A method and system for analyzing tremor frequency spectrum of a pen-holding posture based on a smart wearable device

Abstract

The application discloses a kind of based on smart wearable device's pen posture tremor spectrum analysis method and system, belong to smart wearable device technical field, including in habitual hand wrist deployment first equipment and as first wrist, in auxiliary hand wrist deployment second equipment and as second wrist, establish double-wrist sampling clock alignment, synchronously collect the three-axis acceleration sequence and three-axis gyroscope sequence of first wrist and second wrist and other steps.The application solves the technical problem that single-wrist device cannot distinguish real microvibration and writing pseudo-tremor and the dynamic change of pen posture leads to the distortion of spectral characteristics, realizes the physical separation of real physiological microvibration and writing action pseudo-tremor and the extraction of standardized spectrum under gravity coordinate system;Solve the technical problem that the existing technology directly discards writing pseudo-tremor, leading to information waste and only being able to evaluate writing quality from a single physiological dimension, realizes the energy recovery of discarded signal and physiological mechanics two-dimensional writing quality evaluation.

Description

Technical Field

[0001] This invention belongs to the field of smart wearable device technology, specifically relating to a method and system for analyzing the spectral tremor of pen-holding posture based on smart wearable devices. Background Technology

[0002] As smart wearable devices become increasingly prevalent in people's daily lives, users can monitor their daily activities and analyze fine motor skills by wearing them. In the high-frequency daily scenario of writing, users expect to use wrist-worn wearable devices to assess their writing posture and quality in real time, in order to correct writing direction and improve writing stability. Current technologies primarily focus on medical diagnostic scenarios such as Parkinson's disease, typically using single-wrist acquisition to assess tremor signals for disease grading. A non-medical writing quality analysis system for everyday writing scenarios in healthy individuals has not yet been established.

[0003] Currently, in pen-writing scenarios, existing single-wrist smart wearable devices collect wrist acceleration signals that simultaneously contain genuine physiological micro-vibrations and pseudo-tremors from writing movements. These two frequencies highly overlap, making effective separation impossible with a single wrist device. Furthermore, current technology lacks a pen-holding posture compensation mechanism, leading to distortions in the micro-vibration spectrum due to dynamic posture changes. Additionally, existing technology directly discards pseudo-tremors from writing movements as noise, wasting writing dynamics information and lacking non-medical dynamic assessment methods for fine motor skills in everyday writing scenarios for healthy individuals. For these reasons, current technology cannot accurately assess writing quality from both physiological and mechanical dimensions, thus hindering the support for correcting writing direction and dynamically monitoring writing stability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention aims to provide a method and system for analyzing the spectral characteristics of pen-holding posture tremors based on smart wearable devices. This solves the technical problems of single-wrist devices being unable to distinguish between genuine micro-vibrations and pseudo-tremors during writing, as well as the distortion of spectral characteristics caused by dynamic changes in pen-holding posture. It achieves the physical separation of genuine physiological micro-vibrations from pseudo-tremors during writing and the standardized spectral extraction under a gravity coordinate system. The specific solution is as follows: In a first aspect, embodiments of this application provide a method including the following steps: The first device is deployed on the dominant wrist and serves as the first wrist, and the second device is deployed on the auxiliary wrist and serves as the second wrist. The sampling clock is aligned between the two wrists, and the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist are collected synchronously. Real-time attitude angles are calculated based on the three-axis gyroscope and three-axis acceleration sequences of the first wrist, and a projection matrix is ​​constructed. Cross-power spectral density analysis was performed on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action; The real physiological micro-vibration components and the pseudo-tremor components of the writing action are projected onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum. A windowed fast Fourier transform is performed on the real micro-vibration spectrum to calculate the energy integrals in the low-frequency, mid-frequency, and high-frequency bands, yielding the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index. The main frequency is extracted and the energy concentration is analyzed on the pseudo-tremor spectrum to obtain the pen stroke force index, velocity stability index, and acceleration change rate index. The low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index are input into a pre-trained multilayer perceptron regression model to obtain a comprehensive score for writing quality. By summarizing the overall writing quality scores, the results of the pen-holding posture tremor spectrum analysis were obtained.

[0005] Secondly, embodiments of this application provide a pen-holding posture tremor spectrum analysis system based on a smart wearable device, comprising: The data acquisition unit is used to deploy a first device on the dominant wrist as the first wrist, and deploy a second device on the auxiliary wrist as the second wrist, establish dual-wrist sampling clock alignment, and synchronously acquire the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist. The attitude calculation unit is used to calculate the real-time attitude angles and construct the projection matrix based on the three-axis gyroscope sequence and three-axis acceleration sequence of the first wrist. The pseudo-tremor separation unit is used to perform cross-power spectral density analysis on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action. The posture compensation unit is used to project the real physiological micro-vibration components and the pseudo-tremor components of writing action onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum. The spectrum decomposition unit is used to perform windowed fast Fourier transform on the real micro-vibration spectrum, calculate the energy integrals of the low-frequency, mid-frequency, and high-frequency bands, and obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index. The energy recovery unit is used to extract the main frequency and analyze the energy concentration of the pseudo-tremor spectrum to obtain the pen stroke strength index, velocity stability index, and acceleration change rate index. The coupled evaluation unit is used to input the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index into the pre-trained multilayer perceptron regression model to obtain a comprehensive score of writing quality. The dynamic monitoring unit is used to summarize the comprehensive score of writing quality and obtain the results of the pen-holding posture tremor spectrum analysis. Beneficial effects

