture movements in real time;

[0120] S35. The coordinated phase space state vector , trajectory neighborhood density distribution and the local Lyapunov exponent Together they form a set of phase space reconstruction trajectories with acupuncture action dynamic stability constraints. .

[0121] In this implementation, S4 includes the following steps:

[0122] S41. According to the phase space reconstruction trajectory set and the first-order derivative of the three-axis displacement information in the axial direction, the three key stages in the whole acupuncture process are identified, including the needle insertion stage, the dwelling stage and the needle removal stage. By analyzing the change trend of the displacement in the time series, the interval boundary of the speed continuously rising, stabilizing to zero or continuously decreasing is determined, and the time period index of the corresponding stage is dynamically divided, and the time period index is respectively classified into the needle insertion stage index set, the dwelling stage index set and the needle removal stage index set. According to the three index sets, the corresponding cooperative phase space state vector sequence is extracted from the phase space reconstruction trajectory set, and the needle insertion stage trajectory sequence, the dwelling stage trajectory sequence and the needle removal stage trajectory sequence are respectively formed to form a stage trajectory sequence set;

[0123] S42. For each stage trajectory sequence in the stage trajectory sequence set, the state vectors of two adjacent cooperative phase spaces are subtracted in time order to obtain a state transfer direction vector sequence in the current stage, and the state transfer direction vector sequence reflects the evolution direction trend of the acupuncture action in the cooperative phase space in the stage;

[0124] S43. In each stage, the angle information between the adjacent state transfer direction vectors is used to calculate the trajectory curvature sequence of the trajectory of that stage. The larger the trajectory curvature value, the more obvious the curvature change of the trajectory of that stage, which reflects the instability of the acupuncture action trajectory and is used to measure the local nonlinearity of the evolution trajectory of the action in the collaborative phase space.

[0125] S44. Combine the co-phase space state vector sequence, state transition direction vector sequence, and trajectory curvature sequence corresponding to each stage to form a needle insertion stage action trajectory segment object. The needle insertion stage action trajectory segment object includes a needle insertion stage trajectory sequence, a state transition direction vector sequence, and a needle insertion stage curvature sequence. The staying stage action trajectory segment object includes a staying stage trajectory, a staying stage direction vector, and a staying stage curvature sequence. The needle withdrawal stage action trajectory segment object includes a needle withdrawal stage trajectory, a needle withdrawal stage direction vector, and a needle withdrawal stage curvature sequence;

[0126] S45. Combine the action trajectory segment objects corresponding to the needle insertion stage, staying stage, and needle withdrawal stage to form an action trajectory segment object set.

[0127] In this embodiment, S5 includes the following steps:

[0128] S51. Call the standard acupuncture action data from the teaching reference library. The standard acupuncture action data is composed of multi-modal information during the operation of an acupuncture expert, including three-axis displacement information, three-axis attitude information, and axial force information, and is consistent with the format of the student's preprocessed action signal dataset, denoted as the standard action original dataset ;

[0129] S52. Perform the same processing steps as the student data processing process on the standard action original dataset Finally, generate a standard action trajectory segment object set , including the standard action trajectory segment objects of the needle insertion stage, staying stage, and needle withdrawal stage, which are respectively composed of a standard co-state vector sequence, a standard state transition direction vector sequence, and a standard trajectory curvature sequence;

[0130] S53. Take the student action trajectory segment object set and the standard action trajectory segment object set as the two sides of the matching respectively, and construct the student-side preference sequence. The preference degree of the student action trajectory segment for the standard action trajectory segment is jointly determined by two dimensions: one is the structural similarity between the co-state vector sequences, which is used to measure the consistency of the action trajectory evolution path, and the other is the difference between the average trajectory curvatures, which is used to reflect the deviation of the action trajectory in terms of stability; the two are combined by weighting to form the student-side preference scoring result, and the student-side preference sequence is generated by arranging in descending order according to the preference score;

[0131] S54. Construct the standard-side preference sequence based on the same method. When generating the preference for the standard action trajectory segment, the preference score is also formed according to the structural similarity of the co-state vector sequence and the difference in average trajectory curvature, and the standard-side preference sequence is generated by sorting according to the preference score. The student-side preference sequence and the standard-side preference sequence together form a bilateral preference sequence.

