Rehabilitation movement recognition and quantitative assessment method, device, equipment and storage medium

Rehabilitation action recognition is performed through the cascade model of preset random forest network and gated loop unit, and combined with the convolutional neural network to calculate the weight of motion characteristics, the low degree of intelligence of rehabilitation action recognition and quantitative evaluation in the existing technology is solved. The problem of insufficient accuracy has achieved refined quantitative scoring and personalized treatment support for rehabilitation actions.

CN118228145BActive Publication Date: 2024-11-01XIAN LIBANG CONTMEDU MEDICAL TECHNOLOGY CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410354962.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-27
Publication Date
2024-11-01
Estimated Expiration
2044-03-27

AI Technical Summary

Technical Problem

The existing rehabilitation action recognition and quantitative assessment methods have low intelligence and insufficient accuracy, and cannot achieve assessments that are completely free from the participation of doctors. The scoring results cannot accurately reflect the patient's actual recovery level and progress.

Method used

The cascading model of preset random forest network, gated cyclic unit and fully connected network is used to identify rehabilitation actions. The quantitative evaluation results of motion signals are calculated through logical feature extraction, motion feature extraction and convolutional neural network, and the gradient weighted class activation is used. Map calculates the weight of signal characteristics and generates a refined quantitative scoring formula.

Benefits of technology

It realizes intelligent identification and meticulous quantitative scoring of rehabilitation actions, improves the accuracy of assessment, saves medical resources, and provides more comprehensive information support for the personalized formulation of rehabilitation treatment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118228145B_ABST
    Figure CN118228145B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, equipment and storage medium for identifying and quantitatively evaluating rehabilitation movements. The method includes: obtaining the target motion signal of the patient during the rehabilitation process, inputting the target motion signal into a preset rehabilitation movement recognition model for processing, generating a target recognition result corresponding to the target motion signal, obtaining the motion feature value and weight corresponding to the target motion signal, and calculating the quantitative evaluation result of the target motion signal based on the motion feature value and weight. This scheme successfully realizes the intelligent recognition of rehabilitation movements by making full use of the multi-dimensional target motion signal collected by the inertial sensor and using the cascade classification model of the preset random forest and the preset gated recurrent unit; at the same time, the weight of the signal feature is calculated by using the gradient weighted class activation mapping, and a refined quantitative scoring formula for rehabilitation movements is proposed through the feature value and weight, which realizes the detailed quantitative scoring of rehabilitation movements and improves the accuracy of quantitative evaluation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of network security technology, and in particular to a method, device, equipment and storage medium for identifying and quantitatively evaluating rehabilitation movements. Background Art

[0002] The typical feature of stroke is neurological deficits caused by acute focal damage to the central nervous system due to vascular causes, including cerebral infarction, intracerebral hemorrhage and subarachnoid hemorrhage, which is the leading cause of disability and death worldwide. Stroke patients have limb motor dysfunction, which seriously affects their ability to take care of themselves and their quality of life. At present, a well-organized multidisciplinary approach can be used for post-acute rehabilitation treatment to achieve better results.

[0003] Traditionally, visual sensors have been widely used in motion analysis and rehabilitation goals for the recognition and quantitative assessment of rehabilitation movements. Through limb motion tracking algorithms, body movements can be captured comprehensively, non-invasively and in real time, and three-dimensional skeletal models can be generated. Kinematic parameters for specific motion tasks can be derived from these models to achieve upper limb motion assessment and lower limb gait analysis. This enables patients to assess the rehabilitation effect during home rehabilitation, and provides feedback on the training effect to patients, thereby correcting their training misunderstandings.

[0004] However, the intelligence level of rehabilitation movement recognition using the above method is low, and the accuracy of quantitative evaluation is low. Summary of the invention

[0005] The present application aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of the present application proposes a method for identifying and quantitatively evaluating rehabilitation movements, the method comprising:

[0006] Obtain the patient's target movement signal during the rehabilitation process;

[0007] The target motion signal is input into a preset rehabilitation action recognition model for processing to generate a target recognition result corresponding to the target motion signal; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network;

[0008] Based on the target recognition result, the motion feature value and weight corresponding to the target motion signal are obtained, and the quantitative evaluation result of the target motion signal is calculated based on the motion feature value and the weight.

[0009] In a possible implementation, the target motion signal is input into a preset rehabilitation motion recognition model for processing to generate a target recognition result corresponding to the target motion signal, including:

[0010] The target motion signal is input into a preset rehabilitation motion recognition model to extract logical features and generate multiple logical feature values; wherein each logical feature value is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude;

[0011] Based on the target motion signal and a plurality of logical feature values, a target recognition result corresponding to the target motion signal is generated.

[0012] In a possible implementation, generating a target recognition result corresponding to the target motion signal based on the target motion signal and a plurality of logical feature values ​​includes:

[0013] Inputting multiple logical feature values ​​into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal;

[0014] The target motion signal and the initial recognition result are input into a preset gated recurrent unit for processing, and the processing result is input into a preset fully connected network to generate a target recognition result corresponding to the target motion signal.

[0015] In a possible implementation, obtaining a motion feature value and a weight corresponding to the target motion signal based on the target recognition result includes:

[0016] Based on the target recognition result, the motion feature of the target motion signal is extracted to generate multiple motion feature data and motion feature values ​​corresponding to each motion feature data; wherein the multiple motion feature data include four feature data, the four feature data include amplitude, dynamic time warp, smoothness and approximate entropy, the amplitude is used to characterize the action amplitude corresponding to the target motion signal, the dynamic time warp is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the action fluency corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal;

[0017] Based on the motion feature values ​​corresponding to each motion feature data, a target feature matrix of a preset size is generated;

[0018] The target feature matrix is ​​input into the preset convolutional neural network model for processing to generate weights corresponding to each motion feature value.

