A wearable device-based limb rehabilitation exercise data evaluation system

Abstract

The present application relates to medical rehabilitation monitoring and data processing technical field, specifically to a kind of limb rehabilitation exercise data evaluation system based on wearable device, comprising: phase space reconstruction unit: fusion inertia data constructs state vector, mapping generates phase space state trajectory;Trajectory deviation analysis unit: point cloud is projected to standard reference trajectory, quantization distribution state generates trajectory out-of-bound index;Nonlinear stability analysis unit: based on trajectory evolution calculation maximum Lyapunov index, generates movement stability classification label;Multi-dimensional feature fusion evaluation unit: decoupling identification compensation and pathology by index and label, triggers real-time feedback;Benchmark parameter self-adapting update unit: utilize topological feature to update pipeline parameter;The present application realizes the accurate decoupling and individualized evaluation of rehabilitation exercise quality.

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation monitoring and data processing technology, specifically to a limb rehabilitation movement data assessment system based on wearable devices. Background Technology

[0002] With the development of rehabilitation medicine and smart wearable technology, the demand for quantitative assessment of limb motor function is increasing. However, the complex nonlinear dynamic characteristics contained in limb motor data pose many challenges to the accurate determination of rehabilitation status.

[0003] Currently, threshold-based assessment methods using fundamental kinematic parameters or subjective visual observation by therapists are commonly employed. Existing technologies typically utilize statistical measures such as the amplitude and variance of acceleration or angular velocity to measure movement quality and judge whether a movement meets standards using simple linear indicators. However, traditional assessment methods struggle to delve into the dynamic structural characteristics behind movement time series. Algorithms based solely on velocity or acceleration thresholds often fail to effectively distinguish between pathological tremors and rigid convergence states, easily leading to misjudgments of rehabilitation status. Furthermore, traditional algorithms struggle to identify stable compensatory movements—that is, situations where patients utilize incorrect muscle groups but exhibit relatively stable trajectories—often misjudging them as standard rehabilitation movements, making it difficult to detect latent pathological risks and failing to meet the needs of personalized and precise rehabilitation.

[0004] Therefore, how to deeply explore the dynamic characteristics of motion data, accurately decouple and identify the compensatory types and pathological states of limb movements has become an urgent problem to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a limb rehabilitation exercise data assessment system based on wearable devices. Specifically, the technical solution of this invention includes:

[0006] Phase space reconstruction unit: acquires limb motion time series data collected by wearable devices, which includes triaxial acceleration components and triaxial angular velocity components; calculates the optimal delay time and embedding dimension of the time series data based on mutual information method and false neighbor method; constructs state vector based on optimal delay time and embedding dimension, maps the one-dimensional time series data to a high-dimensional phase space, and generates phase space state trajectory;

[0007] Trajectory Deviation Analysis Unit: Loads a preset standard reference trajectory feature matrix, which includes the pipeline center trajectory line and pipeline radius threshold constructed based on healthy limb movement data; projects the phase space state trajectory into the topological space of the standard reference trajectory; detects the spatial distribution state of the phase space state trajectory relative to the inner wall of the standard reference trajectory, quantifies the degree to which the point cloud distribution exceeds the inner wall of the pipeline, and generates a trajectory boundary crossing index.

[0008] Nonlinear stability analysis unit: Based on the trajectory evolution characteristics of the phase space state trajectory in phase space, calculate the maximum Lyapunov exponent; load a preset dynamic stability threshold range, map the maximum Lyapunov exponent to the corresponding stability level, and generate motion stability classification labels;

[0009] Multidimensional feature fusion assessment unit: Combining trajectory boundary crossing index and motion stability classification label, it decouples and identifies the compensatory type and pathological state of limb movement through multidimensional feature fusion calculation; Based on the identification results, it generates a rehabilitation exercise quality assessment report and triggers real-time feedback instructions;

[0010] The baseline parameter adaptive update unit: In response to the rehabilitation exercise quality assessment report determining that the exercise is in an effective rehabilitation state, it extracts the topological features of the current phase space state trajectory; and uses the topological features to perform a weighted update of the standard reference trajectory feature matrix to generate personalized reference trajectory parameters.

[0011] As a further aspect of the present invention, a state vector is constructed based on the optimal delay time and embedding dimension, and one-dimensional time-domain time-series data is mapped to a high-dimensional phase space to generate phase space state trajectories. The methods include:

[0012] Time series data is defined as a collection of time series data.

[0013] By applying time delay processing to a time series set using the optimal delay time, multiple delayed sequences are generated.

[0014] Based on the embedding dimension, the values ​​of the time series set and the delayed sequence at the same time are extracted to form a high-dimensional state vector;

[0015] Arrange the high-dimensional state vectors of all moments in chronological order to construct the phase space trajectory matrix, and use the phase space trajectory matrix as the phase space state trajectory.

[0016] As a further aspect of the present invention, the spatial distribution state of the phase space state trajectory relative to the inner wall of the standard reference trajectory is detected, the degree to which the point cloud distribution exceeds the inner wall of the pipe is quantified, and the trajectory boundary crossing index is generated by means of:

[0017] Analyze the standard reference trajectory feature matrix to obtain the pipeline center trajectory line and pipeline radius threshold;

[0018] Calculate the Euclidean distance from each point in the phase space state trajectory to the center trajectory line of the pipeline;

[0019] By comparing the Euclidean distance with the pipe radius threshold, escape points where the Euclidean distance is greater than the pipe radius threshold are identified.

[0020] The number of escape points is counted, and the proportion of the number of escape points to the total number of points in the phase space state trajectory is calculated. This proportion is then determined as the trajectory boundary crossing index.

[0021] As a further aspect of the present invention, loading a preset dynamic stability threshold range and mapping the maximum Lyapunov exponent to the corresponding stability level to generate motion stability classification labels includes the following methods:

[0022] Load the preset first entropy threshold and second entropy threshold, wherein the first entropy threshold is greater than the second entropy threshold;

[0023] If the maximum Lyapunov exponent is greater than the first entropy threshold, the motion state is determined to be chaotic divergence, and a risk warning label is generated as a motion stability classification label.

[0024] If the maximum Lyapunov exponent is less than or equal to the first entropy threshold and greater than or equal to the second entropy threshold, the motion state is determined to be a stable period, and a valid training label is generated as a motion stability classification label.

[0025] If the maximum Lyapunov exponent is less than the second entropy threshold, the motion state is determined to be rigid convergence, and an invalid motion label is generated as the motion stability classification label.