[0132] In this embodiment, S6 includes the following steps:

[0133] S61. Based on the bilateral preference sequence and the specific requirements for the analysis of the stability of acupuncture teaching actions, an improved stable marriage algorithm that incorporates the constraints of acupuncture action dynamics characteristics and the adaptive adjustment of the stage action stability weight is constructed. The improved stable marriage algorithm introduces a dynamic update mechanism for the dynamic preferences between acupuncture action segments, and updates the preference list in real time during each round of matching to meet the requirements for the evaluation of action stability in the acupuncture teaching process;

[0134] S62. In the round of matching, for each trainee action trajectory segment object , the preference value is dynamically updated according to its historical matching status with each standard action trajectory segment object :

[0135] ;

[0136] Among them, , are the collaborative state vector sequences of the th trainee action trajectory segment object and the th standard action trajectory segment object in the matching edge, respectively, reflecting the structural similarity degree between acupuncture action trajectories, represents the cosine similarity of the trajectory sequence, , represent the average trajectory curvatures of the corresponding trajectory segments respectively, is the fixed weight of the trajectory structure similarity, is the dynamic weight, which is adaptively adjusted as the number of matching rounds increases, represents the cumulative historical number of rejections or acceptances of the trainee action trajectory segment object for the standard action trajectory segment object in the previous round of matching, is the adjustment factor for the influence of the historical rejection or acceptance times on the preference value update;

[0137] S63. In each round of stable matching process of the round of matching, the local stability values of each matching edge are calculated and recorded in real time: ;

[0138] Among them, is the local Lyapunov exponent, indicating the trainee action trajectory segment object for the standard action trajectory segment object The mean of the local Lyapunov exponents, which is used to characterize the differences in local dynamic stability between two trajectory segments in the collaborative phase space. The introduction of the exponential function is used to emphasize the decisive role of the local stability characteristics of the acupuncture action dynamics in the stable marriage matching process;

[0139] S64. When the absolute value of the difference between the local stability values corresponding to the matching edges between two adjacent rounds of matching is lower than the preset convergence threshold the improved stable marriage algorithm stops iterating and determines the set of final stable matching relationships and outputs the corresponding global one-to-many stable matching relationships;

[0140] S65. Weighted aggregation is performed on the local stability values of all matching edges in the set of final stable matching relationships to obtain the comprehensive stability metric value for the analysis of the stability of acupuncture actions in acupuncture teaching :

[0141] ; where is the number of matching edges in the set of final stable matching relationships.

[0142] In this embodiment, S7 includes the following steps:

[0143] S71. Based on the comprehensive stability metric value and combined with the teaching evaluation requirements for different stability levels in the acupuncture teaching scenario, set the stability scoring interval and establish a hierarchical scoring rule to convert the numerical stability metric result into a discrete grade feedback to form the stability scoring result ;

[0144] S72. For the stability scoring result and combined with the matching pairs in the set of final stable matching relationships calculate the set of difference vectors between each pair of trainee trajectories and the standard trajectory in the collaborative phase space, and extract the following feedback information based on the set of difference vectors:

[0145] Structural deviation description: Determine the overall deviation degree of the trajectory according to the average Euclidean distance value between the collaborative state vectors;

[0146] Stability defect localization: Locate the key time periods where unstable fluctuations occur based on the differences in local Lyapunov exponents;

[0147] Force control anomaly detection: Identify the physical channels where instability occurs by combining the direction of the vector with the largest difference in the axial force embedding dimension;

[0148] S73. The stability scoring result , Integrate the structural offset description, stability defect location, and force control anomaly detection results to generate a structured evaluation report for teaching feedback, and push it to the student side and teacher side through the interaction interface of the teaching platform to achieve a feedback closed-loop.

[0149] In this embodiment, the hierarchical scoring rule is based on the value range of the comprehensive stability metric value to divide the interval and correspondingly set five grades: A, B, C, D, E. The specific rules are as follows:

[0150] Grade A: , the structure of the acupuncture action trajectory is higher than the first coincidence threshold, the dynamic stability is extremely high, and the action execution is highly consistent with the standard action;

[0151] Grade B: , the structure of the acupuncture action trajectory is higher than the second coincidence threshold, and there are small local dynamic deviations;

[0152] Grade C: , the structure of the acupuncture action trajectory is higher than the third coincidence threshold, and it is necessary to locally correct the action path or force control rhythm;

[0153] Grade D: , the structure of the acupuncture action trajectory is lower than the third coincidence threshold, and there are multi-stage trajectory offsets or severe instability points;

[0154] Grade E: , the structure of the acupuncture action trajectory is lower than the fourth coincidence threshold, and there are significant teaching risks or operation misunderstandings.