[0019] In a possible implementation, the target feature matrix is ​​input into a preset convolutional neural network model for processing to generate weights corresponding to each motion feature value, including:

[0020] The target feature matrix is ​​input into the preset convolutional neural network model for processing, and the weights corresponding to each motion feature value are calculated through gradient weighted class activation mapping.

[0021] In a possible implementation manner, the quantitative evaluation result of the target motion signal is calculated based on the motion feature value and the weight, including:

[0022] Obtaining first category information corresponding to each motion feature value and second category information corresponding to each weight; wherein the first category information includes positive coefficient feature values ​​and negative coefficient feature values, and the second category information includes positive coefficient weights and negative coefficient weights; the positive coefficient feature values ​​are used to characterize that the motion feature value is positively correlated with the preset action completion degree, and the negative coefficient feature values ​​are used to characterize that the motion feature value is negatively correlated with the preset action completion degree; the positive coefficient weights correspond to the positive coefficient feature values, and the negative coefficient weights correspond to the negative coefficient feature values;

[0023] Based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight and the negative coefficient weight, the quantitative evaluation result of the target motion signal is calculated.

[0024] In a possible implementation, the method further includes:

[0025] Obtain the original movement signals of patients during rehabilitation;

[0026] The original motion signal is subjected to data preprocessing to generate a target motion signal; wherein the data preprocessing operation at least includes signal denoising, signal segmentation and signal length conversion.

[0027] The second aspect of the present application provides a rehabilitation movement recognition and quantitative assessment device, the device comprising:

[0028] An acquisition module, used to acquire the target motion signal of the patient during the rehabilitation process;

[0029] A generation module, used for inputting the target motion signal into a preset rehabilitation action recognition model for processing, and generating a target recognition result corresponding to the target motion signal; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network;

[0030] The calculation module is used to obtain the motion feature value and weight corresponding to the target motion signal based on the target recognition result, and calculate the quantitative evaluation result of the target motion signal based on the motion feature value and weight.

[0031] In a possible implementation manner, the generation module is specifically used to:

[0032] The target motion signal is input into a preset rehabilitation motion recognition model to extract logical features and generate multiple logical feature values; wherein each logical feature value is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude;

[0033] Based on the target motion signal and a plurality of logical feature values, a target recognition result corresponding to the target motion signal is generated.

[0034] In a possible implementation manner, the generation module is further used to:

[0035] Inputting multiple logical feature values ​​into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal;

[0036] The target motion signal and the initial recognition result are input into a preset gated recurrent unit for processing, and the processing result is input into a preset fully connected network to generate a target recognition result corresponding to the target motion signal.

[0037] In a possible implementation manner, the above-mentioned calculation module is specifically used for:

[0038] Based on the target recognition result, the motion feature of the target motion signal is extracted to generate multiple motion feature data and motion feature values ​​corresponding to each motion feature data; wherein the multiple motion feature data include four feature data, the four feature data include amplitude, dynamic time warp, smoothness and approximate entropy, the amplitude is used to characterize the action amplitude corresponding to the target motion signal, the dynamic time warp is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the action fluency corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal;

[0039] Based on the motion feature values ​​corresponding to each motion feature data, a target feature matrix of a preset size is generated;

[0040] The target feature matrix is ​​input into the preset convolutional neural network model for processing to generate weights corresponding to each motion feature value.

[0041] In a possible implementation manner, the above-mentioned calculation module is further used for:

[0042] The target feature matrix is ​​input into the preset convolutional neural network model for processing, and the weights corresponding to each motion feature value are calculated through gradient weighted class activation mapping.

[0043] In a possible implementation manner, the above-mentioned calculation module is further used for:

[0044] Obtaining first category information corresponding to each motion feature value and second category information corresponding to each weight; wherein the first category information includes positive coefficient feature values ​​and negative coefficient feature values, and the second category information includes positive coefficient weights and negative coefficient weights; the positive coefficient feature values ​​are used to characterize that the motion feature value is positively correlated with the preset action completion degree, and the negative coefficient feature values ​​are used to characterize that the motion feature value is negatively correlated with the preset action completion degree; the positive coefficient weights correspond to the positive coefficient feature values, and the negative coefficient weights correspond to the negative coefficient feature values;

[0045] Based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight and the negative coefficient weight, the quantitative evaluation result of the target motion signal is calculated.

[0046] In a possible implementation manner, the above-mentioned rehabilitation movement recognition and quantitative assessment device is also used for:

[0047] Obtain the original movement signals of patients during rehabilitation;

[0048] The original motion signal is subjected to data preprocessing to generate a target motion signal; wherein the data preprocessing operation at least includes signal denoising, signal segmentation and signal length conversion.

[0049] The third aspect of the present application proposes an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the rehabilitation movement recognition and quantitative assessment method as described in the first aspect.

[0050] The fourth aspect of the present application proposes a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the rehabilitation movement recognition and quantitative assessment method as described in the first aspect.