[0026] As a further aspect of this invention, by combining the trajectory boundary crossing index and motion stability classification labels, and through multi-dimensional feature fusion calculation, the decoupling identification of limb movement compensation types and pathological states includes:

[0027] Load the preset collision tolerance threshold;

[0028] If the motion stability classification label is a valid training label and the trajectory boundary crossing index is greater than the collision tolerance threshold, then the limb movement is determined to be a stable compensatory movement.

[0029] If the motion stability classification label is a valid training label and the trajectory boundary crossing index is less than or equal to the collision tolerance threshold, then the limb movement is determined to be a standard rehabilitation exercise.

[0030] If the movement stability classification label is a risk warning label, then the trajectory boundary crossing index is ignored, and the limb movement is directly judged to be out of control pathological state;

[0031] If the movement stability classification label is an invalid movement label, then the limb movement is determined to be in a stiff and invalid state.

[0032] As a further aspect of the present invention, the system is integrated into a wearable device, which includes a microprocessor, a storage module, and a wireless transmission module.

[0033] The storage module is used to store the standard reference trajectory feature matrix and various preset threshold parameters;

[0034] The wireless transmission module is used to send the generated rehabilitation exercise quality assessment report and real-time feedback instructions to an external terminal;

[0035] The phase space reconstruction unit utilizes the matrix operation accelerator of the microprocessor to perform phase space mapping operations.

[0036] As a further aspect of the present invention, the methods for triggering real-time feedback commands include:

[0037] In response to the determination that the limb movement is a stable compensatory movement, corrective prompts are generated for the force-producing muscle groups;

[0038] In response to the determination that limb movement is in a pathological state of loss of control, an interruption alarm command is generated to immediately stop training;

[0039] In response to the determination that the limb movement is a standard rehabilitation exercise, an incentive instruction to maintain the training intensity is generated;

[0040] In response to the determination that limb movement is in a stiff and ineffective state, a guiding instruction is generated to increase the range of motion.

[0041] As a further aspect of the present invention, the method of generating personalized reference trajectory parameters by weighting and updating the standard reference trajectory feature matrix using topological features includes:

[0042] Calculate the topological similarity between the current phase space state trajectory and the feature matrix of the standard reference trajectory;

[0043] Generate and update weight factors based on topological similarity;

[0044] By using updated weighting factors, the spatial distribution characteristics of the current phase space state trajectory are fused into the standard reference trajectory feature matrix, the pipeline center trajectory line is smoothed and corrected, and saved as personalized reference trajectory parameters.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] 1. This invention uses a phase space reconstruction unit to calculate the optimal delay time and embedding dimension based on the mutual information method and the false neighbor method, constructs a state vector, and maps the limb motion time series data containing triaxial acceleration and angular velocity into a high-dimensional phase space state trajectory. This method utilizes phase space reconstruction technology to completely preserve the trajectory evolution characteristics of limb motion in phase space, ensuring the consistency and accuracy of the phase space state trajectory topology.

[0047] 2. This invention utilizes a multi-dimensional feature fusion evaluation unit, combining the trajectory boundary crossing index generated by the trajectory deviation analysis unit with the motion stability classification label generated by the nonlinear stability analysis unit, to perform multi-dimensional feature fusion calculation. Based on preset dynamic stability threshold ranges and collision tolerance thresholds, the system can accurately decouple and identify standard rehabilitation movements, stable compensatory movements, out-of-control pathological states, and rigid ineffective states through logical judgment. This effectively solves the problem that existing technologies struggle to distinguish between stable compensation and standard rehabilitation, or between pathological states and rigid ineffective states, thus improving the reliability of the evaluation.

[0048] 3. This invention introduces a benchmark parameter adaptive update unit. When the movement is determined to be in an effective rehabilitation state, the topological features of the current phase space state trajectory are used to weight and update the standard reference trajectory feature matrix. This mechanism can generate personalized reference trajectory parameters, so that the assessment benchmark can be dynamically adjusted with the patient's rehabilitation process, and achieve adaptive matching between rehabilitation standards and the patient's actual ability.

[0049] 4. This invention runs on the microprocessor of a wearable device, utilizes the matrix operation accelerator of the microprocessor to perform phase space mapping operations, and sends evaluation reports and real-time feedback instructions through a wireless transmission module; this architecture realizes efficient feature calculation based on the local microprocessor, and combined with real-time feedback instructions for correcting force exertion muscle groups, blocking alarms, stimulation and guidance, establishes a fast-response human-computer interaction closed loop, and improves the safety and effectiveness of rehabilitation training. Attached Figure Description

[0050] The following drawings, which illustrate embodiments of this application and are incorporated herein by reference as part of this application, are provided for understanding the application and explain its principles.

[0051] Figure 1 This is a structural diagram of a limb rehabilitation exercise data assessment system based on wearable devices according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] Please see Figure 1This invention relates to a limb rehabilitation exercise data evaluation system based on wearable devices. The system takes limb movement time-series data collected by the wearable device as input and outputs a rehabilitation exercise quality evaluation report and real-time feedback instructions. The system includes: a phase space reconstruction unit: acquiring limb movement time-series data collected by the wearable device, the time-series data including triaxial acceleration components and triaxial angular velocity components; calculating the optimal delay time and embedding dimension of the time-series data based on the mutual information method and the false neighbor method; constructing a state vector based on the optimal delay time and embedding dimension, mapping the one-dimensional time-domain time-series data to a high-dimensional phase space, and generating a phase space state trajectory;

[0054] Trajectory Deviation Analysis Unit: Loads a preset standard reference trajectory feature matrix, which includes the pipeline center trajectory line and pipeline radius threshold constructed based on healthy limb movement data; projects the phase space state trajectory into the topological space of the standard reference trajectory; detects the spatial distribution state of the phase space state trajectory relative to the inner wall of the standard reference trajectory, quantifies the degree to which the point cloud distribution exceeds the inner wall of the pipeline, and generates a trajectory boundary crossing index.

[0055] Nonlinear stability analysis unit: Based on the trajectory evolution characteristics of the phase space state trajectory in phase space, calculate the maximum Lyapunov exponent; load a preset dynamic stability threshold range, map the maximum Lyapunov exponent to the corresponding stability level, and generate motion stability classification labels;

[0056] Multidimensional feature fusion assessment unit: Combining trajectory boundary crossing index and motion stability classification label, it decouples and identifies the compensatory type and pathological state of limb movement through multidimensional feature fusion calculation; Based on the identification results, it generates a rehabilitation exercise quality assessment report and triggers real-time feedback instructions;

[0057] The baseline parameter adaptive update unit: In response to the rehabilitation exercise quality assessment report determining that the exercise is in an effective rehabilitation state, it extracts the topological features of the current phase space state trajectory; and uses the topological features to perform a weighted update of the standard reference trajectory feature matrix to generate personalized reference trajectory parameters.