[0155] An intelligent analysis-based acupuncture teaching assistance platform is applied to an intelligent analysis-based acupuncture teaching assistance method, including the following modules:

[0156] The multi-modal action acquisition module is used to collect the three-axis displacement information, three-axis attitude information, and axial force information of the students during the acupuncture teaching process, and output to form a multi-modal action original data set;

[0157] The signal preprocessing and synchronization module is used to perform wavelet threshold filtering, linear interpolation resampling, and timestamp alignment operations on the multi-modal action original data set to generate a preprocessed action signal data set;

[0158] The delay parameter and embedding dimension estimation module is used to calculate the delay time parameter set and embedding dimension set for each modal information in the preprocessed action signal data set respectively based on the average mutual information function and the pseudo-nearest neighbor ratio function;

[0159] The collaborative phase space reconstruction module is used to perform multi-dimensional embedding reconstruction of each modal signal according to the delay time parameter set and the embedding dimension set, construct a collaborative phase space state vector sequence integrating the three-axis displacement, three-axis posture and axial force, and output a phase space reconstruction trajectory set;

[0160] The motion trajectory segmentation and feature extraction module is used to divide the needle insertion, retention and needle removal into three stages according to the derivative change trend of the three-axis displacement information, and extract the coordinated state vector sequence, state transfer direction vector sequence and trajectory curvature sequence in each stage to construct the student's motion trajectory segment object set;

[0161] The standard action processing module is used to extract standard acupuncture action data from the teaching reference library and perform processing steps consistent with the student data, generate a set of standard action trajectory fragment objects, and construct a bilateral preference sequence based on structural similarity and stability indicators;

[0162] Improved stable marriage matching module, used to execute the improved stable marriage algorithm that introduces local Lyapunov exponent and historical matching feedback adjustment, generate the final stable matching relationship set under multiple rounds of preference iteration and dynamic weight update mechanism, and calculate the local stability value of each pair of matching edges, and summarize them to form a comprehensive stability measurement index value;

[0163] The teaching feedback generation module is used to generate stability scoring results based on the preset scoring interval mapped by the comprehensive stability metric index value, and generate a structured evaluation report based on the difference vector set, structural offset, stability defects and force control anomalies. The report is pushed to the student and teacher ends through the teaching platform interface to achieve a closed feedback loop.

[0164] Example 1

[0165] In Training Room No. 6 of the Experimental Teaching Center of the College of Acupuncture and Massage of A City University of Traditional Chinese Medicine, the acupuncture simulation teaching session started on time. The theme of the training course was "Standard Operation Procedure Training of Zusanli Acupoint". The participants were 14 undergraduate students from the 21st grade of the "Basic Operation Class of Acupuncture". The operation supervision and data analysis were supported by the intelligent acupuncture teaching platform VerAI-TCM V1.4. The main lecturer was Associate Professor Gao Haitao, who has 11 years of teaching experience.

[0166] Trainee B performed the first needle insertion training at operating station No. 3. The platform equipment access number was "APT-TCM-003", the acquisition serial number was "LZM-1", and the system multimodal collector started working, continuously recording timestamps, three-axis position vectors, posture angles, and axial forces. The sampling frequency was set to 120 Hz, the total action duration was 32.4 seconds, and the total number of sampled frames was 3888.

[0167] The platform automatically recognizes the needle insertion stage. The starting frame is the 48th frame. The system prompts "The action enters the fast stage". Later, the system monitors that the acceleration of its Z-axis displacement is higher than the upper limit threshold of the standard sample (standard: 0.42 m / s², measured value: 0.61 m / s²), and the attitude angle θ fluctuates continuously by more than 5.8 degrees within 1.6 seconds. The local Lyapunov exponent value rises to 0.0387 (the standard threshold is below 0.025). The system background automatically marks this period as a "highly unstable section" in the "local unstable state list".