[0006] (1) This invention deploys smart wearable devices on the first and second wrists and establishes dual-wrist sampling clock alignment. It performs cross-power spectral density analysis on the dual-wrist triaxial acceleration sequence to extract the real physiological micro-vibration component and the pseudo-tremor component of writing action. Based on the triaxial gyroscope sequence and triaxial acceleration sequence of the first wrist, it calculates the real-time attitude angle to construct a projection matrix. The two types of components are projected onto the gravity coordinate system through the projection matrix. This solves the technical problem that single-wrist devices cannot distinguish between real micro-vibration and pseudo-tremor of writing action and the distortion of spectral characteristics caused by dynamic changes in pen holding posture. It realizes the physical separation of real physiological micro-vibration and pseudo-tremor of writing action and the standardized spectrum extraction under the gravity coordinate system. (2) This invention obtains low-frequency stability index, mid-frequency fatigue index and high-frequency control accuracy index by performing frequency band division on the real micro-vibration spectrum, and obtains pen stroke force index, speed stability index and acceleration change rate index by extracting the main frequency and analyzing the energy concentration of the pseudo-vibration spectrum. The above six indices are input into the pre-trained multilayer perceptron regression model to obtain a comprehensive score of writing quality. This solves the technical problems of the prior art that directly discards pseudo-vibration of writing, resulting in information waste and can only evaluate writing quality from a single physiological dimension. It realizes the energy recovery of discarded signals and the physiological and mechanical dual-dimensional writing quality evaluation. (3) This invention establishes a baseline model of personal writing micro-vibration spectrum based on the comprehensive writing quality score and the mid-frequency fatigue index, tracks the monotonically rising trend of the mid-frequency fatigue index and the decay trend of the high-frequency control precision index, and calculates the deviation from the baseline model. When the deviation exceeds the safety boundary threshold, it triggers a fine motor fatigue warning. After the rest interval, it calculates the recovery rate of the low-frequency stability index and the high-frequency control precision index relative to the baseline model. This solves the technical problem that existing tremor detection technology only supports medical static disease assessment and cannot cover non-medical dynamic monitoring of writing quality. It realizes dynamic monitoring of fine motor ability and early warning of fatigue in daily writing scenarios for healthy people. Attached Figure Description

[0007] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0008] Figure 1 A flowchart illustrating the pen-holding posture tremor spectrum analysis method based on smart wearable devices provided by the present invention; Figure 2 This is a schematic block diagram of the pen-holding posture tremor spectrum analysis system based on a smart wearable device provided by the present invention. Detailed Implementation

[0009] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0010] 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.

[0011] It should be understood that, when used in this specification and the appended claims, terms include and encompass the presence of the described features, integrals, steps, operations, elements, and components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and sets thereof.

[0012] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "one," "an," and "that" are intended to include the plural forms.

[0013] It should also be further understood that the terms used in this application specification and the appended claims refer to any combination of one or more of the associated listed items and all possible combinations, and include such combinations.

[0014] Furthermore, in this application, unless otherwise explicitly specified or limited in the embodiments, the terms installation, connection, linking, and fixing appearing in the embodiments should be interpreted broadly. For example, a connection can be a fixed connection, a detachable connection, or an integral part; it can also be a mechanical connection, an electrical connection, etc. Of course, it can also be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication between two components, or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific implementation.

[0015] In related technologies, the following problems typically exist when analyzing wrist micro-vibration signals during writing: Existing technologies all use single-wrist devices to collect hand motion signals. During the writing process, the wrist acceleration signal contains both real physiological micro-vibrations and pseudo-tremors generated by the writing action itself. Existing technologies use high-pass filtering or simple thresholding to remove so-called motion artifacts. However, the frequency components of the writing action highly overlap with the frequency bands of physiological micro-vibrations and muscle fatigue tremors, causing real micro-vibration signals to be filtered out simultaneously, making it impossible to obtain a pure physiological spectrum of the writing state. The accelerometer coordinate system of existing wearable devices is fixed to the device body. During the writing process, the wrist posture changes continuously and dynamically, causing the projection component of the same real micro-vibration in the device coordinate system to undergo pseudo-changes. Existing technologies have not established a mechanism for calculating the pen-holding posture angle and compensating for the coordinate system, which makes the spectrum analysis results mixed with posture modulation interference. It is impossible to distinguish whether the spectrum change is due to changes in physiological state or changes in pen-holding posture, resulting in a significant reduction in the reliability of writing quality assessment.

[0016] Existing technologies treat the acceleration component generated by the writing action as noise that needs to be suppressed and discard or filter it directly. However, this pseudo-tremor component contains writing dynamics characteristics such as pen force, speed and rate of change of acceleration, which are physically related to writing quality. Existing technologies have not established an energy recovery and reuse mechanism for pseudo-tremors, resulting in waste of signal information and making it impossible to comprehensively evaluate writing quality from both physiological and mechanical dimensions.