[0058] This embodiment elaborates on the core architecture and operating mechanism of the above system. The phase space reconstruction unit, as the front-end sensing center of the system, aims to uncover the dynamic system characteristics hidden behind the time series. This unit acquires limb motion time series data through an inertial measurement unit (IMU) module integrated in the wrist or knee brace. This data set covers six degrees of freedom information of spatial motion. To adapt to subsequent univariate dynamics calculations and fully preserve the translational and rotational characteristics of limb motion, the system performs weighted modulus fusion on the six-axis data and calculates the Euclidean norm of the three-axis accelerations. The calculation formula is as follows:

[0059]

[0060] The Euclidean norm of the triaxial angular velocity is calculated using the following formula:

[0061]

[0062] in, They are respectively Acceleration components along the X, Y, and Z axes at any given time; They are respectively The angular velocity components along the X, Y, and Z axes at any given time are calculated; and a one-dimensional scalar sequence is constructed, the calculation formula of which is:

[0063]

[0064] The connection between the terms in the formula represents a weighted summation, where, These are preset dimensionless weighting coefficients; in order to eliminate acceleration With angular velocity The physical dimension differences are considered, and a dimensionless one-dimensional scalar sequence is constructed. The weighting coefficient is defined as the reciprocal of the upper limit of the corresponding sensor range: let the maximum range of the accelerometer be... ,For example The maximum range of the angular velocity meter is ,For example Then set Its value is Dimensions are ,set up Its value is Dimensions are ;

[0065] Through this dimensional cancellation process, the term is made and It is converted into a pure numerical value that reflects the proportion relative to the range, so as to ensure that the two types of physical quantities have equal numerical weights in the fused scalar sequence.

[0066] Here, the reciprocal of the maximum range is chosen as the weight instead of the signal variance or energy, based on considerations of the real-time performance and the stability of phase space reconstruction of the embedded system: if variance normalization is used, long-term historical data needs to be cached to calculate dynamic statistics, which not only introduces computational delay, but also causes the normalization factor to fluctuate with the intensity of motion, resulting in nonlinear distortion of the geometric shape of the phase space attractor and destroying the consistency of the topology.

[0067] Normalization based on the maximum range is a static linear transformation that maps physical quantities to the sensor capability utilization space [-1,1], ensuring the stability of the attractor shape. Meanwhile, regarding concerns about whether the signal is weakened when the motion features do not reach the range, since phase space reconstruction depends on the topological embedding of the delay vector, its dynamic features, such as the Lyapunov exponent, depend on the evolution structure of the trajectory rather than the absolute amplitude. As long as the signal strength is higher than the sensor noise floor, this linear scaling can completely preserve the differential geometric features of acceleration and angular velocity, and will not lose dynamic information due to the small amplitude. On the contrary, it avoids the excessive amplification of noise by variance normalization in the static or micro-motion state, thereby ensuring the accuracy of phase space reconstruction.

[0068] Furthermore, from the perspective of topological dynamics, the reciprocal of the maximum range is used as the weight. This forms a diagonal scaling matrix According to a corollary of the Tukens embedding theorem, as long as It is a non-singular linear transformation, and the reconstructed attractor manifold With the original dynamic manifold Preserve differential homeomorphism;

[0069] In contrast, variance-based normalization introduces a nonlinear fluctuation factor dependent on the sample window, which may violate the affine invariance of phase space distance metrics. Therefore, choosing the reciprocal of a fixed physical range is a necessary mathematical condition to ensure the comparability of topological features of rehabilitation assessments across different time sections;

[0070] The system uses the mutual information method to calculate the average mutual information function of time series data. Specifically, during the calculation, the time series data... The range of values ​​is divided into An equally probable interval It is a positive integer, and in this embodiment, it is taken as... The statistical data point falls into the first The marginal probabilities of each interval and data points Falling into The marginal probabilities of each interval and statistical data points Falling into interval and Falling into Joint probability of intervals Mutual information value is calculated using the Shannon entropy formula. The calculation formula is as follows:

[0071]

[0072] The system calculates different delays Below The value is selected, and the first local minimum point corresponding to the function curve as the delay increases is chosen. As the optimal delay time; the spurious neighbor method is used to solve the embedding dimension for a given embedding dimension. Construct state vectors in phase space Calculate each vector Its nearest neighbor Euclidean distance between And examine when the dimension increases to The rate of change of the distance between two points The calculation formula is as follows:

[0073]

[0074] in, The current embedding dimension; For discrete-time indexes on a time series; In order to be in In 3D phase space, with vector The index of the nearest neighbor in the time series using Euclidean distance, i.e. yes The nearest neighbor; and These represent increasing the embedding dimension to [number]. At that time, the original time index and The corresponding new high-dimensional state vector is used to detect whether the original proximity relationship has been destroyed due to the increase in dimension. Represents Euclidean norm operations;

[0075] like Greater than the preset discrimination threshold In this embodiment, the threshold value is set to 15. The selection criteria are based on the theoretical boundary of the spurious neighbor algorithm by Kennel et al., and the values ​​typically range from [value range missing]. In this embodiment, we select The noise floor measurement is based on an IMU sensor;

[0076] The specific measurement method is as follows: sensor data is collected under static conditions. Calculate the standard deviation of the signal ;set up The criterion is to ensure that the change in the Euclidean distance between the nearest neighbors is significantly higher than the noise level, i.e., satisfying:

[0077]

[0078] in, The preset noise tolerance coefficient is typically set to 2; under normalized scaling, it is taken as... It can effectively filter out false divergences caused by the inherent white noise of the sensor, while retaining the subtle nonlinear tremor characteristics in rehabilitation movements;

[0079] If a neighboring point can retain the subtle dynamic characteristics of rehabilitation movements while filtering out the interference of high-frequency vibration noise on the topology, then the neighboring point is determined to be a false neighboring point.

[0080] Different system iterative calculations The proportion of false neighbors is selected as the minimum dimension when the proportion converges to zero or below a preset threshold such as 5%. The convergence threshold of 5% is based on Abbarbanel's empirical criterion. Considering that actual physiological signals inevitably contain measurement noise, complete convergence to zero is usually not feasible. A tolerance of 5% is sufficient to ensure the complete unfolding of the topology and is used as the embedding dimension.

[0081] Based on these two key topological parameters, the system performs phase space reconstruction, mapping the one-dimensional discrete signal into a phase space state trajectory in a high-dimensional Euclidean space. The trajectory deviation analysis unit introduces a standard reference trajectory feature matrix, which is essentially a set of tubular trajectories with a certain tolerance radius formed in phase space when a healthy limb performs standard rehabilitation movements. By projecting the phase space state trajectory generated by the patient onto this topological space, the system can accurately detect the distribution of the point cloud relative to the inner wall of the pipe and quantify the trajectory boundary excursion index. This index intuitively reflects the degree of deviation of the current movement from the spatial structure.