[0168] Meanwhile, the matching engine performs trajectory matching based on the standard sample "STD-FZL2020". The system log shows that within the frame number #1208 - #1452 section, the average value of the state vector difference is 1.183 (unitless), and the structure deviation score is 3.6 (score range: 0 - 10). The system automatically executes the 3rd round of matching of the improved stable marriage algorithm, outputs the preference update value Pi,j(r)=2.73, and finally confirms that the deviation range of the paired action sequence from the standard sample trajectory is moderate, generating a scoring grade "C".

[0169] The system pops up a window to prompt the feedback report. The teacher - end report ID is "FBK - 003 - LZM1", and it contains the following structural content:

[0170] Action paragraph: 9:02:21 9:02:25; Abnormal type: Loss of control of attitude stability; Cause analysis: Attitude control error + sudden change in axial force application; Suggestion: During the needle insertion, the left - wrist control angle needs to be stabilized within ±3° in the tangent direction, and the continuous force application should not exceed 0.4N / s.

[0171] After receiving the push, the student - end enters the retraining mode of the simulation operation console.

[0172] The second - round training number is "LZM - 2". The platform samples 3620 frames of data again. The total operation duration is 29.6 seconds. After the platform reconstructs the phase space, the monitored data shows that the average value of the overall Lyapunov exponent drops to 0.0198, the needle - insertion stability is significantly improved, the change in the attitude angle is controlled within 2.1°, and the standard deviation of the trajectory curvature drops to 0.0075 (compared with 0.0152 in the first time). The scoring system updates the stability metric value Stotal to 0.803, and the stability level is upgraded to "B+".

[0173] In the overall class summary analysis, the average increase in Stotal for the second - round operation of the intelligent system group (7 people using the method of the present invention) is 0.174, while that of the traditional teaching group (oral guidance by teachers) is only 0.062 on average; the "average positioning time of the problem paragraph" marked by the system is 0.46 seconds, and the average time visually estimated by the teacher group is 3.2 seconds, with an obvious difference in accuracy (95.1% vs 62.8%).

[0174] Table 1 shows the comparison and summary of some data exported from the experimental platform

[0175] Student ID First Stotal Score Second Stotal Score Number of Feedback in Problem Sections Average Duration for Error Location (seconds) LZM 0.578 0.803 2 0.42 DXT 0.613 0.782 1 0.53 AXL 0.547 0.762 2 0.49 FYR 0.598 0.801 1 0.37

[0176] Table 2 shows the data of the comparison with the traditional group

[0177] Student ID First Stotal Score Second Stotal Score B 0.591 0.624 C 0.572 0.618

[0178] Example 1 fully verifies three key problems solved by the technology of the present invention: First, there is a lack of dynamic process modeling in traditional teaching, making it difficult to identify abnormal fluctuations in specific time periods during operations; second, the oral feedback from teachers is lagged and rough, lacking an accurate dynamic positioning mechanism; third, there is a lack of a data-driven scoring closed-loop, making it difficult to quantify the improvement effect. The present invention effectively fills the above technical gaps through three algorithmic links of collaborative phase space analysis, stable marriage matching, and structured feedback, significantly improving the teaching real-time performance, accuracy, and the self-correction efficiency of trainees.

[0179] Based on the traditional action signal reconstruction method, the present invention introduces the coupling relationship of triaxial displacement information, triaxial attitude information, and axial force information. By constructing a collaborative phase space state vector and calculating the trajectory neighborhood density and local Lyapunov exponent, it comprehensively reflects the temporal evolution law and stability characteristics during the action process, and significantly enhances the quantitative modeling ability of the core index of action stability while maintaining data integrity.

[0180] The present invention proposes to construct a bilateral preference sequence based on structural similarity + average trajectory curvature difference, and on this basis, designs a stable marriage matching algorithm that dynamically adjusts weights and historical feedback factors, which can achieve an adaptive preference update mechanism and multi-round matching iterations, effectively improving the global stability of action segment matching, and significantly superior to the traditional static scoring-based matching method.