[0051] The embodiments of the present application have the following beneficial effects:

[0052] The embodiment of the present application provides a method for identifying and quantitatively evaluating rehabilitation movements, the method comprising: obtaining a target motion signal of a patient during rehabilitation, inputting the target motion signal into a preset rehabilitation motion recognition model for processing, generating a target recognition result corresponding to the target motion signal, obtaining a motion feature value and a weight corresponding to the target motion signal based on the target recognition result, and calculating a quantitative evaluation result of the target motion signal based on the motion feature value and the weight. This scheme successfully realizes the intelligent recognition of rehabilitation movements by making full use of the multi-dimensional target motion signal collected by the inertial sensor and using a cascade classification model of a preset random forest and a preset gated recurrent unit; in addition, by real-time recognition of the patient's current rehabilitation movements, the need for doctors in the process of rehabilitation training or rehabilitation evaluation is solved, thereby effectively saving medical resources, and effective rehabilitation training and rehabilitation evaluation can be performed in the absence of a doctor, such as at home, remotely, in a rehabilitation hall, etc.; at the same time, the weight of the signal feature is calculated using the gradient weighted class activation mapping, and a refined quantitative scoring formula for rehabilitation movements is proposed through the feature value and weight, thereby realizing a detailed quantitative scoring of rehabilitation movements. This algorithm not only improves the accuracy of quantitative evaluation, but also provides more comprehensive information support for the personalized formulation of rehabilitation treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A block diagram of a computer device provided in an embodiment of the present application;

[0054] Figure 2 A flowchart of the steps of the rehabilitation movement recognition and quantitative assessment method provided in the embodiment of the present application;

[0055] Figure 3 A flowchart of the steps for generating a target motion signal provided in an embodiment of the present application;

[0056] Figure 4 A schematic diagram of the structure of a cascade model provided in an embodiment of the present application;

[0057] Figure 5 A schematic diagram of the structure of a GRU unit provided in an embodiment of the present application;

[0058] Figure 6 A flowchart of the steps for generating a target recognition result provided in an embodiment of the present application;

[0059] Figure 7 Another step flow chart for generating target recognition results provided in an embodiment of the present application;

[0060] Figure 8 A flowchart of the steps for obtaining motion feature values ​​and weights provided in an embodiment of the present application;

[0061] Fig. 9A schematic diagram of a network structure of a preset convolutional neural network model provided in an embodiment of the present application;

[0062] Fig.10 A flowchart of the steps for calculating the quantitative evaluation results provided in an embodiment of the present application;

[0063] Fig.11 An overall framework diagram of a rehabilitation movement recognition and quantitative assessment method provided in an embodiment of the present application;

[0064] Fig.12 A structural block diagram of a rehabilitation movement recognition and quantitative assessment device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0066] When performing rehabilitation movement recognition and quantitative assessment, visual sensors have been widely used in motion analysis and rehabilitation goals. Through limb motion tracking algorithms, body movements can be captured comprehensively, non-invasively and in real time, and a three-dimensional skeletal model can be generated. Kinematic parameters for specific motion tasks can be derived from the model to achieve upper limb motion assessment and lower limb gait analysis. This enables patients to assess the rehabilitation effect during home rehabilitation, and provides feedback on the training effect to patients, thereby correcting their training misunderstandings.

[0067] However, there are still some obvious problems when using the above method for rehabilitation action recognition:

[0068] 1. Lack of motion recognition system: The current intelligent assessment system can only achieve semi-automation and lacks a complete motion recognition system, which makes it impossible to achieve an assessment that is completely independent of the participation of doctors, so the degree of intelligence is low.

[0069] 2. Insufficient assessment accuracy: The existing assessment methods need to be improved in terms of scoring accuracy, which may affect the effectiveness and targetedness of the treatment plan.

[0070] 3. Rough scoring of movement standards: The current scoring system is not detailed enough in assessing the standard of movements, resulting in a ceiling effect in the scoring process, that is, the scoring results cannot accurately reflect the patient's actual rehabilitation level and progress.

[0071] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values ​​may be based on additional conditions or values ​​beyond the described values ​​in practice.

[0072] The rehabilitation movement recognition and quantitative assessment method provided in the present application can be applied to a computer device (electronic device), and the computer device can be a server or a terminal, wherein the server can be a single server or a server cluster composed of multiple servers. The embodiments of the present application do not make specific limitations on this. The terminal can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices.

[0073] Take the computer device as a server as an example. Figure 1 A block diagram of a server is shown, such as Figure 1 As shown, the server may include a processor and a memory connected via a system bus. The processor of the server is used to provide computing and control capabilities. The memory of the server includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. When the computer program is executed by the processor, a method for recognizing and quantitatively evaluating rehabilitation movements is implemented.

[0074] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the server to which the solution of the present application is applied. Optionally, the server may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0075] It should be noted that the execution subject of the embodiments of the present application may be a computer device or a rehabilitation movement recognition and quantitative assessment device. The following method embodiments are described with a computer device as the execution subject.

[0076] Figure 2 A flowchart of a method for identifying and quantitatively evaluating rehabilitation movements provided in an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0077] Step 202: Acquire the target motion signal of the patient during the rehabilitation process.

[0078] Among them, before collecting the target motion signal, four inertial sensors can be properly installed on key parts of the patient's body, including the affected arm and upper arm of the upper limb, and the affected thigh and calf of the lower limb, so that all target motion signals of all limb parts and joints including the shoulder, elbow, wrist, hand, hip, knee, ankle, etc. can be collected. In addition, sensor calibration can be performed to check accuracy and correct any possible offset to ensure that the generated data is reliable and accurate.

[0079] Next, the data acquisition program can be started to record the motion data of various movements of the patient during rehabilitation training, which is about four cycles. Among them, one motion cycle refers to the patient completing a rehabilitation assessment movement or rehabilitation training movement. Clinical evaluation usually requires the patient to complete the movement multiple times, so there are multiple cycles. The data acquisition system requires the movement to be completed four times, so there are four cycles.

[0080] Each inertial sensor can collect data at a sampling rate of 30Hz, which can include Euler angles, acceleration, and angular velocity. The data size collected by the inertial sensor is the number of sensors * the number of signal types * the signal length L, for example, it can be 4*9*L. Among them, the Euler angle includes pitch angle, yaw angle, and roll angle; acceleration includes calibrated data in the three directions of X, Y, and Z; angular velocity includes calibrated data in the three directions of X, Y, and Z.

[0081] In addition, the collected target motion signals can also be stored in a computer hard disk or cloud database to ensure data security and accessibility. At the same time, the data may also need to be transmitted in real time to an intelligent rehabilitation identification and evaluation unit or other processing equipment.