[0082] Meanwhile, the nonlinear stability analysis unit works in parallel, calculating the maximum Lyapunov exponent based on the evolutionary separation rate of the point cloud trajectory over time; specifically, the system uses the Rosenstein data-volume method for solution: finding each reference point in phase space. nearest neighbor , here This represents the index of the nearest neighbor in the time series, to distinguish it from the reference point index. And satisfy the constraints. ;

[0083] And constrain the separation step size of the two in the time series to be greater than the preset Taylor window value. Typically, the time frame is taken from the relevant time to eliminate the effects of autocorrelation; the time step size of the evolution of this pair of neighboring points is tracked. Euclidean distance And calculate the average logarithmic divergence of all reference point pairs, the formula of which is:

[0084]

[0085] in, This is the index of the reference point in the phase space trajectory matrix. This represents the total number of valid reference points; here, the total number of valid reference points... Determined as the total number of points in phase space Subtract the maximum evolution step size ,Right now Maximum evolutionary step size The value of is crucial for capturing linear divergence regions;

[0086] In this embodiment, the motion frequency is extracted by performing a Fast Fourier Transform on the preprocessed time-series data. Calculate the average orbital period and will Set to correspond to to The number of sampling points per average orbital period, selected to The basis for this cycle is:

[0087] like Less than In each cycle, trajectory separation is mainly affected by local geometric curvature and cannot reflect the global dynamic stretching and folding characteristics; if Exceed Each cycle, due to the boundedness of chaotic systems, the separation distance When the scale reaches the attractor diameter and enters the saturation region, the slope approaches zero, causing the exponential calculation to fail. Simulation verification shows that... to Within a period range, the logarithmic distance curve To achieve optimal linearity, the linear correlation coefficient is [value missing]. ;

[0088] This ensures that the evolution time is long enough to exhibit exponential separation characteristics, while avoiding entering the saturation region of large-scale folding, so as to ensure that all reference points have complete evolutionary trajectories.

[0089] At the same time, distance Specifically defined as a reference point Its nearest neighbor After The state vector after one time step and The Euclidean distance between them is calculated using the following formula:

[0090]

[0091] in, Indicates reference point nearest neighbor go through The state vector after each time step evolution; Indicates reference point go through The state vector after each time step evolution;

[0092] Fitting using the least squares method To ensure fitting accuracy, the system automatically limits the time window for linear fitting to a linear region that varies over time. Within the range, i.e., the evolutionary step size From 0 to ,in The sampling frequency of the wearable device's sensor, for example, 100Hz, is used to convert time units into distances from walks, utilizing this range. The slope as the maximum Lyapunov exponent This eliminates nonlinear interference caused by trajectory folding or saturation;

[0093] The system maps the maximum Lyapunov exponent to a preset range of dynamic stability thresholds, thereby generating a motor stability classification label that reflects neural control ability.

[0094] Based on this, the multi-dimensional feature fusion evaluation unit serves as the decision core, combining the trajectory boundary crossing index and the motion stability classification label to perform multi-dimensional feature fusion calculation. This calculation process is not a simple numerical superposition, but rather constructs a feature fusion model based on hierarchical logic: the discrete motion stability classification labels, namely effective training, risk warning, and ineffective motion, are used as the primary decision key, and the continuous trajectory boundary crossing index is used as the secondary discriminant. When the primary decision key is effective training, the system further determines whether the trajectory boundary crossing index exceeds the preset tolerance threshold. If it does, the fusion output is stable compensation; otherwise, the output is standard rehabilitation.

[0095] When the primary decision key is a risk warning or invalid movement, the secondary discriminant is ignored, and the output is directly fused as out-of-control pathology or rigid invalidity. Through this multi-level logic solution, stable compensatory or latent pathological states that cannot be detected by simple trajectory comparison are identified. When the baseline parameter adaptive update unit determines that the movement is valid, it extracts the current topological features to perform weighted updates on the standard reference trajectory, thereby realizing the dynamic accompaniment of rehabilitation standards.

[0096] In a preferred embodiment of the present invention, a state vector is constructed based on an optimal delay time and an embedding dimension to map one-dimensional time-domain time-series data to a high-dimensional phase space. The method for generating a phase space state trajectory includes: defining the time-series data as a set of time series; performing time delay processing on the set of time series using the optimal delay time to generate multiple delayed sequences; extracting the values ​​of the set of time series and the delayed sequences at the same moment according to the embedding dimension to form a high-dimensional state vector; arranging the high-dimensional state vectors of all moments in chronological order to construct a phase space trajectory matrix, and using the phase space trajectory matrix as the phase space state trajectory.

[0097] This embodiment further refines the computation process of the phase space reconstruction unit, aiming to construct a high-dimensional structural system that can truly reflect the dynamic characteristics of limbs; the acquired time-series data is defined as a time series set. This includes the total number of sampling points. Discrete values; the optimal delay time obtained using the aforementioned solution. A time-shifting operation is performed on the set to generate a series of delayed sequences; this step aims to eliminate linear correlations between the data; based on the embedding dimension... The system extracts the original sequence and each delayed sequence at the same time. The numerical values ​​are combined to generate a high-dimensional state vector. The calculation formula is as follows:

[0098]

[0099] in, For the first phase space The coordinate vectors of each state point originate from reconstruction calculations, and their physical meaning is the coordinate vector of the system in [the following context]. Instantaneous state in 3D space; For the original time series data at time... The amplitude is obtained from sensor data. The optimal delay time is derived from the mutual information method calculation and is expressed in units of the number of sampling points. The embedding dimension is derived from the results of the spurious neighbor method and is a dimensionless integer; the system arranges the high-dimensional state vectors of all time steps in chronological order to construct the phase space trajectory matrix. The matrix is ​​then identified as the phase space state trajectory, and its matrix form is as follows:

[0100]

[0101] in, This is the matrix transpose symbol. The number of valid points in phase space is determined by the total number of points. Subtracting the reconstruction loss, we get the result. The specific calculation of the reconstruction loss is as follows: ,Right now This indicates that the original sequence cannot be fully constructed due to time-delayed embedding. The number of data points in a dimensional vector;

[0102] This embodiment utilizes Takens' embedding theorem to achieve lossless recovery from low-dimensional observation signals to high-dimensional dynamic states. When processing complex rehabilitation data such as Parkinson's tremor or post-stroke rigidity, this time-delay reconstruction method can unfold signals that originally appeared chaotic in three-dimensional physical space into a high-dimensional phase space, eliminating trajectory overlap caused by projection. This processing makes pathological micro-tremors or compensatory muscle synergy abnormalities present clearly discernible geometric differences in high-dimensional space, providing a solid mathematical foundation for subsequent topological manifold analysis.