[0181] The present invention generates a feedback report containing multiple dimensions such as "structural deviation description", "stability defect positioning", and "force control anomaly detection" by structurally modeling the core indexes of local Lyapunov difference and collaborative state vector difference obtained in the matching results, and sets a five-level scoring system based on the stable metric index Stotal, realizing a closed-loop output from quantitative indexes to teaching suggestions, which can accurately identify the unstable segments and weak control points existing in the trainees' operations, and the teaching intervention is more targeted.

[0182] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An acupuncture teaching assistance method based on intelligent analysis, characterized in that It includes the following steps: S1. Collect a multi-modal action data set for the whole process of acupuncture teaching, and perform preprocessing operations to output a preprocessed action signal data set; S2. Calculate the delay time parameter and the embedding dimension parameter for the preprocessed action signal data set to obtain a set of delay parameters and an embedding dimension value; S3. Perform phase space reconstruction on the preprocessed action signal data set according to the set of delay parameters and the embedding dimension value to form a set of phase space reconstruction trajectory sets; S4. Based on the set of phase space reconstruction trajectories, segment the action segments according to the acupuncture needle insertion stage, the staying stage, and the needle withdrawal stage, extract local dynamic characteristics, and generate a set of action trajectory segment objects; S5. Construct a set of standard action trajectory segment objects according to the teaching reference library, use the set of action trajectory segment objects and the set of standard action trajectory segment objects as the two parties for matching, and construct a bilateral preference sequence according to the dynamic similarity index and the stability difference index; S6. Execute an improved stable marriage algorithm based on the bilateral preference sequence to obtain a comprehensive stability metric value; The S6 includes the following steps: S61. Based on the bilateral preference sequence, for the specific requirements of the stability analysis of acupuncture teaching acupuncture actions, construct an improved stable marriage algorithm that incorporates the dynamic characteristics constraints of acupuncture actions and the adaptive adjustment of the stage action stability weights. The improved stable marriage algorithm introduces a dynamic update mechanism for the dynamic preferences between acupuncture action segments, and updates the preference list in real time during each round of matching to meet the requirements of the action stability evaluation during the acupuncture teaching process; S62. During the r-th round of matching process, for each trainee action trajectory segment object , dynamically update the preference value according to its historical matching status with each standard action trajectory segment object : ； Among them, and are the collaborative state vector sequences of the -th trainee action trajectory segment object and the -th standard action trajectory segment object in the matching edges, respectively, reflecting the structural similarity degree between acupuncture action trajectories. represents the cosine similarity of the trajectory sequences, and represent the average trajectory curvatures of the corresponding trajectory segments, respectively. is the fixed weight of the trajectory structure similarity, is the dynamic weight, which is adaptively adjusted as the number of matching rounds increases. represents the cumulative historical number of rejections or acceptances of the trainee action trajectory segment object to the standard action trajectory segment object in the previous round of matching. is the adjustment factor for the influence of the historical rejection or acceptance times on the preference value update. S63. In each round of the stable matching process of the round of matching, the local stability values of each matching edge are calculated and recorded in real time: ； Among them, is the local Lyapunov exponent, representing the object of the trainee's action trajectory segment and the object of the standard action trajectory segment The mean value of the local Lyapunov exponents is used to characterize the difference in local dynamic stability between the two trajectory segments in the cooperative phase space. The introduction of the exponential function is used to emphasize the decisive role of the local stability characteristics of the acupuncture action dynamics in the stable marriage matching process; S64. When the local stability value corresponding to the matching edge has an absolute value of the difference between adjacent two rounds of matching lower than a preset convergence threshold the improved stable marriage algorithm stops iterating, determines the set of final stable matching relationships and outputs the corresponding global one-to-many stable matching relationship; S65. For the set of final stable matching relationships weightedly aggregate the local stability values of all matching edges in it to obtain the comprehensive stability metric value for the stability analysis of acupuncture teaching acupuncture actions : ； Among them, is the number of matching edges in the final stable matching relationship set; S7. Map the preset scoring interval according to the comprehensive stability metric value to generate a stability scoring result, and push the stability scoring result and the corresponding action gap information to the teaching platform to complete the feedback.