[0017] Existing tremor detection patents all limit their application to disease diagnosis, and their assessment output is the severity level or disease type. Existing technologies do not cover the assessment of fine motor skills, dynamic monitoring of writing fatigue, and tracking of recovery levels in healthy individuals in daily writing scenarios, leaving a clear application gap.

[0018] Therefore, this application provides a pen-holding posture tremor spectrum analysis method based on smart wearable devices, which can provide a theoretical basis for the quality assessment of fine motor skills in daily writing scenarios. The pen-holding posture tremor spectrum analysis method based on smart wearable devices provided in this application is applied to a terminal device, and the method is executed through application software installed on the terminal device. The terminal device can be a smartphone, tablet computer, laptop computer, desktop computer, personal digital assistant, or wearable device, etc.

[0019] It should be noted that the application scenarios described in the following embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0020] The following is a detailed description of the pen-holding posture tremor spectrum analysis method based on smart wearable devices provided in this application.

[0021] like Figure 1 As shown, the method includes the following steps S110 to S180.

[0022] S110. Deploy the first device on the dominant wrist as the first wrist, and deploy the second device on the auxiliary wrist as the second wrist. Establish dual-wrist sampling clock alignment and synchronously collect the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist.

[0023] Specifically, both the first and second devices are smart wearable devices with built-in three-axis MEMS accelerometers and three-axis MEMS gyroscopes. The accelerometer has an upper range of 16g, a lower range of -16g, a 16-bit resolution, and a sampling frequency of 100Hz. The gyroscope has an upper range of 2000dps, a lower range of -2000dps, a 16-bit resolution, and a sampling frequency of 100Hz. The two devices establish a communication link via Bluetooth 5.0 and use the Bluetooth time synchronization protocol to achieve nanosecond-level sampling clock alignment, with a timestamp alignment accuracy better than 1ms.

[0024] The above sampling frequency of 100Hz is based on the following: According to the Nyquist sampling theorem, the main frequency components of writing action and micro-vibration are below 50Hz. A sampling frequency of 100Hz can meet the signal fidelity requirements while taking into account the power consumption limitations of wearable devices.

[0025] During the writing process, the first device collects the three-axis acceleration sequence and the three-axis gyroscope sequence of the first wrist in real time, and the second device collects the three-axis acceleration sequence and the three-axis gyroscope sequence of the second wrist in real time. The above four sets of sequences are packaged with a unified timestamp and transmitted to the edge computing terminal via Bluetooth Low Energy to obtain the dual-wrist synchronous raw data stream.

[0026] S120. Calculate the real-time attitude angle based on the three-axis gyroscope sequence and three-axis acceleration sequence of the first wrist, and construct the projection matrix.

[0027] Specifically, after receiving the three-axis gyroscope sequence and the three-axis acceleration sequence of the first wrist, the edge computing terminal uses a complementary filtering algorithm to calculate the real-time attitude angle of the first wrist relative to the gravity coordinate system. This algorithm uses the component of the three-axis acceleration sequence of the first wrist in the gravity direction to estimate the pitch angle and roll angle, and uses the integral of the angular velocity of the three-axis gyroscope sequence of the first wrist to estimate the rate of change of attitude angle. By combining the advantages of the two through complementary filtering, the filtering coefficient is set to 0.98, and the real-time pitch angle, roll angle and yaw angle sequence with a frequency of 100Hz is output.

[0028] Let the measurement value of the triaxial accelerometer on the first wrist be... The angular velocity measurement value of the three-axis gyroscope is Pitch angle estimated by accelerometer With roll angle for: ; ; The recursive formula for complementary filtering is: In the formula, The filter coefficient is set to 0.98. The sampling time interval is in seconds. =1 / 100=0.01s; k is the sampling time sequence number; Estimate the pitch angle for the accelerometer, in degrees; Estimate the roll angle for the accelerometer, in degrees.

[0029] The filtering coefficient of 0.98 is chosen based on the fact that the accelerometer has a high signal-to-noise ratio in the low-frequency band and the gyroscope has a high signal-to-noise ratio in the high-frequency band. The coefficient of 0.98 makes the gyroscope integration dominate the high-frequency attitude estimation and the accelerometer dominate the low-frequency correction. This value can provide the best attitude tracking accuracy in writing scenarios.

[0030] Based on this real-time attitude angle sequence, a projection matrix from the first wrist acceleration coordinate system to the gravity coordinate system is constructed. Let the real-time pitch angle be... The roll angle is yaw angle is Then the projection matrix R is: ; In the formula, the specific elements of each rotation matrix are: In the formula, R is the projection matrix; Let be the rotation matrix about the x-axis; Let be the rotation matrix about the y-axis; Let be the rotation matrix about the z-axis; The pitch angle is expressed in degrees (°). The roll angle is expressed in degrees (°). Yaw angle, in degrees.

[0031] S130. Perform cross-power spectral density analysis on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action.

[0032] Specifically, the edge computing terminal performs windowed Fourier transform on the triaxial acceleration sequences of the first wrist and the second wrist to obtain the self-power spectral density of the first wrist and the self-power spectral density of the second wrist, and calculates the cross-power spectral density, the coherence coefficient of the two wrists and the phase difference between the two wrists.