[0103] In a preferred embodiment of the present invention, the spatial distribution state of the phase space state trajectory relative to the inner wall of the standard reference trajectory is detected, the degree to which the point cloud distribution exceeds the inner wall of the pipe is quantified, and the trajectory boundary excursion index is generated by: parsing the feature matrix of the standard reference trajectory, obtaining the pipe center trajectory line and the pipe radius threshold; and calculating the Euclidean distance from each point in the phase space state trajectory to the pipe center trajectory line.

[0104] The Euclidean distance is compared with the pipeline radius threshold to identify escape points where the Euclidean distance is greater than the pipeline radius threshold. The number of escape points is counted, and the proportion of the number of escape points to the total number of points in the phase space state trajectory is calculated. This proportion is determined as the trajectory boundary crossing index.

[0105] This embodiment details the specific logic of quantifying the degree of freedom overflow of limb movement in the trajectory deviation analysis unit; the system analyzes the standard reference trajectory feature matrix and extracts the pipeline center trajectory line representing the geometric core of the standard movement. and the pipe radius threshold that defines the tolerance range. ;

[0106] Here, the construction of the standard reference trajectory feature matrix adopts a population statistical method: standard action data of multiple groups of healthy subjects are collected, and a statistically representative mean trajectory is generated using a dynamic time warping centroid averaging algorithm as the pipeline center trajectory line. ; and calculate the standard deviation of the spatial position at each alignment node, setting 3 times the standard deviation as the threshold for the pipe radius of that node. This allows for a larger tolerance range in complex phase transitions such as joint switching;

[0107] Specifically, the data structure is defined as: the trajectory line of the pipeline center. Store as An orderly dimensional vector set , here Defined as the normalized length of the standard trajectory, it is obtained by dynamically time-normalizing and aligning the collected health sample data, and then resampling to a fixed number of points using cubic spline interpolation. The results were used to unify the time scale for different samples; pipe radius threshold. For dynamic threshold sequence ;

[0108] For each state point in the phase space state trajectory The system calculates the Euclidean distance between the point and the corresponding point on the pipeline's center trajectory line. In specific calculations, in order to avoid phase jump misjudgments that may be caused by overlapping regions of cyclic trajectories in phase space, the system adopts a nearest neighbor search strategy with time constraints.

[0109] Centerline index matched by the state point at the previous sampling time Based on the first state point of the sequence The system performs a global search to determine the initial optimal matching point. and assign it to To address the randomness of the patient's action initiation phase, a local dynamic search window is set, for example... , here The value was determined as the total number of points in the trajectory. Rounded down to 5%;

[0110] The parameter value is set based on the analysis of the variability rate of movement cycles in human exercise physiology, namely, the phase jitter of the movement execution of rehabilitation patients usually does not exceed the movement cycle. ;

[0111] This ratio is derived from research on the temporal variability of voluntary human movement; based on the extended application of Fitts's Law in periodic movements, it represents the standard deviation of the time period during which the human body completes repetitive rehabilitation movements without external metronome guidance. Typically around the average period of ;

[0112] This invention takes Confidence interval as a local dynamic search window It can cover The above-mentioned normal phase drift effectively prevents the loss of pathological hysteresis characteristics due to excessive forced alignment caused by time warping; that is, corresponding to the concept of time, the system converts it into the corresponding discrete point width, thereby setting the window. In order to handle the boundary problem at the loop connection of periodic rehabilitation exercises, the index calculation of this search window adopts a modular approach. Operation, i.e., the search range covers the set of indexes:

[0113]

[0114] in, This represents the modulo operation with respect to the total number of points L in the standard trajectory. This operation ensures that when searching the index... Exceeding When the range is defined, it can be correctly mapped back to the valid interval of the trajectory, realizing the loop search logic that connects the beginning and the end;

[0115] This ensures a smooth transition to the beginning of the next cycle at the end of the action cycle; searching only within the subset of the center trajectory line points covered by the window. The point with the smallest Euclidean distance As the corresponding point, obtain the radius threshold associated with that point. The calculation logic is as follows: The Euclidean distance calculation formula is:

[0116]

[0117] in, It represents the Euclidean distance of the state point from the center, which is derived from real-time calculation and has the physical meaning of the spatial deviation of the action. This is the vector of state points in the current phase space state trajectory; Distance on the center trajectory line of the standard pipeline The nearest discrete point vector that satisfies the time-series constraints;

[0118] The system executes the escape point identification logic and calculates the... The dynamic radius threshold corresponding to this point Perform comparison; respond to Greater than The system determines this point as an escape point, meaning that the motion state at this moment has broken through the manifold constraints of the standard action in the current phase; the system counts the number of escape points. And calculate its percentage of the total points. The proportion is used to generate a trajectory boundary crossing index. The calculation formula is as follows:

[0119]

[0120] in, The trajectory boundary violation index is derived from statistical calculations. Its physical meaning is the degree of spatial deformation and compensation of the action, and the unit is percentage.

[0121] In a preferred embodiment of the present invention, loading a preset dynamic stability threshold range and mapping the maximum Lyapunov exponent to the corresponding stability level to generate a motion stability classification label includes: loading a preset first entropy threshold and a second entropy threshold, wherein the first entropy threshold is greater than the second entropy threshold; the first entropy threshold and the second entropy threshold are dynamic parameters named based on the positive correlation between the maximum Lyapunov exponent and dynamic entropy in chaotic systems; wherein the first entropy threshold defines the upper boundary of the effective training range, i.e., the critical value that distinguishes between stability and chaos, and the second entropy threshold defines the lower boundary of the effective training range, i.e., the critical value that distinguishes between stability and stiffness, and the two constitute the dynamic stability range that is determined to be an effective rehabilitation state;

[0122] If the maximum Lyapunov exponent is greater than the first entropy threshold, the motion state is determined to be chaotic divergence, and a risk warning label is generated as the motion stability classification label; if the maximum Lyapunov exponent is less than or equal to the first entropy threshold and greater than or equal to the second entropy threshold, the motion state is determined to be stable periodic, and a valid training label is generated as the motion stability classification label.

[0123] If the maximum Lyapunov exponent is less than the second entropy threshold, the motion state is determined to be rigid convergence, and an invalid motion label is generated as the motion stability classification label.