2. The acupuncture teaching assistance method based on intelligent analysis according to claim 1, characterized in that The S1 includes the following steps: S11. Use a multi-modal acquisition device to synchronously collect signals throughout the acupuncture teaching process, obtain three-axis displacement information, three-axis attitude information, and axial force information, and form a multi-modal action original data set ; S12. Perform wavelet threshold denoising on the multi-modal action original dataset, retain the main frequency characteristics of the signal, and perform linear interpolation resampling to align all modal data to a unified time resolution , and align them through timestamp consistency to form an action signal dataset after time alignment. Take the action signal dataset after time alignment as the preprocessed action signal dataset .

3. The acupuncture teaching assistance method based on intelligent analysis according to claim 2, characterized in that, The S2 includes the following steps: S21. For the triaxial displacement information, triaxial attitude information, and axial force information in the preprocessing action signal dataset, respectively use the average mutual information function-based to estimate the optimal delay time parameter, and find the time point when the information independence first appears during the time evolution of the signal by calculating the statistical correlation of the signal at different time intervals . Based on the time point when the information independence first appears, determine the delay time parameter of the triaxial displacement information , the delay time parameter of the triaxial attitude information , and the delay time parameter of the axial force information . The three together form a delay time parameter set ; S22. After obtaining the set of delay time parameters for each type of signal component, the pseudo-nearest neighbor ratio function is used to determine the optimal embedding dimension. By comparing the relative distance changes between neighboring trajectory points in the reconstructed phase space under different embedding dimensions, the proportion of pseudo-nearest neighbors is identified. When the pseudo-nearest neighbor ratio at a certain embedding dimension drops below the threshold, it is considered that the original nonlinear structure has been fully unfolded at this embedding dimension, and the embedding dimensions of the three-axis displacement information , the embedding dimension of the three-axis attitude information and the embedding dimension of the axial force information are obtained respectively. The three together form the embedding dimension set .

4. The acupuncture teaching assistance method based on intelligent analysis according to claim 3, characterized in that, The S3 includes the following steps: S31. Based on the set of delay time parameters and the set of embedding dimensions, for the preprocessed action signal data set, construct a multi-modal collaborative reconstruction phase space that incorporates the dynamic stability constraints of acupuncture actions. On the basis of the independent reconstruction of a single signal, introduce the dynamic interaction between the three-axis displacement information, the three-axis attitude information, and the axial force information as the collaborative reconstruction constraint conditions; S32. For the data at the -th moment in the preprocessing action signal dataset, construct a collaborative phase space state vector that fuses multi-dimensional information coupling of acupuncture actions : ； Among them, represents the value of the three-axis displacement information at the -th moment, where is the delay time parameter of the three-axis displacement information, , which is used to construct the time-delay embedding sequence of the three-axis displacement information, is the embedding dimension of the three-axis displacement information, indicating how many delayed embedding components are used to construct the state space trajectory of the signal, represents the value of the three-axis attitude information at the -th moment, is the delay time parameter of the three-axis attitude information, , which is used to construct the time-delay embedding sequence of the attitude signal, is the embedding dimension of the three-axis attitude information, indicating the dimension used when constructing the phase space of the signal, represents the value of the axial force information at the -th moment, is the delay time parameter of the axial force, , which is used to construct the time-delay embedding sequence of the force signal, is the embedding dimension of the axial force information; S33. Calculate the trajectory neighborhood density distribution of multi-dimensional information of acupuncture actions in the collaborative phase space based on the collaborative phase space state vector The trajectory neighborhood density distribution reflects the stability of the trainee's acupuncture actions in the dynamic state space: The trajectory neighborhood density distribution reflects the stability of the trainee's acupuncture actions in the dynamic state space: ； Among them, is the volume of the -neighborhood sphere centered at the current collaborative phase space state vector ; is the Heaviside step function, is the scale of the set of collaborative phase space state vectors; in the topic of analyzing the stability of acupuncture actions in acupuncture teaching; S34. According to the trajectory neighborhood density distribution Calculate the local Lyapunov exponent of the collaborative phase space trajectory , and analyze the local stability of the dynamic evolution process of acupuncture actions: ； Among them, is the sampling interval, is the frame number increment by which the trajectory advances along the time axis, represents the average of all point pairs in the neighborhood of the current point; S35. Combine the collaborative phase space state vector , the density distribution of the trajectory neighborhood and the local Lyapunov exponent to jointly form a set of phase space reconstruction trajectories with the stability constraint of acupuncture action dynamics .