[0033] The windowed Fourier transform uses the Hanning window, and the window function... Defined as: ; In the formula, n is the sample number within the window, n=0,1,…,N−1; N is the window length, which is 1024.

[0034] Let the triaxial acceleration sequence of the first wrist be... The second wrist triaxial acceleration sequence is as follows Windowed Fourier transform of both yields and Then the cross power spectral density for: ; In the formula, Cross-power spectral density, in g 2 / Hz; Fourier transform of the first wrist acceleration sequence; Fourier transform of the second wrist acceleration sequence; for The complex conjugate of f; f is the frequency in Hz; t is the time in seconds.

[0035] Calculation of coherence coefficient of the two arms from cross power spectral density Phase difference with both wrists : In the formula, The coherence coefficient of the two wrists; The power spectral density of the first wrist, in g. 2 / Hz; The power spectral density of the second wrist, in g. 2 / Hz; is the phase difference between the two wrists, in degrees; arg is the complex argument operation.

[0036] The coherence threshold was set to 0.75, the phase difference threshold to 30°, and the amplitude ratio threshold to 3dB. These thresholds were set based on the following: In healthy individuals, the acceleration coherence coefficient of both wrists at rest is typically higher than 0.8. During writing, the coherence coefficient decreases slightly due to passive transmission from the non-pen-holding wrist; a value of 0.75 ensures accurate extraction of physiological vibrations while avoiding excessive inclusion of writing artifacts. Physiological tremors are synchronously transmitted through the central nervous system, and the phase difference between the two wrists is typically less than 20°; considering sampling noise and Bluetooth synchronization errors, this is relaxed to 30°. The writing action is primarily driven by the pen-holding wrist, and the passive acceleration amplitude of the non-pen-holding wrist is typically less than 50% of that of the pen-holding wrist, i.e., attenuation exceeds 6dB; a conservative boundary of 3dB is chosen.

[0037] Within a frequency band where the coherence coefficient of both wrists is higher than the coherence determination threshold and the phase difference between the two wrists is in a synchronous phase state, the synchronous micro-vibration component of both wrists is extracted as the real physiological micro-vibration component; within a frequency band where the amplitude of the self-power spectral density of the first wrist is higher than the amplitude of the self-power spectral density of the second wrist and the amplitude difference exceeds the amplitude ratio determination threshold, the dominant vibration component of the first wrist is extracted as the pseudo-tremor component of the writing action.

[0038] S140. Project the real physiological micro-vibration component and the pseudo-tremor component of the writing action onto the gravity coordinate system through the projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum.

[0039] Specifically, the real physiological micro-vibration components are projected onto the gravity coordinate system through a projection matrix to obtain a unified coordinate system real micro-vibration sequence. A windowed fast Fourier transform is performed on the unified coordinate system real micro-vibration sequence to obtain the real micro-vibration spectrum. The pseudo-tremor components of the writing action are projected onto the gravity coordinate system through a projection matrix to obtain a unified coordinate system pseudo-tremor sequence. A windowed fast Fourier transform is performed on the unified coordinate system pseudo-tremor sequence to obtain the pseudo-tremor spectrum.

[0040] The discrete calculation formula for the windowed Fast Fourier Transform is as follows: In the formula, k is the frequency index, k=0,1,…,N−1; N is the number of transformation points, taken as 1024; j is the imaginary unit; For the Hanning window function; To unify the coordinate system for the nth sampling point of the real micro-vibration sequence; The nth sampling point of the pseudo-tremor sequence in a unified coordinate system.

[0041] The Fast Fourier Transform (FFT) described above uses a 1024-point Hanning window with a 50% overlap. The 1024-point window size is chosen because, at a sampling rate of 100Hz, 1024 points correspond to a 10.24-second time window, with a frequency resolution of approximately 0.098Hz, sufficient to distinguish low-frequency flutter components around 0.5Hz. The 50% overlap is chosen because, in time-frequency analysis, 50% overlap can reduce the variance of spectral estimation between adjacent windows by approximately 30% without significantly increasing computational load.

[0042] S150. Perform frequency band division on the real micro-vibration spectrum, calculate the energy integral of the low-frequency band, mid-frequency band, and high-frequency band, and obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index.

[0043] Specifically, the low-frequency band is set to 0.5Hz to 2Hz, the mid-frequency band to 2Hz to 8Hz, and the high-frequency band to 8Hz to 15Hz. The above frequency band boundaries are set based on the following: 0.5Hz to 2Hz corresponds to the dominant frequency of physiological tremor, 2Hz to 8Hz corresponds to muscle fatigue tremor, and 8Hz to 15Hz corresponds to the frequency band of neuromuscular fine adjustment.

[0044] In the actual micro-vibration spectrum, spectral energy integration is performed for the low-frequency, mid-frequency, and high-frequency bands respectively. The discrete calculation of the spectral energy integration adopts the trapezoidal summation method: In the formula, It is a low-frequency stability index; This refers to the mid-frequency fatigue index. This refers to the high-frequency control accuracy index. These are discrete frequency points, in Hz. Frequency resolution, in Hz. =100 / 1024≈0.097Hz, which is the sampling frequency.