[0124] This embodiment details the decision logic of the nonlinear stability analysis unit. This logic sets two key entropy thresholds for the system based on the statistical distribution of a large amount of clinical case data: the first entropy threshold... With the second entropy threshold And satisfy ;

[0125] The specific statistical determination process for the threshold is as follows: construct a system containing the number of samples. Clinical exercise databases of patients at different stages of rehabilitation, such as Calculate the maximum Lyapunov exponential distribution of the samples; perform cluster analysis on the distribution using a Gaussian mixture model, with the specific parameters set as follows: set the number of mixture components. This corresponds to three modes: stable, chaotic, and rigid. The covariance type is set to full covariance to accommodate the correlation of features with different dimensions. An expectation-maximization algorithm is used for iterative solution, with a convergence threshold set to... To prevent overfitting, the Bayesian information criterion is used to select the optimal model parameters; three cluster centers are identified: stable, chaotic, and rigid.

[0126] set up For chaotic clustering and stable clustering in Bayesian decision boundary values ​​under confidence intervals, set To distinguish between stable clustering and rigid clustering The Bayesian decision boundary value under the confidence interval, thereby ensuring that the state division is statistically significant;

[0127] The system calculates the maximum Lyapunov exponent. Execute hierarchical determination; respond to Greater than This indicates that adjacent trajectories separate at an extremely fast exponential rate in phase space, and the system is in a chaotic state that is extremely sensitive to initial conditions. At this time, the system determines the motion state as chaotic divergence and generates a risk warning label.

[0128] In response to lie in and The range indicates that the trajectory separation rate is within a reasonable range, the system has a stable periodic attractor structure, and at this point the system determines the motion state to be stable and generates effective training labels; in response to Less than This indicates that the trajectory has converged excessively and the system lacks the necessary physiological variability. At this point, the system determines the motion state as rigid convergence and generates invalid motion labels.

[0129] This embodiment utilizes the maximum Lyapunov exponent as the gold standard for measuring the stability of a dynamic system and constructs a dual threshold determination mechanism. This mechanism can not only identify chaotic state (random movement) but also identify rigid state (immobility), effectively solving the problem that traditional velocity or acceleration threshold-based methods cannot distinguish between pathological tremor and rigid convergence.

[0130] This entropy-based assessment method delves into the control quality of the neuromuscular system, enabling early warning of rehabilitation risks such as impending loss of control, falls, or severe spasms in patients.

[0131] In a preferred embodiment of the present invention, the method of decoupling and identifying the compensatory type and pathological state of limb movement by combining the trajectory boundary crossing index and the motion stability classification label and through multi-dimensional feature fusion calculation includes: loading a preset collision tolerance threshold.

[0132] If the motion stability classification label is a valid training label and the trajectory out-of-bounds index is greater than the collision tolerance threshold, then the limb movement is determined to be a stable compensatory movement; if the motion stability classification label is a valid training label and the trajectory out-of-bounds index is less than or equal to the collision tolerance threshold, then the limb movement is determined to be a standard rehabilitation movement.

[0133] If the movement stability classification label is a risk warning label, the trajectory boundary crossing index is ignored, and the limb movement is directly determined to be in an uncontrolled pathological state; if the movement stability classification label is an invalid movement label, the limb movement is determined to be in a stiff and invalid state.

[0134] This embodiment details the multi-dimensional feature fusion logic of the multi-dimensional feature fusion evaluation unit; the system loads a preset collision tolerance threshold. For example, it can be set to 15%; the system will classify the motion stability labels as LLE dimension and trajectory out-of-bounds index. That is, the topological dimension is constructed as a two-dimensional decision matrix, and the following decoupled recognition is performed: in response to the label being a valid training label and Greater than The system identifies this as a stable compensatory movement, which corresponds to the patient using incorrect muscle groups to exert force in pursuit of movement stability.

[0135] In response to the label being a valid training label and Less than or equal to The system classifies it as a standard rehabilitation exercise, indicating that the movement is both stable and accurate; in response to the risk warning label, regardless of... Regardless of the numerical value, the system directly determines it as an uncontrolled pathological state, because when neural control fails, the accidental overlap of trajectories has lost its clinical significance; in response to the label being an invalid movement label, the system determines it as a stiff and ineffective state, corresponding to joint adhesions or limited range of motion caused by fear and pain;

[0136] This embodiment achieves true decoupling of morphology and dynamics through multi-dimensional feature fusion. In complex clinical rehabilitation scenarios, traditional algorithms often misjudge stable compensation as standard motion because the trajectories seem similar, or misjudge pathological tremors as simple trajectory errors. This solution accurately distinguishes these four distinct clinical states, and in particular, it effectively detects the highly concealed stable compensation, providing reliable data support for developing accurate and personalized rehabilitation prescriptions.

[0137] In a preferred embodiment of the present invention, the system is integrated into a wearable device, which includes a microprocessor, a storage module, and a wireless transmission module. The storage module is used to store a standard reference trajectory feature matrix and various preset threshold parameters. The wireless transmission module is used to send the generated rehabilitation exercise quality assessment report and real-time feedback instructions to an external terminal. The phase space reconstruction unit uses the matrix operation accelerator of the microprocessor to perform phase space mapping operations.

[0138] This embodiment details the hardware operating environment and architecture of the system. The system is integrated into wearable devices such as smart rehabilitation bracelets. The wearable device includes a microprocessor, a storage module, and a wireless transmission module. The microprocessor is equipped with a matrix operation processing unit, such as an integrated or independent digital signal processor, a floating-point operation unit, or a general-purpose computing unit, which is specifically used to carry out high-dimensional matrix multiplication and eigenvalue decomposition operations in the phase space reconstruction unit.

[0139] In addition, the system is equipped with a storage module for local storage of standard reference trajectory feature matrices and various preset threshold parameters to ensure rapid access to the comparison benchmark;

[0140] Meanwhile, the system integrates a wireless transmission module, such as Bluetooth Low Energy, which is responsible for sending the generated rehabilitation exercise quality assessment report and real-time feedback instructions to external terminals, such as mobile applications or rehabilitation physician workstations.

[0141] This embodiment utilizes the hardware acceleration capabilities of the microprocessor to achieve real-time computing at the edge. In motion rehabilitation scenarios requiring millisecond-level feedback, this architecture avoids the risk of feedback lag caused by network latency by eliminating the need to upload raw big data to the cloud for processing. At the same time, it ensures the privacy and security of patient biometric data and achieves efficient assessment with low power consumption and low latency.