5. A method for assisting acupuncture teaching based on intelligent analysis according to claim 4, characterized in that, The S4 includes the following steps: S41. According to the set of phase space reconstruction trajectories and the first derivative of the three-axis displacement information in the axial direction, identify the three key stages in the whole process of acupuncture, including the needle insertion stage, the staying stage, and the needle withdrawal stage. By analyzing the change trend of the displacement in the time series, judge the interval boundaries where the speed continuously rises, stabilizes to zero, or continuously decreases, dynamically divide the time period indexes corresponding to the stages, and classify the time period indexes into the needle insertion stage index set, the staying stage index set, and the needle withdrawal stage index set respectively. According to the three index sets, extract the corresponding collaborative phase space state vector sequences from the set of phase space reconstruction trajectories to form the needle insertion stage trajectory sequence, the staying stage trajectory sequence, and the needle withdrawal stage trajectory sequence respectively, and constitute a set of stage trajectory sequences; S42. For each stage trajectory sequence in the stage trajectory sequence set, the state vectors of two adjacent cooperative phase spaces are subtracted in time order to obtain a state transfer direction vector sequence in the current stage, and the state transfer direction vector sequence reflects the evolution direction trend of the acupuncture action in the cooperative phase space in the stage; S43. In each stage, the angle information between the adjacent state transfer direction vectors is used to calculate the trajectory curvature sequence of the trajectory of that stage. The larger the trajectory curvature value, the more obvious the curvature change of the trajectory of that stage, which reflects the instability of the acupuncture action trajectory and is used to measure the local nonlinearity of the evolution trajectory of the action in the collaborative phase space. S44. The state vector sequence of the coordinated phase space, the state transfer direction vector sequence and the trajectory curvature sequence corresponding to each stage are collectively formed into a needle insertion stage motion trajectory fragment object, wherein the needle insertion stage motion trajectory fragment object includes the needle insertion stage trajectory sequence, the state transfer direction vector sequence and the needle insertion stage curvature sequence, the stay stage motion trajectory fragment object includes the stay stage trajectory, the stay stage direction vector and the stay stage curvature sequence, and the needle removal stage motion trajectory fragment object includes the needle removal stage trajectory, the needle removal stage direction vector and the needle removal stage curvature sequence; S45. Combine the motion track segment objects corresponding to the needle insertion phase, the stay phase, and the needle removal phase to form a motion track segment object set.

6. The acupuncture teaching assistance method based on intelligent analysis according to claim 5, wherein The S5 comprises the following steps: S51. Call the standard acupuncture action data from the teaching reference library. The standard acupuncture action data consists of multi-modal information during the operation of acupuncture experts, including three-axis displacement information, three-axis attitude information, and axial force information, and is consistent with the format of the preprocessed action signal data set of the trainees, denoted as the standard action original data set ; S52. For the original standard action dataset Execute the processing steps consistent with the trainee data processing flow, and finally generate a set of standard action trajectory segment objects , including standard action trajectory segment objects in the needle insertion stage, residence stage, and needle withdrawal stage, which are respectively composed of a standard collaborative state vector sequence, a standard state transition direction vector sequence, and a standard trajectory curvature sequence; S53. The set of trainee action trajectory segment objects and the set of standard action trajectory segment objects are respectively used as the two parties for matching to construct the trainee-side preference sequence. The preference degree of the trainee action trajectory segment for the standard action trajectory segment is jointly determined by two dimensions: one is the structural similarity between the collaborative state vector sequences, which is used to measure the consistency of the action trajectory evolution path, and the other is the difference between the average trajectory curvatures, which is used to reflect the deviation of the action trajectory in terms of stability; the two are combined through weighted combination to form the trainee-side preference scoring result, and the trainee-side preference sequence is generated by arranging them in descending order according to the preference score; S54. The standard side preference sequence is constructed based on the same method. When generating preferences, the standard action trajectory fragments also form a preference score based on the structural similarity of the collaborative state vector sequence and the difference in the average trajectory curvature, and the standard side preference sequence is generated by sorting according to the preference score. The student side preference sequence and the standard side preference sequence together constitute a bilateral preference sequence.

