Cloud platform-based dyskinesia monitoring system

Abstract

The invention discloses a dyskinesia monitoring system based on a cloud platform, and relates to the field of dyskinesia detection.The dyskinesia monitoring system comprises the steps that wearable equipment and a sensor are used for collecting motion state data and physiological parameter data of a user; calculating an uploading priority weight based on an adaptive entropy model, calculating an uploading period by using a dynamic adjustment algorithm, and adjusting a data uploading frequency and a data uploading priority; preprocessing the motion state data and the physiological parameter data through edge calculation; compressing the exercise function data, and uploading the exercise function data to a cloud end by using a wireless network according to the data uploading frequency and the data uploading priority; training and predicting through a convolutional neural network to generate a personalized dyskinesia early prediction result; personalized exercise rehabilitation suggestions are generated and fed back to the user through the user equipment side. Personalized prediction is carried out through the convolutional neural network, the potential risk of dyskinesia is found in advance, and early warning is provided for a user.

Description

Technical Field

[0001] The present invention relates to the field of movement disorder detection, and particularly to a movement dysfunction monitoring system based on a cloud platform. Background Art

[0002] With the continuous development of wearable devices and sensor technologies, health monitoring systems based on cloud platforms have been widely applied in fields such as healthcare, sports, and elderly care. These systems achieve continuous monitoring of users' health conditions by collecting physiological parameters and movement state data of the human body in real time. Especially in the early diagnosis and intervention of movement dysfunction, wearable devices provide valuable information for disease warning by monitoring physiological data in real time, such as heart rate, acceleration, electromyogram, etc. For example, functions such as heart rate monitoring and gait analysis have been verified in various applications, helping doctors and patients identify potential risks of movement disorders in a timely manner. In addition, the introduction of edge computing has improved data processing efficiency and real-time performance. By performing data preprocessing and analysis at the device end, it greatly reduces the dependence on cloud computing, improves the system response speed, and saves bandwidth.

[0003] However, despite certain progress in the prior art, there are still many challenges and deficiencies. First, in terms of the scheduling of data uploading and computing tasks, existing movement function monitoring systems often cannot dynamically adjust according to the actual power and network conditions of users. As a result, when the power is low or the network environment is poor, the monitoring function of the system is limited, affecting the integrity and real-time performance of the data. Second, existing movement function analysis mostly relies on fixed sensor data, and the collection and processing of these data often lack sufficient intelligence and are difficult to perform personalized analysis according to the individual differences of users. For example, the prediction analysis methods in most current monitoring systems are too single and cannot make full use of complex movement state data for accurate early diagnosis. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a movement dysfunction monitoring system based on a cloud platform to solve the problem of difficult personalized analysis according to the individual differences of users.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a motion dysfunction monitoring system based on a cloud platform, which includes a data acquisition module that uses wearable devices and sensors to collect the user's motion state data and physiological parameter data; a cloud-edge collaboration module that calculates the upload priority weight based on an adaptive entropy model, calculates the upload period using a dynamic adjustment algorithm, and adjusts the data upload frequency and data upload priority according to the battery power and network environment of the user device; an edge computing module that preprocesses the motion state data and physiological parameter data on the user device through edge computing and integrates them into motion function data; a data upload module that compresses the motion function data and uploads the motion function data to the cloud using a wireless network according to the data upload frequency and data upload priority and stores it in a cloud database; a prediction analysis module that trains and predicts based on the motion function data in the cloud database through a convolutional neural network to generate an early prediction result of personalized motion dysfunction; and a real-time feedback module that generates a personalized motion rehabilitation recommendation based on the early prediction result of personalized motion dysfunction and feeds it back to the user through the user device.