[0082] In some optional embodiments, when generating a target motion signal, such as Figure 3 As shown, Figure 3 A flowchart of steps for generating a target motion signal provided in an embodiment of the present application includes:

[0083] Step 302: Acquire the original motion signal of the patient during the rehabilitation process.

[0084] Step 304: Perform data preprocessing on the original motion signal to generate a target motion signal.

[0085] The data preprocessing operation includes at least signal denoising, signal segmentation and signal length transformation, and may also include missing value processing and other operations, which are not specifically limited in the present embodiment.

[0086] Optionally, the original motion signal may contain some random noise, so a moving average filter may be used to denoise the original motion signal. The calculation formula of the moving average filter output is shown in formula (1).

[0087]

[0088] Among them, y k Represents the output result, x k represents the original motion signal, N represents the length of the moving average filter, and i is a loop parameter in the summation formula and has no specific physical meaning.

[0089] When performing signal segmentation, since the original motion signal is approximately four cycles, the first-order difference of the original signal can be calculated first, so as to detect the change points of the differential signs corresponding to the peaks and troughs: the point from positive to negative is the peak, and the point from negative to positive is the trough. Then, the period is segmented by the detected peaks and troughs, and the second to third period signals are used as the output signal. This section of the signal is the most stable and is used for subsequent analysis and calculation.

[0090] When performing signal length transformation processing, the signal length can be unified, for example, the data length after signal segmentation can be unified to 200 sampling points. Optionally, the length of the segmented signal can be unified by using a piecewise linear interpolation method, and the calculation formula is shown in formula (2).

[0091]

[0092] Among them, (x1, y1) and (x2, y2) represent two adjacent points in the signal, f(x) is a piecewise linear interpolation function, and the value interval of the independent variable x is [x1, x2].

[0093] The size of the target motion signal after final data preprocessing is 4*9*200. The data acquisition process of each inertial sensor is independent of each other. The size of each target motion signal can be expressed as 36*200, which is the number of signals * signal length.

[0094] Step 204: input the target motion signal into a preset rehabilitation action recognition model for processing, and generate a target recognition result corresponding to the target motion signal.

[0095] The preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network, such as Figure 4 As shown, Figure 4A schematic diagram of the structure of a cascade model provided in an embodiment of the present application, wherein the random forest module is a preset random forest network, the gated unit module is a preset gated recurrent unit, the fully connected layer is a preset fully connected network, and the logical feature vector is composed of multiple logical feature values.

[0096] Multiple preset gated recurrent units (GRU) and preset fully connected networks can form a gated recurrent unit. The GRU unit can be composed of a reset gate and an update gate, and its structure is as follows: Figure 5 As shown, Figure 5 A schematic diagram of the structure of a GRU unit provided in an embodiment of the present application.

[0097] The reset gate is used to control the last timestamp state h entering the GRU unit. t-1 The number of gate vectors g r Input x from the current timestamp t and the previous timestamp state h t-1 It is obtained by transforming the following formulas (3) and (4). The update gate is used to control the previous timestamp state h t-1 and the new input h t ~ For the new state vector h t The degree of influence can be obtained according to the following formulas (5) and (6).

[0098] g r =σ(W r [h t-1 ,x t ]+b r ) (3)

[0099] h t ~=tanh(W h [g r h t-1 ,x t ]+b h ) (4)

[0100] g z =σ(W z [h t-1 ,x t ]+b z ) (5)

[0101] h t =(1-g z )h t-1 +g z h t ~ (6)

[0102] Among them, Wr ,W z ,W h ,b r ,b h ,b z The parameters of the reset gate and update gate are automatically optimized by the back propagation algorithm; g z and g r are the reset gate control vector and the update gate control vector respectively; σ is the activation function, and the Sigmoid function is used.

[0103] When the preset random forest network, preset gated recurrent unit and preset fully connected network are trained, the maximum depth in the structural parameters of the preset random forest network can be 10, and the input length can be 8. The structural parameters of the preset gated recurrent unit can adopt a bidirectional strategy, the number of network layers can be 2, the number of units in each layer can be 200, the input length can be 36, and the hidden length can be 128. The input length in the structural parameters of the preset fully connected network can be 128, and the output length can be 46.

[0104] In addition, the loss function in the training parameters of the above three networks can adopt the cross-entropy loss function, that is, Cross-entropy, the optimizer can be the Adam optimizer, the learning rate can be 0.001, and the Epochs can be 50. Epochs refers to how many times the data will be "rounded" during the training process. The complete data set is sent to the neural network training, and the return is called an epoch. The batch size can be 32.

[0105] In some optional embodiments, Figure 6 As shown, Figure 6 A flowchart of steps for generating a target recognition result provided in an embodiment of the present application includes:

[0106] Step 602: Input the target motion signal into a preset rehabilitation motion recognition model to extract logical features and generate a plurality of logical feature values.

[0107] Step 604: Generate a target recognition result corresponding to the target motion signal based on the target motion signal and the multiple logical feature values.

[0108] Among them, each logical characteristic value is used to characterize whether there is a valid signal on the different limb structures of the patient, as well as the signal fluctuation amplitude information. When performing logical feature extraction, the fluctuation amplitude can be first judged for the Euler angle signal in the target motion signal. Fluctuations less than 20 degrees are regarded as baseline signals and can be classified as invalid signals. In order to distinguish the strength of the angle signal, among the valid signals with a fluctuation amplitude greater than or equal to 20 degrees, the amplitude between 20-80 degrees is defined as a small signal, and the amplitude greater than 80 degrees is defined as a large signal.