[0045] The low-frequency energy integral value is used as the low-frequency stability index, the mid-frequency energy integral value is used as the mid-frequency fatigue index, and the high-frequency energy integral value is used as the high-frequency control accuracy index.

[0046] S160. Extract the main frequency and analyze the energy concentration of the pseudo-tremor spectrum to obtain the pen stroke strength index, velocity stability index, and acceleration change rate index.

[0047] Specifically, the frequency corresponding to the amplitude peak in the pseudo-tremor spectrum is searched as the main frequency, the amplitude of the main frequency is extracted, and the pen stroke strength index is defined as: In the formula, This refers to the index of brushstroke strength. This is the force amplitude conversion coefficient; The main frequency amplitude, in grams (g).

[0048] The above force amplitude conversion coefficient The calibration method is as follows: Multiple sets of writing samples with known pen pressure are collected. The actual pen pressure value is obtained through a pressure sensor. Simultaneously, the amplitude of the main frequency of the corresponding pseudo-vibration spectrum is extracted. A mapping relationship between the main frequency amplitude and pen pressure is established through linear regression. The regression slope is the... .

[0049] Calculate the uniformity of energy distribution of the pseudo-tremor spectrum within the horizontal plane of the gravity coordinate system. Let the spectral energy in the x-direction within the horizontal plane be... The spectral energy in the y-direction is It is defined as the power spectral density integral of the pseudo-tremor spectrum in the horizontal plane of the gravity coordinate system along the x-axis and y-axis: In the formula, The frequency component of the pseudo-tremor spectrum in the x-axis direction of the gravity coordinate system; The frequency component in the y-axis direction; This represents frequency resolution, measured in Hz.

[0050] It is obtained by summing the spectral energy in the x-direction and the spectral energy in the y-direction within the horizontal plane: ; The speed stability index is defined as: ; In the formula, The velocity stability index; The energy of the spectrum in the x-direction, in g. 2 / Hz; The energy of the spectrum in the y-direction, in grams. 2 / Hz; It is the sum of the spectral energy in the x-direction and the spectral energy in the y-direction within the horizontal plane, in units of g. 2 / Hz.

[0051] The above It is obtained by summing the spectral energy in the x-direction and the spectral energy in the y-direction within the horizontal plane.

[0052] The pseudo-tremor spectrum is transformed to the time domain to obtain a time-domain pseudo-tremor sequence. First-order difference is performed on the time-domain pseudo-tremor sequence to extract the peak value of the rate of change of acceleration. The rate of change of acceleration exponent is defined as: ; In the formula, The rate of change of acceleration is expressed in m / s². 3 ; The current sampling point of the time-domain pseudo-tremor sequence, in m / s. 2 ; The previous sampling point, in m / s 2 ; At the current sampling time, The previous sampling time; The sampling time interval is set to 0.01s; max is the maximum value calculation.

[0053] S170. Input the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index into the pre-trained multilayer perceptron regression model to obtain a comprehensive writing quality score.

[0054] Specifically, a multilayer perceptron regression model is constructed. This model includes an input layer, a hidden layer, and an output layer. The input layer has 6 nodes, corresponding to the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index, respectively. The hidden layer has 3 layers, with 64 nodes in each layer. The output layer has 1 node and outputs a comprehensive score of writing quality.

[0055] The forward propagation calculation process of this model is as follows: In the formula, x is the input vector. ; This is the weight matrix from the input layer to the first hidden layer, with dimensions 64×6; This is the bias vector for the first hidden layer, with dimensions 64×1; This is the weight matrix from the 1st hidden layer to the 2nd hidden layer, with a dimension of 64×64; This is the bias vector for the second hidden layer, with dimensions 64×1; This is the weight matrix for the 2nd to 3rd hidden layers, with a dimension of 64×64; This is the bias vector for the 3rd hidden layer, with dimensions 64×1; This is the weight matrix from the 3rd hidden layer to the output layer, with a dimension of 1×64; is the output layer bias, with a dimension of 1×1; f is the modified linear unit activation function.

[0056] weight matrix to The bias vector is initialized using the Xavier initialization method. to and Initialize to a zero vector.

[0057] The hidden layer uses the modified linear unit activation function, which is defined as: ; In the formula, f(x) is the output of the modified linear unit activation function; x is the input of the hidden layer neuron.

[0058] Model training uses the mean squared error loss function, which is defined as: ; In the formula, L is the mean squared error loss function value; N is the total number of training samples; The expert-labeled scores for the i-th sample are dimensionless. Let i be the model prediction score for the i-th sample; i is the sample number.

[0059] In the backpropagation algorithm, the update formulas for weights and biases are: In the formula, The learning rate is set to 0.001. For layer number, =1,2,3,4; L is the mean squared error loss function value.

[0060] The model was trained using the Adam optimizer with an initial learning rate of 0.001, a batch size of 32, and 100 training epochs. L2 regularization was also used with a regularization coefficient of 0.001.