[0008] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: the wearable devices include smart watches, bracelets, smart insoles, and smart clothing, the sensors include accelerometers, gyroscopes, pressure sensors, electromyography sensors, heart rate sensors, temperature sensors, blood oxygen sensors, and blood pressure sensors, the motion state data includes acceleration changes, angle changes, pressure distribution, electromyography signals, and heart rate change data during exercise, and the physiological parameter data includes body temperature, blood oxygen saturation, and blood pressure.

[0009] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: the specific steps of using the wearable devices and sensors to collect the user's motion state data and physiological parameter data are as follows:

[0010] Set the sampling frequency and acquisition period of the wearable devices and sensors;

[0011] According to the sampling frequency and acquisition period, convert the user's motion physical signals into electrical signals through a sensor integrated circuit;

[0012] Through an analog-to-digital converter, convert the electrical signals into digital signals to obtain the user's motion state data and physiological parameter data.

[0013] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: the specific steps of calculating the upload priority weight based on the adaptive entropy model are as follows:

[0014] Regularly obtain the remaining battery power, current network latency, and bandwidth, and store them in the local cache;

[0015] An adaptive entropy model is used to calculate the priority and frequency of the uploaded data,

[0016] and the upload priority is comprehensively calculated.

[0017] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: the dynamic adjustment algorithm is used to calculate the upload period, and the specific steps are as follows,

[0018] Obtain the basic power upload period;

[0019] Based on the basic power upload period and the remaining battery power currently, a dynamic adjustment algorithm is dynamically constructed;

[0020] Adjust the upload period according to the dynamic adjustment algorithm.

[0021] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: according to the power and network environment of the user device end, the data upload frequency and the data upload priority are adjusted, and the specific steps are as follows,

[0022] Obtain the remaining battery power from the local cache and set the power threshold;

[0023] Adjust the data upload frequency according to the power threshold and in combination with the dynamic adjustment algorithm;

[0024] Detect the network environment data of the user device end and analyze the current network environment situation;

[0025] Further adjust the data upload frequency and the data upload priority according to the network environment situation and in combination with the upload priority weight.

[0026] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: through edge computing, the motion state data and the physiological parameter data are preprocessed at the user device end and integrated into motion function data, and the specific steps are as follows,

[0027] Remove the sensor noise of the acceleration change and the angle change through a low-pass filter;

[0028] Remove the baseline drift and noise in the heart rate change data through a high-pass filter and retain the heartbeat fluctuation part;

[0029] Use a waveform filter to remove the external electromagnetic interference of the electromyogram signal;

[0030] Perform data standardization on the motion state data and the physiological parameter data;

[0031] Use the linear interpolation method to fill in the missing values of the motion state data and the physiological parameter data;

[0032] Perform time windowing on the motion state data and physiological parameter data, and integrate them into motion function data in JSON format.

[0033] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform described in the present invention, wherein: perform data compression on the motion function data, and upload the motion function data to the cloud using a wireless network according to the data upload frequency and data upload priority, and store it in the cloud database. The specific steps are as follows:

[0034] Use the LZW lossless compression algorithm to perform data compression on the motion function data;

[0035] Use the MQTT protocol to configure the data chunk upload, breakpoint resumption, and data retransmission parameters;

[0036] According to the data upload frequency and data upload priority, upload the compressed motion function data to the cloud server using a wireless network;

[0037] On the cloud server, receive the uploaded compressed motion function data, decompress it, extract the key values of the motion function data from the JSON, and store them in the database.

[0038] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform described in the present invention, wherein: based on the motion function data in the cloud database, perform training and prediction through a convolutional neural network to generate an early prediction result of personalized motion dysfunction. The specific steps are as follows:

[0039] Obtain historical motion function data from the database and perform timestamp alignment;

[0040] Extract time-frequency features from the motion function data, normalize them to the same range, and obtain a time-frequency feature vector set;

[0041] Based on the time-frequency feature vector set, construct the input layer, adaptive convolutional layer, and adaptive pooling layer of the convolutional neural network;

[0042] Introduce a gating mechanism to control the dynamic change of weights;

[0043] Adopt adaptive pooling, and dynamically adjust the pooling window according to the size of the feature map and data changes;

[0044] Introduce a self-attention mechanism;

[0045] Adopt a weighted cross-entropy loss function to train the convolutional neural network model;

[0046] Introduce a dynamic learning rate adjustment strategy to accelerate the convergence of the convolutional neural network model;

[0047] Use the trained convolutional neural network to predict the newly collected motion function data and generate early prediction results of personalized motion dysfunction.