[0109] Based on the above rules, eight logical features can be extracted from each target motion signal, representing large upper arm signal, small upper arm signal, large forearm signal, small forearm signal, large thigh signal, small thigh signal, large calf signal, and small calf signal. For example, the characteristic values ​​of these eight features are 1, 0, 1, 0, 0, 0, 0, 0, indicating that the action is a large overall upper limb action. Therefore, in the subsequent cascade model, only the inertial sensor of the upper limb can be considered to reduce the interference of other inertial sensors. It should be noted that these eight logical features represent different types of actions, not signals. The values ​​of these features are logical values ​​0 or 1, where 0 indicates no valid signal or baseline signal, and 1 indicates a valid signal. Therefore, each target motion signal will be converted into a feature vector of length 8, where the vector element value is 0 or 1.

[0110] Then, based on the target motion signal and the multiple logical feature values, a target recognition result corresponding to the target motion signal can be generated. Figure 7 As shown, Figure 7 Another step flow chart for generating a target recognition result provided in an embodiment of the present application includes:

[0111] Step 702: Input multiple logical feature values ​​into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal.

[0112] Step 704: input the target motion signal and the initial recognition result into a preset gated recurrent unit for processing, and input the processing result into a preset fully connected network to generate a target recognition result corresponding to the target motion signal.

[0113] Among them, multiple logical feature values ​​can be input into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal. Then the target motion signal and the initial recognition result are input into a preset gated recurrent unit for processing, where x1 to x n In addition, the preset gated recurrent unit can adopt a bidirectional strategy, and the processing results of the forward and reverse outputs are spliced ​​as the input of the preset fully connected network.

[0114] The processing result is input into a preset fully connected network to generate a target recognition result corresponding to the target motion signal. The processing result can be processed in the preset fully connected network and then converted into a probability distribution to output the target recognition result.

[0115] Step 206: acquiring motion feature values ​​and weights corresponding to the target motion signal based on the target recognition result, and calculating a quantitative evaluation result of the target motion signal based on the motion feature values ​​and weights.

[0116] Among them, Figure 8 As shown, Figure 8 A flowchart of steps for obtaining motion feature values ​​and weights provided in an embodiment of the present application includes:

[0117] Step 802: extract motion features of the target motion signal based on the target recognition result, and generate a plurality of motion feature data and motion feature values ​​corresponding to each motion feature data.

[0118] Step 804: Generate a target feature matrix of a preset size based on the motion feature values ​​corresponding to each motion feature data.

[0119] Step 806: Input the target feature matrix into a preset convolutional neural network model for processing to generate weights corresponding to each motion feature value.

[0120] Among them, the multiple motion feature data include four feature data, including amplitude (AMP), dynamic time warping (DTW), smoothness (SMO) and approximate entropy (APEN). The amplitude is used to characterize the motion amplitude corresponding to the target motion signal, the dynamic time warping is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the motion smoothness corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal.

[0121] Motion feature extraction can be performed on each of the above 36 target motion signals to generate multiple motion feature data and motion feature values ​​corresponding to each motion feature data. Based on the motion feature values ​​corresponding to each motion feature data, a target feature matrix of a preset size is generated. The size of the target feature matrix can be 36×4. In addition, in order to meet specific processing or model requirements, the target feature matrix can be reshaped into a 13×13 form. This transformation can be filled with zero padding to ensure the consistency of the matrix dimension. This method of feature extraction and matrix adjustment is intended to provide a richer data representation for subsequent models, especially for algorithms suitable for processing two-dimensional data structures, such as convolutional neural networks.

[0122] The amplitude can be calculated by formula (7), the dynamic time warp can be calculated by formula (8), the smoothness can be calculated by formula (9), and the approximate entropy can be calculated by formulas (10), (11), (12), (13), and (14).

[0123] AMP=max(x)-min(x) (7)

[0124]

[0125]

[0126] APEN=Φm(r)-Φm+1(r) (10)

[0127]

[0128]

[0129] d[x(i),x(j)]=maxk[|x(i+k)-x(j+k)|] (13)

[0130] x(i)={xi,xi+1,...,xi+m-1},i=1,2...M-m+1 (14)

[0131] Among them, x is the target motion signal; i is the cycle parameter, which has no specific physical meaning; x i(n) and i(n) is the point on signal A and signal B that matches the path; d is the Euclidean distance between the two points; m is the pre-defined embedding dimension or segmentation parameter; r is the pre-defined threshold parameter; other parameters will be offset during the calculation process, or have no specific physical meaning and are cyclic parameters.

[0132] Finally, the target feature matrix can be input into the preset convolutional neural network model (CNN) for processing to generate weights corresponding to each motion feature value. When establishing the preset convolutional neural network model, the 13×13 target feature matrix can be used as input, and the doctor's score label can be used as the supervised network label to train the preset convolutional neural network model. The network structure is as follows: Fig. 9 As shown, Fig. 9 A schematic diagram of the network structure of a preset convolutional neural network model provided in an embodiment of the present application. The doctor scoring label may include 0, 1, and 2, where 0 indicates that the action cannot be completed at all; 1 indicates that the action can be completed to a certain extent; and 2 indicates that the action can be completed in a standard manner.

[0133] In some optional embodiments, when the target feature matrix is ​​input into a preset convolutional neural network model for processing, the weights corresponding to each motion feature value can be calculated by gradient-weighted class activation mapping (Grad-CAM for short). Grad-CAM is proposed for the traditional CNN architecture and is used to explain the decisions made based on the abstract features learned by the last convolutional layer. Grad-CAM explains why the CNN network makes decisions at the original image layer. In an embodiment of the present application, the content of the CNN decision is the score of the rehabilitation assessment action, and Grad-CAM can explain why the sample score is 0, 1, or 2 at the original feature matrix level. Mathematically, this explanation is expressed as the importance of each pixel in the original image. First, consider a convolutional layer of the network, let A k is the kth feature map of this layer. Next, the score of category c can be calculated with respect to feature map A k Gradient y c is the score of category c.