[0061] The initial learning rate of 0.001 is chosen based on the following: For small multilayer perceptron networks, an initial learning rate of 0.001 can ensure the stability of gradient descent while avoiding slow convergence. This value has been verified as optimal within the range of 0.0001 to 0.01 through grid search. The batch size of 32 is chosen based on the following: A batch of 32 samples can provide stable gradient direction estimation within a single batch, while avoiding excessive memory consumption caused by excessively large batches. The number of hidden layer nodes of 64 is chosen based on the following: 64 hidden nodes can provide approximately 384 parameters for the first hidden layer, which is sufficient to fit the nonlinear relationship between the six indices and the handwriting quality score, while avoiding overfitting. The number of training epochs of 100 is chosen based on the following: Within 100 epochs, the validation set loss usually converges to a plateau. Continuing training will not significantly improve performance but will increase the risk of overfitting. The L2 regularization coefficient of 0.001 is chosen based on the following: A decay coefficient of 0.001 can suppress overfitting while retaining sufficient model expressive power.

[0062] The training process includes the following steps: Low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index of historical writing quality samples are used as input features, and the corresponding expert-annotated scores are used as the target output; the dataset is divided into a training set and a validation set, with the training set accounting for 80% and the validation set accounting for 20%; during training, the model output is calculated through forward propagation, the prediction error is calculated through the mean squared error loss function, and the gradient is calculated and the network weights are updated through the backpropagation algorithm; training is stopped when the validation set loss decreases by less than 0.001 for 10 consecutive rounds or when the training rounds reach 100, resulting in a pre-trained multilayer perceptron regression model.

[0063] The expert-annotated scores mentioned above are based on the quality assessment of the writing samples by calligraphy teachers or professional writing evaluators, with a score range of 0 to 100.

[0064] During the online inference phase, the edge computing terminal inputs the six indices of the current writing event into the pre-trained multilayer perceptron regression model to calculate a comprehensive writing quality score. The score ranges from 0 to 100, with higher values ​​indicating better writing quality.

[0065] S180. Summarize the comprehensive writing quality score and obtain the results of the pen-holding posture tremor spectrum analysis.

[0066] Specifically, a baseline model of the personal handwriting micro-vibration spectrum is established based on the comprehensive handwriting quality score and the mid-frequency fatigue index. In the personal handwriting micro-vibration spectrum baseline model, the average mid-frequency fatigue index is the individual baseline value. Compared with individual baseline standard deviation The calculation formula is: In the formula, M represents the number of historical writing sessions, and M≥10; The average mid-frequency fatigue index for the m-th historical writing session. The individual baseline mean of the overall writing quality score. The calculation formula is: In the formula, The average score of the overall writing quality for the m-th historical writing session.

[0067] The sliding time window tracks the monotonically increasing trend of the mid-frequency fatigue index and the decreasing trend of the high-frequency control precision index during the current writing session. The sliding window length is 60 seconds, and the step size is 10 seconds. The above values ​​of 60-second sliding window length and 10-second step size are based on the following: according to the time constant of muscle fatigue accumulation, short-term writing fatigue is significantly manifested within 30 to 120 seconds; the 60-second window can capture a stable trend, and the 10-second step size ensures the update frequency.

[0068] Calculate the standardized deviation of the mid-frequency fatigue index relative to the individual baseline mean within the current window: ; In the formula, D is the standardized deviation. This represents the mid-frequency fatigue index for the current window. This represents the individual baseline mean of the mid-frequency fatigue index. The individual baseline standard deviation is the mid-frequency fatigue index.

[0069] The safety boundary threshold is set at 2.0. The above threshold is set based on the 95% confidence interval of the normal distribution, i.e., the mean ± 2 standard deviations. When the deviation exceeds 2.0, the fatigue state is judged to be significantly deviating from the individual's normal baseline.

[0070] When the deviation exceeds the safety boundary threshold, a fine motor fatigue warning is triggered, which notifies the user through vibration of the wearable device or a screen notification.

[0071] The period from the triggering of the fine motor fatigue warning to the resumption of the writing session is designated as the rest interval. After the rest interval, the low-frequency stability index and high-frequency control accuracy index are re-acquired, and their recovery rate relative to the personal writing micro-vibration spectrum baseline model is calculated: ; In the formula, The recovery rate; The index is the result of a rest period; The index before the rest period; Rest duration, measured in seconds (s).

[0072] The recovery rate is used to assess the user's fine motor skills recovery level, where the historical average recovery rate is calculated using the following formula: ; In the formula, H represents the number of historical rest intervals; Let be the recovery rate at the h-th historical rest interval. If the current recovery rate... Higher than the historical average recovery rate If the recovery rate is good, it is considered good; if it is lower than the historical average recovery rate. If so, the user is prompted to extend their rest or adjust their writing posture.

[0073] The entire system forms a complete closed loop for non-medical writing quality analysis, from raw data acquisition from both wrists, pen-holding posture compensation, pseudo-tremor separation, spectral decomposition, two-dimensional coupling assessment to dynamic monitoring of individual baselines.

[0074] This application also provides a pen-holding posture tremor spectrum analysis system based on a smart wearable device, which is used to perform any of the aforementioned pen-holding posture tremor spectrum analysis methods based on smart wearable devices.

[0075] Specifically, please see Figure 2 , Figure 2 This is a schematic block diagram of a pen-holding posture tremor spectrum analysis device based on a smart wearable device provided in an embodiment of this application.