[0048] As a preferred solution of the motion dysfunction monitoring system based on the cloud platform according to the present invention, wherein: according to the early prediction results of personalized motion dysfunction, generate personalized motion rehabilitation suggestions and feedback them to the user through the user device side. The specific steps are as follows:

[0049] Based on the early prediction results of personalized motion dysfunction, identify the areas that require rehabilitation intervention and set the motion rehabilitation goals for the user.

[0050] According to the user's motion rehabilitation goals, select the motion type, determine the training intensity, customize the training time and frequency, provide action guidance and precautions, and generate personalized motion rehabilitation suggestions.

[0051] Through the user device side, feedback the personalized motion rehabilitation suggestions to the user.

[0052] The beneficial effects of the present invention are as follows: The data acquisition module uses wearable devices and sensors to collect the user's motion state and physiological parameter data in real time, providing an accurate data basis for subsequent analysis. The cloud-edge collaboration module dynamically adjusts the data upload frequency and priority according to the battery power and network environment of the user device, ensuring the stable operation of the system under various conditions. The edge computing module preprocesses the data on the user device side, reducing the computing pressure on the cloud, improving the response speed, and monitoring the user's health status in real time. The data upload module uses data compression and an efficient transmission protocol to ensure the accurate transmission of data in an unstable network environment. Through the prediction analysis module, personalized prediction is performed based on cloud big data and convolutional neural networks, and potential risks of motion dysfunction are discovered in advance, providing early warnings for users. The real-time feedback module generates personalized motion rehabilitation suggestions for users according to the prediction results and feedbacks them to the users through the device side, helping users implement targeted rehabilitation plans. Description of the Drawings

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is a module diagram of the motion dysfunction monitoring system based on the cloud platform in Embodiment 1.

[0055] Figure 2 It is a functional schematic diagram of the cloud-edge collaboration module in Embodiment 1. Detailed implementation manners

[0056] To make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific implementation manners of the present invention will be given with reference to the accompanying drawings of the specification.

[0057] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0058] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0059] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a motion dysfunction monitoring system based on a cloud platform, including the following steps:

[0060] A data acquisition module, using wearable devices and sensors, acquires the motion state data and physiological parameter data of the user.

[0061] Specifically, it includes the following steps:

[0062] Set the sampling frequency and acquisition period of the wearable devices and sensors.

[0063] The wearable devices include smart watches, bracelets, smart insoles and smart clothing, and the sensors include accelerometers, gyroscopes, pressure sensors, electromyography sensors, heart rate sensors, temperature sensors, blood oxygen sensors and blood pressure sensors.