[0134] Next, these gradients can be globally averaged pooled using formula (15) to obtain the importance weight of each feature map

[0135]

[0136] Among them, i and j are feature maps A k The width and height index of A. k The total number of elements in .

[0137] Then for all feature maps A k Perform weighted summation and apply a ReLU function to get the calculation result As shown in formula (16).

[0138]

[0139] The result is a matrix of the same size as the input target feature matrix, which highlights the most important areas for class prediction, that is, the larger the weight, the larger the value. The weight of each pixel can be calculated by normalizing it, that is, the weight corresponding to each motion feature value.

[0140] Finally, when the quantitative evaluation results of the target motion signal are calculated based on the motion feature values ​​and weights, such as Fig.10 As shown, Fig.10 A flowchart of the steps for calculating the quantitative evaluation results provided in the embodiment of the present application includes:

[0141] Step 1002: Obtain first category information corresponding to each motion feature value and second category information corresponding to each weight.

[0142] Step 1004: Calculate the quantitative evaluation result of the target motion signal based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight and the negative coefficient weight.

[0143] Among them, the first category of information includes positive coefficient eigenvalues ​​and negative coefficient eigenvalues, and the second category of information includes positive coefficient weights and negative coefficient weights; the positive coefficient eigenvalue is used to characterize the positive correlation between the motion characteristic value and the preset action completion degree, and the negative coefficient eigenvalue is used to characterize the negative correlation between the motion characteristic value and the preset action completion degree; the positive coefficient weight corresponds to the positive coefficient eigenvalue, and the negative coefficient weight corresponds to the negative coefficient eigenvalue. Thus, a scoring equation can be obtained as shown in formula (17), and the final quantitative evaluation result is Score. The preset action completion degree is pre-set.

[0144]

[0145] Among them, w i is the positive coefficient weight, w j is a negative coefficient weight, f i is a positive coefficient eigenvalue, f j is the negative coefficient eigenvalue, n is the number of positive contribution features, and m is the number of negative contribution features. During the calculation process, each eigenvalue is normalized to between 0 and 100, so the final quantization result is also in the range of 0-100.

[0146] For example, 36*4 features are matrixed and distributed on a 13*13 target feature matrix A. For example, the first row and the first column represent the value of feature s1 of A11. By calculating (16), a new 13*13 matrix B is obtained, which is This matrix, the first row and first column B11 of the matrix represents the weight of feature s1. A and B correspond to each other. The A matrix represents the numerical value of the feature, that is, the eigenvalue, and the B matrix represents the weight of the feature. In addition, regarding positive and negative contributions, for example, for upper limb forearm movements, the upper arm and lower limbs need to be kept still. Therefore, the amplitude characteristics of the sensor signal at the forearm are positive contributions, and the amplitude characteristics at the upper arm and lower limbs are negative contributions. The number of positive and negative contribution features of each action is different, and specific analysis is required. m and n are used to refer to the number.

[0147] For example, suppose there are only two features, s1 and s2, which are positive and negative respectively. s1 is A11 in matrix A, i.e., the positive coefficient eigenvalue, and B11 in matrix B, i.e., the positive coefficient weight; s2 is A22 in matrix A, i.e., the negative coefficient eigenvalue, and B22 in matrix B, i.e., the negative coefficient weight. The final quantitative evaluation result is Score = A11*B11+(100-A22)*B22.

[0148] The embodiment of the present application provides a method for identifying and quantitatively evaluating rehabilitation movements, the method comprising: obtaining a target motion signal of a patient during rehabilitation, inputting the target motion signal into a preset rehabilitation motion recognition model for processing, generating a target recognition result corresponding to the target motion signal, obtaining a motion feature value and a weight corresponding to the target motion signal based on the target recognition result, and calculating a quantitative evaluation result of the target motion signal based on the motion feature value and the weight. This scheme successfully realizes the intelligent recognition of rehabilitation movements by making full use of the multi-dimensional target motion signal collected by the inertial sensor and using a cascade classification model of a preset random forest and a preset gated recurrent unit; in addition, by real-time recognition of the patient's current rehabilitation movements, the need for doctors in the process of rehabilitation training or rehabilitation evaluation is solved, thereby effectively saving medical resources, and effective rehabilitation training and rehabilitation evaluation can be performed in the absence of a doctor, such as at home, remotely, in a rehabilitation hall, etc.; at the same time, the weight of the signal feature is calculated using the gradient weighted class activation mapping, and a refined quantitative scoring formula for rehabilitation movements is proposed through the feature value and weight, thereby realizing a detailed quantitative scoring of rehabilitation movements. This algorithm not only improves the accuracy of quantitative evaluation, but also provides more comprehensive information support for the personalized formulation of rehabilitation treatment.

[0149] Fig.11 The overall framework diagram of a rehabilitation action recognition and quantitative assessment method provided in the embodiment of the present application, wherein the original signal is the original motion signal. By making full use of the multi-dimensional target motion signal collected by the inertial sensor and using the cascade classification model of the preset random forest and the preset gated recurrent unit, the intelligent recognition of rehabilitation actions is successfully realized; at the same time, the weight of the signal feature is calculated using the gradient weighted class activation mapping, and the refined quantitative scoring formula of the rehabilitation action is proposed through the eigenvalue and weight, which realizes the detailed quantitative scoring of the rehabilitation action and improves the accuracy of the quantitative assessment.

[0150] Fig.12 A structural block diagram of a rehabilitation movement recognition and quantitative assessment device provided in an embodiment of the present application.

[0151] like Fig.12 As shown, the rehabilitation movement recognition and quantitative assessment device 1200 includes:

[0152] The acquisition module 1202 is used to acquire the target motion signal of the patient during the rehabilitation process.