[0142] In a preferred embodiment of the present invention, the method of triggering real-time feedback instructions includes: generating a corrective prompt instruction for the exerting muscle group in response to determining that the limb movement is a stable compensatory movement; generating an interruption alarm instruction to immediately stop training in response to determining that the limb movement is an uncontrolled pathological state; generating an incentive instruction to maintain training intensity in response to determining that the limb movement is a standard rehabilitation movement; and generating a guidance instruction to increase the range of motion in response to determining that the limb movement is a stiff and ineffective state.

[0143] This embodiment describes in detail the closed-loop feedback mechanism based on the evaluation results; the system generates a targeted feedback strategy based on the judgment results of the aforementioned multidimensional feature decoupling: in response to the judgment result being stable compensatory movement, the system generates corrective prompts for the force-generating muscle groups, such as please relax your shoulders and use only your upper arms to exert force, aiming to break the patient's incorrect compensatory habits;

[0144] In response to the judgment result being an out-of-control pathological state, the system generates a blocking alarm command to immediately stop training, such as triggering the device to vibrate violently and display a red light to prevent falls or muscle strains;

[0145] In response to a judgment result indicating standard rehabilitation exercise, the system generates incentive instructions to maintain training intensity, such as "The movement is perfect, please maintain the current pace"; in response to a judgment result indicating stiffness and ineffectiveness, the system generates guiding instructions to increase the range of motion, such as "Please try to increase the swing range, don't tense up."

[0146] This embodiment transforms complex algorithm evaluation results into execution instructions that patients can directly understand, establishing a cognitive closed loop for human-computer interaction. In particular, the corrective prompts for stable compensation can help patients perceive and correct erroneous neuromuscular control patterns in real time during training, effectively preventing the solidification of incorrect movements and significantly improving the effectiveness and compliance of rehabilitation training.

[0147] In a preferred embodiment of the present invention, the method of generating personalized reference trajectory parameters by weighted updating of the standard reference trajectory feature matrix using topological features includes: calculating the topological similarity between the current phase space state trajectory and the standard reference trajectory feature matrix; generating an update weight factor based on the topological similarity; using the update weight factor, fusing the spatial distribution features of the current phase space state trajectory into the standard reference trajectory feature matrix, smoothing and correcting the pipeline center trajectory line, and saving it as personalized reference trajectory parameters.

[0148] This embodiment details the dynamic update logic of the baseline parameter adaptive update unit, aiming to address the issue of individual differences. When the system determines that the current exercise is effective rehabilitation, it executes a trajectory extraction step to obtain data that can be used for calculation. Considering that rehabilitation training typically involves repetitive periodic movements, the system's discrete phase space state trajectory... Perform periodic segmentation operations, utilizing Poincaré sections, for example, selecting a hyperplane in phase space with zero velocity and positive acceleration as the section; or using the zero-intersection velocity method to segment the point cloud into... Each independent segment of a motion cycle;

[0149] The dynamic time warping centroid averaging algorithm is used to analyze this. Each segment is time-aligned and centroids are calculated to generate a statistically representative current sub-average trajectory, which is then spline smoothed and resampled. Points are used to obtain the final current sub-average trajectory. ;

[0150] calculate The original center trajectory line in the standard reference trajectory feature matrix Topological similarity The specific calculation method employs a dynamic time warping algorithm to construct the cumulative distance matrix. Its recursive definition is as follows:

[0151]

[0152] in, The th in the cumulative distance matrix Line number The element value of the column, the first Line number The columns are respectively and Time step index of the sequence and They are respectively and The first in the sequence and the Feature point vectors, boundary conditions And for have ,for have The calculated normalized minimum cumulative distance The calculation formula is as follows:

[0153]

[0154] Mapping it to a normalized similarity value, the calculation formula is:

[0155]

[0156] in, For topological similarity, For example, the preset sensitivity coefficient This makes the similarity closer to 1 the smaller the distance;

[0157] based on Generate updated weight factors To prevent low-quality motion data from contaminating the model, a similarity-based nonlinear mapping function was established. The specific form is as follows:

[0158]

[0159] in, For a maximum update step size of 0.1, It is a natural constant. A similarity threshold of 0.8 is used. The steepness coefficient is set to 15 here. This parameter is based on the derivative properties of the sigmoid function, aiming to make the weighting factor... A steep nonlinear cutoff effect is generated nearby, ensuring that when similarity... When the value is below 0.8, the update weight decays rapidly. Below the order of magnitude, thus preventing low-quality data from contaminating the model;

[0160] and Setting the threshold to 0.8 is based on the statistical distribution characteristics of the clinical rehabilitation movement database, serving as an effectiveness boundary. Movements below this threshold typically contain significant compensatory patterns. This function ensures that only when the movement similarity is high... Only then can significant values ​​be achieved, thus realizing a renewal mechanism of survival of the fittest;

[0161] use The pipeline center trajectory line is smoothed and corrected by fusing the spatial distribution features of the current point cloud into the matrix to generate a new center trajectory line. It is worth noting that, in order to prevent trajectory ambiguity caused by phase mismatch, the system uses the optimal curved path generated in the aforementioned dynamic time warping calculation before performing weighted summation. Remap the data point indexes to make them consistent with... To achieve precise spatial phase alignment, the following calculation is performed:

[0162]

[0163] It should be noted that the above formula expresses a weighted linear interpolation operation for each corresponding point in the sequence; that is, for each index in the trajectory sequence ( ), new trajectory point vector It is derived from the vector of the old trajectory point. With the current trajectory point vector The result is obtained by weighted summation, thereby updating the shape of the entire trajectory;

[0164] in, The updated personalized reference trajectory parameters, in physical terms, represent a baseline trajectory adapted to the current stage of rehabilitation. The average trajectory of the current motion after dynamic time warping and phase calibration; The center trajectory line of the standard reference trajectory before the update; The updated weight factor is calculated based on dynamic time warping similarity. Its source is similarity calculation, and its physical meaning is the model learning rate. Its value range is... ;

[0165] In addition, to achieve adaptive adjustment of the manifold pipe envelope and ensure the integrity of parameter updates, the system synchronously updates the pipe radius threshold. ;

[0166] Calculate the position of each point in the current phase space state trajectory relative to the new center trajectory line. The normal dispersion is specifically calculated by utilizing the node correspondence established during the aforementioned dynamic time warping centroid average alignment process. A motion segment in Each node Spatial distribution standard deviation Construct the current second standard deviation sequence ;

[0167] Using the same update weight factor For the original radius threshold The weighted iteration is performed, and the calculation formula is as follows:

[0168]

[0169] in, The updated personalized pipe radius threshold; The standard pipe radius threshold before the update; This is the current standard deviation sequence;

[0170] This allows the pipe volume to be contracted or expanded while updating the trajectory shape, completing a closed-loop update of the personalized reference trajectory parameters; the system will then update the center trajectory line. With radius threshold Save the combined data as a benchmark for the next evaluation.