[0064] Specifically, the accelerometer sampling frequency is set to 50 - 100 Hz, and the acquisition period is set to 10 - 20 ms. A higher sampling frequency helps capture rapid motion changes; the gyroscope sampling frequency is set to 50 - 100 Hz, and the acquisition period is set to 10 - 20 ms, which is used for monitoring posture and rotation angles to ensure a high enough sampling rate to capture subtle angle changes; the pressure sensor sampling frequency is set to 20 - 50 Hz, and the acquisition period is set to 20 - 50 ms, which is used for monitoring pressure distribution. The sampling frequency does not need to be too high, mainly to capture gait changes; the electromyography sensor sampling frequency is set to 200 - 1000 Hz, and the acquisition period is set to 1 - 5 ms, which is used for monitoring muscle activity. A higher sampling frequency is used to capture rapid electromyogram signal fluctuations; the heart rate sensor sampling frequency is set to 1 - 2 Hz, and the acquisition period is set to 500 - 1000 ms. Since the heart rate signal changes slowly, a lower sampling frequency is sufficient; the temperature sensor sampling frequency is set to 1 - 2 Hz, and the acquisition period is set to 500 - 1000 ms. Since the temperature changes slowly, the sampling frequency is relatively low; the blood oxygen sensor sampling frequency is set to 1 - 2 Hz, and the acquisition period is set to 500 - 1000 ms. Since the change in blood oxygen saturation is slow, low-frequency sampling can meet the requirements; the blood pressure sensor sampling frequency is set to 1 - 2 Hz, and the acquisition period is set to 500 - 1000 ms. Since the change in blood pressure is relatively slow, a lower sampling frequency is sufficient.

[0065] According to the sampling frequency and acquisition period, the user's motion physical signals are converted into electrical signals through a sensor integrated circuit.

[0066] Specifically, the accelerometer output range is from ±2g to ±16g, and the gain is 1x - 16x; the gyroscope output range is from ±250° / s to ±2000° / s, and the gain is 1x - 2x; the pressure sensor output range is 0 - 100 kPa, and the gain is 1x; the electromyography sensor output range is 1 μV - 10 mV, and the gain is 100x - 1000x; the heart rate sensor output range is 0 - 100% (heart rate is extracted through photoplethysmography (PPG)), and the gain is 1x - 10x; the temperature sensor output range is -40°C to 100°C, and the gain is 1x; the blood oxygen sensor output range is 0 - 100% (blood oxygen saturation range), and the gain is 1x; the blood pressure sensor output range is 0 - 300 mmHg, and the gain is 1x - 10x.

[0067] Through an analog-to-digital converter, the electrical signals are converted into digital signals to obtain the user's motion state data and physiological parameter data.

[0068] The motion state data includes acceleration changes, angle changes, pressure distribution, electromyogram signals, and heart rate change data during motion.

[0069] Physiological parameter data includes body temperature, blood oxygen saturation, and blood pressure.

[0070] It should be understood that the above information has been obtained with the user's consent and is used for legal purposes.

[0071] Specifically, the resolution of the analog-to-digital converter (ADC) needs to be 16 bits to provide high precision; the sampling rate and input range are set according to the sampling frequency and output range of the sensor.

[0072] Preferably, by precisely setting the sampling frequencies and acquisition periods of different sensors and using a high-precision analog-to-digital converter (ADC) for data conversion, it is possible to efficiently collect the user's motion state data and physiological parameter data. High-frequency sampling sensors (such as accelerometers, gyroscopes, and electromyography sensors) ensure that subtle changes in the motion state can be accurately captured, while low-frequency sampling sensors (such as heart rate, temperature, blood oxygen, and blood pressure sensors) optimize power consumption and meet the measurement accuracy requirements. In addition, flexible sampling parameter settings can provide the most suitable data acquisition scheme according to different sensor characteristics and application scenarios, thereby improving efficiency and stability while ensuring accuracy. The comprehensive acquisition of multi-dimensional data not only provides comprehensive support for the monitoring of movement function disorders but also provides a reliable data basis for subsequent personalized rehabilitation recommendations, enhancing the user experience and ensuring privacy protection and compliance.

[0073] The cloud-edge collaboration module adjusts the data upload frequency and data upload priority according to the battery power and network environment of the user device.

[0074] Specifically, it includes the following steps:

[0075] Regularly obtain the remaining battery power, current network latency, and bandwidth, and store them in the local cache.

[0076] Use an adaptive entropy model to calculate the priority and frequency of uploading data. The expression for the entropy value of the power state is:

[0077]

[0078] Among them, H1(t) is the entropy value of the power state, m is the number of power state categories, r is the index of the power state category, and q r (t) is the probability distribution of the rth power state at time t.