[0153] The generation module 1204 is used to input the target motion signal into the preset rehabilitation action recognition model for processing, and generate a target recognition result corresponding to the target motion signal; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit and a preset fully connected network.

[0154] The calculation module 1206 is used to obtain the motion feature value and weight corresponding to the target motion signal based on the target recognition result, and calculate the quantitative evaluation result of the target motion signal based on the motion feature value and weight.

[0155] Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here. Each module in the above rehabilitation movement recognition and quantitative assessment device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations of the above modules.

[0156] In one embodiment of the present application, a computer device is provided, the computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0157] Obtain the patient's target movement signal during the rehabilitation process;

[0158] The target motion signal is input into a preset rehabilitation action recognition model for processing to generate a target recognition result corresponding to the target motion signal; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network;

[0159] Based on the target recognition result, the motion feature value and weight corresponding to the target motion signal are obtained, and the quantitative evaluation result of the target motion signal is calculated based on the motion feature value and the weight.

[0160] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0161] The target motion signal is input into a preset rehabilitation motion recognition model to extract logical features and generate multiple logical feature values; wherein each logical feature value is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude;

[0162] Based on the target motion signal and a plurality of logical feature values, a target recognition result corresponding to the target motion signal is generated.

[0163] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0164] Inputting multiple logical feature values ​​into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal;

[0165] The target motion signal and the initial recognition result are input into a preset gated recurrent unit for processing, and the processing result is input into a preset fully connected network to generate a target recognition result corresponding to the target motion signal.

[0166] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0167] Based on the target recognition result, the motion feature of the target motion signal is extracted to generate multiple motion feature data and motion feature values ​​corresponding to each motion feature data; wherein the multiple motion feature data include four feature data, the four feature data include amplitude, dynamic time warp, smoothness and approximate entropy, the amplitude is used to characterize the action amplitude corresponding to the target motion signal, the dynamic time warp is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the action fluency corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal;

[0168] Based on the motion feature values ​​corresponding to each motion feature data, a target feature matrix of a preset size is generated;

[0169] The target feature matrix is ​​input into the preset convolutional neural network model for processing to generate weights corresponding to each motion feature value.

[0170] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0171] The target feature matrix is ​​input into the preset convolutional neural network model for processing, and the weights corresponding to each motion feature value are calculated through gradient weighted class activation mapping.

[0172] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0173] Obtaining first category information corresponding to each motion feature value and second category information corresponding to each weight; wherein the first category information includes positive coefficient feature values ​​and negative coefficient feature values, and the second category information includes positive coefficient weights and negative coefficient weights; the positive coefficient feature values ​​are used to characterize that the motion feature value is positively correlated with the preset action completion degree, and the negative coefficient feature values ​​are used to characterize that the motion feature value is negatively correlated with the preset action completion degree; the positive coefficient weights correspond to the positive coefficient feature values, and the negative coefficient weights correspond to the negative coefficient feature values;

[0174] Based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight and the negative coefficient weight, the quantitative evaluation result of the target motion signal is calculated.

[0175] In one embodiment of the present application, when the processor executes the computer program, the processor further implements the following steps:

[0176] Obtain the original movement signals of patients during rehabilitation;

[0177] The original motion signal is subjected to data preprocessing to generate a target motion signal; wherein the data preprocessing operation at least includes signal denoising, signal segmentation and signal length conversion.

[0178] The computer device provided in the embodiment of the present application has similar implementation principles and technical effects to those of the above-mentioned method embodiment, and will not be described in detail here.

[0179] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0180] Obtain the patient's target movement signal during the rehabilitation process;

[0181] The target motion signal is input into a preset rehabilitation action recognition model for processing to generate a target recognition result corresponding to the target motion signal; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network;

[0182] Based on the target recognition result, the motion feature value and weight corresponding to the target motion signal are obtained, and the quantitative evaluation result of the target motion signal is calculated based on the motion feature value and the weight.

[0183] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0184] The target motion signal is input into a preset rehabilitation motion recognition model to extract logical features and generate multiple logical feature values; wherein each logical feature value is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude;

[0185] Based on the target motion signal and a plurality of logical feature values, a target recognition result corresponding to the target motion signal is generated.

[0186] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0187] Inputting multiple logical feature values ​​into a preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal;

[0188] The target motion signal and the initial recognition result are input into a preset gated recurrent unit for processing, and the processing result is input into a preset fully connected network to generate a target recognition result corresponding to the target motion signal.

[0189] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0190] Based on the target recognition result, the motion feature of the target motion signal is extracted to generate multiple motion feature data and motion feature values ​​corresponding to each motion feature data; wherein the multiple motion feature data include four feature data, the four feature data include amplitude, dynamic time warp, smoothness and approximate entropy, the amplitude is used to characterize the action amplitude corresponding to the target motion signal, the dynamic time warp is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the action fluency corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal;

[0191] Based on the motion feature values ​​corresponding to each motion feature data, a target feature matrix of a preset size is generated;

[0192] The target feature matrix is ​​input into the preset convolutional neural network model for processing to generate weights corresponding to each motion feature value.

[0193] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0194] The target feature matrix is ​​input into the preset convolutional neural network model for processing, and the weights corresponding to each motion feature value are calculated through gradient weighted class activation mapping.

[0195] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0196] Obtaining first category information corresponding to each motion feature value and second category information corresponding to each weight; wherein the first category information includes positive coefficient feature values ​​and negative coefficient feature values, and the second category information includes positive coefficient weights and negative coefficient weights; the positive coefficient feature values ​​are used to characterize that the motion feature value is positively correlated with the preset action completion degree, and the negative coefficient feature values ​​are used to characterize that the motion feature value is negatively correlated with the preset action completion degree; the positive coefficient weights correspond to the positive coefficient feature values, and the negative coefficient weights correspond to the negative coefficient feature values;

[0197] Based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight and the negative coefficient weight, the quantitative evaluation result of the target motion signal is calculated.