[0171] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A limb rehabilitation exercise data assessment system based on wearable devices, characterized in that, include: Phase space reconstruction unit: acquires limb motion time-series data collected by wearable devices, the time-series data including triaxial acceleration components and triaxial angular velocity components; The optimal delay time and embedding dimension of the time series data are calculated based on the mutual information method and the false neighbor method. Based on the optimal delay time and the embedding dimension, a state vector is constructed, and the time-series data in the one-dimensional time domain is mapped to a high-dimensional phase space to generate a phase space state trajectory. The trajectory deviation analysis unit loads a preset standard reference trajectory feature matrix, which includes a pipeline center trajectory line constructed based on healthy limb movement data and a pipeline radius threshold. The phase space state trajectory is projected into the topological space of the standard reference trajectory. The spatial distribution state of the phase space state trajectory relative to the inner wall of the standard reference trajectory is detected, and the degree to which the point cloud distribution exceeds the inner wall of the pipeline is quantified to generate a trajectory boundary violation index. Nonlinear stability analysis unit: Based on the trajectory evolution characteristics of the phase space state trajectory in phase space, calculate the maximum Lyapunov exponent; load a preset dynamic stability threshold range, map the maximum Lyapunov exponent to the corresponding stability level, and generate motion stability classification labels; Multidimensional feature fusion evaluation unit: Combining the trajectory boundary crossing index and the motion stability classification label, it decouples and identifies the compensation type and pathological state of limb movement through multidimensional feature fusion calculation; generates a rehabilitation exercise quality evaluation report based on the identification results and triggers real-time feedback instructions; Baseline parameter adaptive update unit: In response to the rehabilitation exercise quality evaluation report determining that the movement is in an effective rehabilitation state, it extracts the topological features of the current phase space state trajectory; The standard reference trajectory feature matrix is ​​updated using the aforementioned topological features to generate personalized reference trajectory parameters. The method of using the topological features to weight and update the standard reference trajectory feature matrix to generate personalized reference trajectory parameters includes: calculating the topological similarity between the current phase space state trajectory and the standard reference trajectory feature matrix; and generating update weight factors based on the topological similarity. Using the updated weighting factor, the spatial distribution characteristics of the current phase space state trajectory are fused into the standard reference trajectory feature matrix, the pipeline center trajectory line is smoothed and corrected, and saved as the personalized reference trajectory parameters. The method of loading a preset dynamic stability threshold range to map the maximum Lyapunov exponent to the corresponding stability level and generate motion stability classification labels includes: loading a preset first entropy threshold and a second entropy threshold, wherein the first entropy threshold is greater than the second entropy threshold; if the maximum Lyapunov exponent is greater than the first entropy threshold, the motion state is determined to be chaotic divergence, and a risk warning label is generated as the motion stability classification label; if the maximum Lyapunov exponent is less than or equal to the first entropy threshold and greater than or equal to the second entropy threshold, the motion state is determined to be stable periodic, and a valid training label is generated as the motion stability classification label; if the maximum Lyapunov exponent is less than the second entropy threshold, the motion state is determined to be rigid convergence, and an invalid motion label is generated as the motion stability classification label. The method of decoupling the identification of compensatory types and pathological states of limb movements by combining the trajectory boundary crossing index and the movement stability classification label through multi-dimensional feature fusion calculation includes: loading a preset collision tolerance threshold; if the movement stability classification label is a valid training label and the trajectory boundary crossing index is greater than the collision tolerance threshold, then the limb movement is determined to be a stable compensatory movement; if the movement stability classification label is a valid training label and the trajectory boundary crossing index is less than or equal to the collision tolerance threshold, then the limb movement is determined to be a standard rehabilitation movement; if the movement stability classification label is a risk warning label, then the trajectory boundary crossing index is ignored, and the limb movement is directly determined to be an uncontrolled pathological state; if the movement stability classification label is an invalid movement label, then the limb movement is determined to be a stiff and ineffective state; a rehabilitation movement quality assessment report is generated based on the identification results, and a real-time feedback command is triggered.

2. The limb rehabilitation exercise data assessment system based on wearable devices according to claim 1, characterized in that, The method of constructing a state vector based on the optimal delay time and the embedding dimension to map the one-dimensional time-domain time-series data to a high-dimensional phase space and generate a phase space state trajectory includes: defining the time-series data as a time series set; performing time delay processing on the time series set using the optimal delay time to generate multiple delayed sequences; extracting the values ​​of the time series set and the delayed sequences at the same moment according to the embedding dimension to form a high-dimensional state vector; arranging the high-dimensional state vectors at all moments in chronological order to construct a phase space trajectory matrix, and using the phase space trajectory matrix as the phase space state trajectory.

3. The limb rehabilitation exercise data assessment system based on wearable devices according to claim 1, characterized in that, The method for detecting the spatial distribution of the phase space state trajectory relative to the inner wall of the standard reference trajectory, quantifying the degree to which the point cloud distribution exceeds the inner wall of the pipeline, and generating a trajectory boundary crossing index includes: parsing the feature matrix of the standard reference trajectory to obtain the pipeline center trajectory line and the pipeline radius threshold; calculating the Euclidean distance from each point in the phase space state trajectory to the pipeline center trajectory line; comparing the Euclidean distance with the pipeline radius threshold to identify escape points whose Euclidean distance is greater than the pipeline radius threshold; counting the number of escape points and calculating the proportion of the number of escape points to the total number of points in the phase space state trajectory, and determining the proportion as the trajectory boundary crossing index.

4. The limb rehabilitation exercise data assessment system based on wearable devices according to claim 1, characterized in that, The system is integrated into a wearable device, which includes a microprocessor, a storage module, and a wireless transmission module. The storage module is used to store the standard reference trajectory feature matrix and various preset threshold parameters. The wireless transmission module is used to send the generated rehabilitation exercise quality assessment report and the real-time feedback command to an external terminal. The phase space reconstruction unit uses the matrix operation accelerator of the microprocessor to perform phase space mapping operations.

5. The limb rehabilitation exercise data assessment system based on wearable devices according to claim 1, characterized in that, The methods for triggering real-time feedback instructions include: generating corrective prompts for the exerting muscle groups in response to determining that the limb movement is a stable compensatory movement; generating an interruption alarm instruction to immediately stop training in response to determining that the limb movement is an uncontrolled pathological state; generating an incentive instruction to maintain training intensity in response to determining that the limb movement is a standard rehabilitation movement; and generating a guiding instruction to increase the range of motion in response to determining that the limb movement is a stiff and ineffective state.

Images