[0079] The expression for the network entropy value is:

[0080]

[0081] Among them, H2(t) is the network entropy value, o is the number of network state categories, b is the index of the network state category, and q bP(b)(t) is the probability distribution of the b-th network state at time t.

[0082] The upload priority is comprehensively calculated, and the expression is:

[0083]

[0084] where W(t) is the priority weight of the uploaded data, β1 is the weight of the battery power on the upload priority, β2 is the weight of the network on the upload priority, and δ is the normalization coefficient.

[0085] The upload period is dynamically adjusted according to the remaining battery power of the current battery, and the dynamic adjustment algorithm expression is:

[0086]

[0087] where F(t) is the upload period, F0 is the base power upload period, E(t) is the current power, and γ is the power decay factor.

[0088] It should be noted that the base power upload period is the preset upload period before dynamically adjusting the upload period according to the remaining battery power of the current battery, that is, the preset value when the power is higher than the power threshold. The base power upload period is set according to the normal operation requirements of the device, that is, the ideal upload frequency when the battery power is sufficient. Considering various factors such as the usage scenario of the device, the battery life requirement, and the data upload requirement, it is flexibly adjusted according to different working environments and usage scenarios using the developer's experience.

[0089] Obtain the remaining battery power from the local cache and set the power threshold;

[0090] Specifically, the user device terminal detects the remaining battery power in real time through the built-in battery management unit and provides the power percentage, and transmits it to the processing unit (such as MCU or SoC). When the remaining battery power is lower than 20%, it enters the low power consumption mode.

[0091] Adjust the data upload frequency according to the power threshold and in combination with the dynamic adjustment algorithm;

[0092] Specifically, when the remaining battery power is higher than 20%, the device uploads data at the original frequency; when the remaining battery power is lower than 20%, the upload frequency is corrected according to the dynamic adjustment algorithm.

[0093] Detect the network environment data of the user device terminal and analyze the current network environment situation.

[0094] Specifically, the network connection status is monitored in real time through the wireless communication unit, and information such as signal strength, delay, and packet loss rate is obtained. When the Wi-Fi signal is weak or there are high delay and high packet loss rate situations, it is judged that the network state is poor.

[0095] Further adjust the data upload frequency and data upload priority according to the network environment and in combination with the upload priority weight.

[0096] Specifically, when the network status is poor, further reduce the data upload frequency, and give priority to uploading key real-time motion information such as heart rate, acceleration change, and angle change. Non-critical data (such as low-priority background data, non-real-time motion information, temperature, etc.) will be uploaded later.

[0097] Furthermore, when the battery power is below 20% and the network status is poor, suspend data upload and perform local caching, and resume upload when the battery power is restored or the network condition improves.

[0098] Preferably, by real-time monitoring the remaining battery power and network environment of the user device, intelligently adjust the data upload frequency and priority, effectively extend the device usage time and improve data transmission efficiency. When the battery power is below 20%, automatically reduce the upload frequency or suspend the upload to reduce the consumption of the remaining battery power; when the network condition is poor, give priority to uploading key real-time motion data (such as heart rate and acceleration change), and delay uploading low-priority data to ensure the stable transmission of important data. It is also double-optimized according to the remaining battery power and network status to avoid unnecessary energy and bandwidth waste, while improving the adaptability and robustness of the module, ensuring stable operation in different usage environments, and enhancing the user experience and long-term availability of the device.

[0099] The edge computing module preprocesses the motion state data and physiological parameter data at the user device end through edge computing and integrates them into motion function data.

[0100] Specifically, it includes the following steps:

[0101] Remove the sensor noise of the acceleration change and angle change through a low-pass filter.