[0076] like Figure 2 As shown, the pen-holding posture tremor spectrum analysis device based on smart wearable devices includes: a data acquisition unit, a posture calculation unit, a pseudo-tremor separation unit, a posture compensation unit, a spectrum decomposition unit, an energy recovery unit, a coupling evaluation unit, and a dynamic monitoring unit.

[0077] The data acquisition unit is used to deploy a first device on the dominant wrist as the first wrist, and deploy a second device on the auxiliary wrist as the second wrist, establish dual-wrist sampling clock alignment, and synchronously acquire the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist. The attitude calculation unit is used to calculate the real-time attitude angles and construct the projection matrix based on the three-axis gyroscope sequence and three-axis acceleration sequence of the first wrist. The pseudo-tremor separation unit is used to perform cross-power spectral density analysis on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action. The posture compensation unit is used to project the real physiological micro-vibration components and the pseudo-tremor components of writing action onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum. The spectrum decomposition unit is used to divide the real micro-vibration spectrum into frequency bands, calculate the energy integrals of the low-frequency band, mid-frequency band, and high-frequency band, and obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index. The energy recovery unit is used to extract the main frequency and analyze the energy concentration of the pseudo-tremor spectrum to obtain the pen stroke strength index, velocity stability index, and acceleration change rate index. The coupled evaluation unit is used to input the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index into the pre-trained multilayer perceptron regression model to obtain a comprehensive score of writing quality. The dynamic monitoring unit is used to summarize the comprehensive score of writing quality and obtain the results of the pen-holding posture tremor spectrum analysis.

[0078] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned pen-holding posture tremor spectrum analysis device and its various units based on smart wearable devices can be referred to the corresponding descriptions in the foregoing method embodiments. For the sake of convenience and brevity, these details will not be repeated here.

[0079] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method and system for analyzing the spectral density of pen-holding posture based on a smart wearable device, characterized in that, Includes the following steps: The first device is deployed on the dominant wrist and serves as the first wrist, and the second device is deployed on the auxiliary wrist and serves as the second wrist. The sampling clock is aligned between the two wrists, and the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist are collected synchronously. Real-time attitude angles are calculated based on the three-axis gyroscope and three-axis acceleration sequences of the first wrist, and a projection matrix is ​​constructed. Cross-power spectral density analysis was performed on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action; The real physiological micro-vibration components and the pseudo-tremor components of the writing action are projected onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum. The real micro-vibration spectrum is divided into frequency bands, and the energy integrals of the low-frequency, mid-frequency, and high-frequency bands are calculated to obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index. The pseudo-tremor spectrum is subjected to main frequency extraction and energy concentration analysis to obtain the pen stroke force index, velocity stability index, and acceleration change rate index. The low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index are input into a pre-trained multilayer perceptron regression model to obtain a comprehensive score for writing quality. By summarizing the overall writing quality scores, the results of the pen-holding posture tremor spectrum analysis were obtained.

2. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, Based on the three-axis gyroscope sequence and three-axis acceleration sequence of the first wrist, the real-time attitude angle is calculated, and a projection matrix is ​​constructed, including: The pitch and roll angles are estimated using the components of the three-axis acceleration sequence of the first wrist in the direction of gravity, and the attitude angle change rate is estimated using the integral of the angular velocity of the three-axis gyroscope sequence of the first wrist. The real-time pitch, roll, and yaw angle sequences are obtained by complementary filtering and fusion. A projection matrix is ​​constructed based on the real-time pitch, roll, and yaw angle sequences.

3. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, Cross-power spectral density analysis was performed on the triaxial acceleration sequences of the first and second wrists to extract the true physiological micro-vibration components and the pseudo-tremor components of the writing motion, including: Windowed Fourier transforms are performed on the triaxial acceleration sequences of the first and second wrists to obtain the self-power spectral density of the first and second wrists. The cross-power spectral density, coherence coefficient of the two wrists, and phase difference between the two wrists are then calculated. Within the frequency band where the coherence coefficient of both wrists is higher than the empirical judgment value and the phase difference between the two wrists is in a synchronous phase state, the synchronous micro-vibration component of both wrists is extracted as the real physiological micro-vibration component. Within the frequency band where the amplitude of the self-power spectral density of the first wrist is higher than that of the amplitude of the self-power spectral density of the second wrist, the dominant vibration component of the first wrist is extracted as the pseudo-tremor component of the writing action.

4. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, The true physiological micro-vibration components and the pseudo-tremor components of the writing motion are projected onto the gravity coordinate system using a projection matrix to obtain the true micro-vibration spectrum and the pseudo-tremor spectrum, including: The real physiological micro-vibration components are projected onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration sequence in the unified coordinate system after attitude compensation. A windowed fast Fourier transform is then performed on the real micro-vibration sequence in the unified coordinate system to obtain the real micro-vibration spectrum. The pseudo-tremor components of the writing action are projected onto the gravity coordinate system through a projection matrix to obtain the pseudo-tremor sequence in the unified coordinate system after attitude compensation. A windowed fast Fourier transform is then performed on the pseudo-tremor sequence in the unified coordinate system to obtain the pseudo-tremor spectrum.

5. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, The actual micro-vibration spectrum is divided into frequency bands, and the energy integrals of the low-frequency, mid-frequency, and high-frequency bands are calculated to obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index, including: Set the boundary frequencies for the low-frequency band, mid-frequency band, and high-frequency band; In the real micro-vibration spectrum, spectral energy integration is performed on the low-frequency, mid-frequency, and high-frequency bands respectively; The low-frequency energy integral value is used as the low-frequency stability index, the mid-frequency energy integral value is used as the mid-frequency fatigue index, and the high-frequency energy integral value is used as the high-frequency control accuracy index.

6. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, The pseudo-tremor spectrum was analyzed by extracting the dominant frequency and performing energy concentration analysis to obtain the pen stroke force index, velocity stability index, and acceleration rate of change index, including: Search for the frequency corresponding to the peak amplitude in the pseudo-tremor spectrum as the main frequency, extract the main frequency amplitude, and convert the main frequency amplitude into the pen stroke strength index. Calculate the uniformity of energy distribution of the pseudo-tremor spectrum in the horizontal plane of the gravity coordinate system, and convert the uniformity of energy distribution into a velocity stability index; The pseudo-tremor spectrum is transformed to the time domain to obtain the time-domain pseudo-tremor sequence. The time-domain pseudo-tremor sequence is then subjected to first-order difference to extract the peak value of the rate of change of acceleration, which is then converted into the rate of change of acceleration exponent.

7. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 1, characterized in that, The low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index are input into a pre-trained multilayer perceptron regression model to obtain a comprehensive writing quality score, including: A multilayer perceptron regression model is constructed, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer corresponds to the low-frequency stability index, the mid-frequency fatigue index, the high-frequency control precision index, the pen stroke force index, the speed stability index, and the acceleration change rate index. The hidden layer uses a modified linear unit activation function, and the output layer outputs a comprehensive score of writing quality. The low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index of the current writing event are input into the pre-trained multilayer perceptron regression model to calculate the comprehensive writing quality score.

8. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 7, characterized in that, The pre-training steps for a multilayer perceptron regression model include: Collect historical writing quality samples and corresponding expert annotation scores, and use the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index of the historical writing quality samples as input features, and use the expert annotation scores as the target output. Using mean squared error as the loss function, the network weights are iteratively optimized through backpropagation until the loss function converges, resulting in a pre-trained multilayer perceptron regression model.

9. The pen-holding posture tremor spectrum analysis method based on intelligent wearable devices according to claim 8, characterized in that, Also includes: A baseline model of the micro-vibration spectrum of personal writing was established based on the comprehensive writing quality score and the mid-frequency fatigue index. By tracking the monotonically increasing trend of the mid-frequency fatigue index and the decaying trend of the high-frequency control precision index during the current writing session through a sliding time window, the deviation of the mid-frequency fatigue index from the personal writing micro-vibration spectrum baseline model is calculated. When the deviation exceeds the safety boundary threshold, a fine motor fatigue warning is triggered. The period from the triggering of the fine motor fatigue warning to the restart of the writing session is taken as the rest interval. After the rest interval, the low-frequency stability index and the high-frequency control accuracy index are re-collected, and their recovery rate relative to the personal writing micro-vibration spectrum baseline model is calculated to assess the level of fine motor ability recovery.

10. A pen-holding posture tremor spectrum analysis system based on a smart wearable device, characterized in that, include: The data acquisition unit is used to deploy a first device on the dominant wrist as the first wrist, and deploy a second device on the auxiliary wrist as the second wrist, establish dual-wrist sampling clock alignment, and synchronously acquire the three-axis acceleration sequence and three-axis gyroscope sequence of the first wrist and the second wrist. The attitude calculation unit is used to calculate the real-time attitude angles and construct the projection matrix based on the three-axis gyroscope sequence and three-axis acceleration sequence of the first wrist. The pseudo-tremor separation unit is used to perform cross-power spectral density analysis on the triaxial acceleration sequences of the first and second wrists to extract the real physiological micro-vibration components and the pseudo-tremor components of the writing action. The posture compensation unit is used to project the real physiological micro-vibration components and the pseudo-tremor components of writing action onto the gravity coordinate system through a projection matrix to obtain the real micro-vibration spectrum and the pseudo-tremor spectrum. The spectrum decomposition unit is used to perform windowed fast Fourier transform on the real micro-vibration spectrum, calculate the energy integrals of the low-frequency, mid-frequency, and high-frequency bands, and obtain the low-frequency stability index, mid-frequency fatigue index, and high-frequency control accuracy index. The energy recovery unit is used to extract the main frequency and analyze the energy concentration of the pseudo-tremor spectrum to obtain the pen stroke strength index, velocity stability index, and acceleration change rate index. The coupled evaluation unit is used to input the low-frequency stability index, mid-frequency fatigue index, high-frequency control precision index, pen stroke force index, speed stability index, and acceleration change rate index into the pre-trained multilayer perceptron regression model to obtain a comprehensive score of writing quality. The dynamic monitoring unit is used to summarize the comprehensive score of writing quality and obtain the results of the pen-holding posture tremor spectrum analysis.

Images