7. An acupuncture teaching assistance method based on intelligent analysis according to claim 6, characterized in that, The S7 comprises the following steps: S71. Based on the comprehensive stability metric value , combined with the teaching evaluation requirements for different stability levels in the acupuncture teaching scenario, set the stability scoring range and establish a hierarchical scoring rule to convert the numerical stability metric result into a discrete level feedback, forming the stability scoring result ; S72. For the stability scoring results , combined with the matching pairs in the final stable matching relationship set , calculate the set of difference vectors between each pair of trainee trajectories and the standard trajectory in the collaborative phase space , and extract the following feedback information based on the set of difference vectors: Structural deviation description: The overall deviation of the trajectory is determined based on the average Euclidean distance between the collaborative state vectors; Stability defect location: locate the key period of unstable fluctuations based on the difference of local Lyapunov exponents; Force control anomaly detection: Combine the direction of the vector with the largest difference in the axial force embedding dimension to identify the physical channel where instability occurs; S73. Integrate the stability scoring results , the structural deviation description, the stability defect location, and the force control anomaly detection results to generate a structured evaluation report for teaching feedback, and push it to the student side and the teacher side through the interaction interface of the teaching platform to achieve a feedback closed-loop.

8. An acupuncture teaching assistance method based on intelligent analysis according to claim 7, characterized in that, The hierarchical scoring rule is based on the value of the comprehensive stability metric The value range is divided into intervals, and five levels are set correspondingly: A, B, C, D, E. The specific rules are as follows: Level A: , the acupuncture motion trajectory structure is higher than the first anastomosis threshold, with extremely high kinetic stability, and the motion execution is highly consistent with the standard motion; Level B: , the acupuncture movement trajectory structure is higher than the second anastomosis threshold, and there are small local dynamic deviations; Level C: , the acupuncture motion trajectory structure is higher than the third anastomosis threshold, and the motion path or force control rhythm needs to be locally corrected; Grade D: , the needle insertion motion trajectory structure is lower than the third anastomosis threshold, with multi-stage trajectory deviation or severe instability points; Level E: , the structure of the acupuncture movement trajectory is lower than the fourth anastomosis threshold, presenting significant teaching risks or operation misunderstandings.

9. An acupuncture teaching assistance platform based on intelligent analysis, which is applied to an acupuncture teaching assistance method based on intelligent analysis according to any one of claims 1-8, and is characterized in that, Includes the following modules: The multimodal motion acquisition module is used to collect the three-axis displacement information, three-axis posture information and axial force information of students during acupuncture teaching, and output the original data set of multimodal motion; The signal preprocessing and synchronization module is used to perform wavelet threshold filtering, linear interpolation resampling and timestamp alignment operations on the original multimodal action data set to generate a preprocessed action signal data set; A delay parameter and embedding dimension estimation module is used to calculate a delay time parameter set and an embedding dimension set for each modal information in the preprocessed motion signal data set based on an average mutual information function and a pseudo neighbor ratio function; The collaborative phase space reconstruction module is used to perform multi-dimensional embedding reconstruction of each modal signal according to the delay time parameter set and the embedding dimension set, construct a collaborative phase space state vector sequence integrating the three-axis displacement, three-axis posture and axial force, and output a phase space reconstruction trajectory set; The motion trajectory segmentation and feature extraction module is used to divide the needle insertion, retention and needle removal into three stages according to the derivative change trend of the three-axis displacement information, and extract the coordinated state vector sequence, state transfer direction vector sequence and trajectory curvature sequence in each stage to construct the student's motion trajectory segment object set; The standard action processing module is used to extract standard acupuncture action data from the teaching reference library and perform processing steps consistent with the student data, generate a set of standard action trajectory fragment objects, and construct a bilateral preference sequence based on structural similarity and stability indicators; Improved stable marriage matching module, used to execute the improved stable marriage algorithm that introduces local Lyapunov exponent and historical matching feedback adjustment, generate the final stable matching relationship set under multiple rounds of preference iteration and dynamic weight update mechanism, and calculate the local stability value of each pair of matching edges, and summarize them to form a comprehensive stability measurement index value; The teaching feedback generation module is used to generate stability scoring results based on the preset scoring interval mapped by the comprehensive stability metric index value, and generate a structured evaluation report based on the difference vector set, structural offset, stability defects and force control anomalies. The report is pushed to the student and teacher ends through the teaching platform interface to achieve a closed feedback loop.

Images