[0198] In one embodiment of the present application, when the computer program is executed by a processor, the following steps are further implemented:

[0199] Obtain the original movement signals of patients during rehabilitation;

[0200] The original motion signal is subjected to data preprocessing to generate a target motion signal; wherein the data preprocessing operation at least includes signal denoising, signal segmentation and signal length conversion.

[0201] The computer-readable storage medium provided in this embodiment has similar implementation principles and technical effects to those of the above method embodiments, and will not be described in detail here.

[0202] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0203] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0204] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for identifying and quantitatively evaluating rehabilitation movements, characterized in that: The method comprises: Obtain the patient's target movement signal during the rehabilitation process; The target motion signal is input into a preset rehabilitation action recognition model for logical feature extraction to generate a plurality of logical feature values; wherein each of the logical feature values ​​is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude; based on the target motion signal and the plurality of logical feature values, a target recognition result corresponding to the target motion signal is generated; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network; Based on the target recognition result, a motion feature value and a weight corresponding to the target motion signal are obtained, and based on the motion feature value and the weight, a quantitative evaluation result of the target motion signal is calculated.

2. The method according to claim 1, characterized in that The generating a target recognition result corresponding to the target motion signal based on the target motion signal and the multiple logical feature values ​​comprises: Inputting the plurality of logical feature values ​​into the preset random forest network for processing to generate an initial recognition result corresponding to the target motion signal; The target motion signal and the initial recognition result are input into the preset gated recurrent unit for processing, and the processing result is input into the preset fully connected network to generate a target recognition result corresponding to the target motion signal.

3. The method according to claim 1 or 2, characterized in that: The acquiring the motion feature value and the weight corresponding to the target motion signal based on the target recognition result includes: Based on the target recognition result, motion features are extracted from the target motion signal to generate multiple motion feature data and motion feature values ​​corresponding to each of the motion feature data; wherein the multiple motion feature data include four feature data, and the four feature data include amplitude, dynamic time warp, smoothness and approximate entropy, the amplitude is used to characterize the action amplitude corresponding to the target motion signal, the dynamic time warp is used to characterize the similarity between the target motion signal and the standard motion signal, the smoothness is used to characterize the action fluency corresponding to the target motion signal, and the approximate entropy is used to characterize the signal regularity corresponding to the target motion signal; Based on the motion feature values ​​corresponding to each of the motion feature data, generating a target feature matrix of a preset size; The target feature matrix is ​​input into a preset convolutional neural network model for processing to generate weights corresponding to each of the motion feature values.

4. The method according to claim 3, characterized in that The step of inputting the target feature matrix into a preset convolutional neural network model for processing to generate weights corresponding to each of the motion feature values ​​includes: The target feature matrix is ​​input into a preset convolutional neural network model for processing, and the weights corresponding to the motion feature values ​​are calculated by gradient weighted class activation mapping.

5. The method according to claim 3, characterized in that: The quantitative evaluation result of the target motion signal is calculated based on the motion feature value and the weight, comprising: Obtaining first category information corresponding to each of the motion feature values, and second category information corresponding to each of the weights; wherein the first category information includes positive coefficient feature values ​​and negative coefficient feature values, and the second category information includes positive coefficient weights and negative coefficient weights; the positive coefficient feature values ​​are used to characterize that each of the motion feature values ​​is positively correlated with a preset action completion degree, and the negative coefficient feature values ​​are used to characterize that each of the motion feature values ​​is negatively correlated with the preset action completion degree; the positive coefficient weights correspond to the positive coefficient feature values, and the negative coefficient weights correspond to the negative coefficient feature values; Based on the positive coefficient eigenvalue, the negative coefficient eigenvalue, the positive coefficient weight, and the negative coefficient weight, a quantitative evaluation result of the target motion signal is calculated.

6. The method according to claim 1 or 2, characterized in that: The method further comprises: Obtain the original movement signals of patients during rehabilitation; The original motion signal is subjected to a data preprocessing operation to generate the target motion signal; wherein the data preprocessing operation at least includes signal denoising processing, signal segmentation processing and signal length conversion processing.

7. A rehabilitation movement recognition and quantitative assessment device, characterized in that: The device comprises: An acquisition module, used to acquire the target motion signal of the patient during the rehabilitation process; A generation module, used for inputting the target motion signal into a preset rehabilitation action recognition model to extract logical features and generate multiple logical feature values; wherein each of the logical feature values ​​is used to characterize whether there is a valid signal on different limb structures of the patient, as well as information on the signal fluctuation amplitude; based on the target motion signal and the multiple logical feature values, a target recognition result corresponding to the target motion signal is generated; wherein the preset rehabilitation action recognition model includes a cascade model obtained by cascading a preset random forest network, a preset gated recurrent unit, and a preset fully connected network; A calculation module is used to obtain a motion feature value and a weight corresponding to the target motion signal based on the target recognition result, and calculate a quantitative evaluation result of the target motion signal based on the motion feature value and the weight.

8. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the rehabilitation movement recognition and quantitative assessment method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the rehabilitation movement recognition and quantitative assessment method as described in any one of claims 1-6.

Citation Information

Patent Citations

  • Upper limb rehabilitation robot motion intention recognition model training and recognition method and device

    CN114159080A

  • Post-stroke rehabilitation evaluation deep learning model construction method based on brain muscle network graph theory characteristics

    CN114587385A

  • Cerebral stroke rehabilitation evaluation method based on brain muscle function network characteristics

    CN115316998A

  • Human body lower limb rehabilitation training intelligent identification method and system

    CN115349853A

  • Rehabilitation training evaluation assisting method and device, computer equipment and storage medium

    CN117392746A