[0102] Specifically, use a low-pass filter to denoise the sensor signals of the acceleration change and angle change. Set the cut-off frequency of the low-pass filter to 50 - 100 Hz. The low-pass filter can effectively filter out high-frequency noise (such as environmental interference, sensor noise, etc.), and only retain the low-frequency effective motion data. Use typical filtering algorithms such as Butterworth or Chebyshev filters to ensure that the signal is smooth and distortion-free.

[0103] Remove the baseline drift and noise in the heart rate change data through a high-pass filter, and retain the heartbeat fluctuation part.

[0104] Specifically, a high-pass filter is used to remove baseline drift and low-frequency noise in the heart rate change data. To retain the heartbeat fluctuation part, the cut-off frequency of the high-pass filter is set to 0.5 Hz - 1 Hz, which can remove the low-frequency drift in the heart rate signal and retain the actual heartbeat changes. The filter uses an FIR (finite impulse response) or IIR (infinite impulse response) filter.

[0105] A waveform filter is used to remove external electromagnetic interference from the electromyogram signal.

[0106] Specifically, to remove external electromagnetic interference from the electromyogram signal (EMG), a waveform filter or a band-pass filter is used, and the frequency band range of the filter is set to 20 - 500 Hz. It can remove environmental electromagnetic interference, ensure the clarity and accuracy of the electromyogram signal data, and thus better reflect muscle activities.

[0107] Data standardization is performed on the motion state data and physiological parameter data.

[0108] It should be understood that the z-score standardization method is used to standardize all motion state data and physiological parameter data to ensure that the data is within the same dimension range, which is convenient for subsequent processing and analysis.

[0109] The linear interpolation method is used to fill in the missing values in the motion state data and physiological parameter data.

[0110] Specifically, for each missing value, a linear interpolation is calculated through the two adjacent valid data points before and after it to obtain an estimated value. This can avoid analysis errors caused by data missing and ensure the continuity and integrity of the data.

[0111] The motion state data and physiological parameter data are windowed in time and integrated into motion function data in JSON format.

[0112] Specifically, the collected motion state data and physiological parameter data are windowed according to the time stamp. The time window length is defined as 1 - 2 seconds, and the sliding window method (moving 500 ms each time) is used to segment the motion state data and physiological parameter data. The motion state data and physiological parameter data in each time window will be integrated into JSON format, which includes statistical quantities such as the average value, standard deviation, and maximum value of all data in the window. This time windowing process can effectively capture the time-varying characteristics of the motion state and physiological data and ensure the accuracy of subsequent data analysis and modeling.

[0113] Preferably, by preprocessing the motion state data and physiological parameter data on the user device side, the data quality and analysis efficiency are effectively improved. Noise is removed through low-pass, high-pass, and waveform filters to ensure smooth and interference-free data. Z-score normalization is used to make the data from different sensors consistent. Linear interpolation is used to fill in missing values to ensure data continuity. At the same time, time windowing is used to capture the time-varying characteristics of the data. All preprocessing operations are completed on the device side, reducing network bandwidth consumption and transmission delay, and ensuring real-time feedback and efficient data transmission.

[0114] The data upload module compresses the motion function data and uploads the motion function data to the cloud using a wireless network according to the data upload frequency and data upload priority, and stores it in the cloud database.

[0115] Specifically, it includes the following steps:

[0116] Use the LZW lossless compression algorithm to compress the motion function data.

[0117] It should be understood that the motion function data usually contains high-frequency sampling data from multiple sensors, so the data volume is huge, and direct upload will consume more bandwidth. Using the LZW (Lempel-Ziv-Welch) lossless compression algorithm to compress the motion function data can effectively reduce the data volume. Lossless compression ensures the integrity of the data and does not lose any important motion or physiological information.

[0118] Use the MQTT protocol to configure data chunk upload, resume interrupted upload, and data retransmission parameters.

[0119] Specifically, MQTT (Message Queuing Telemetry Transport) is a lightweight, low-bandwidth message transmission protocol suitable for data exchange between Internet of Things devices. Based on the MQTT protocol, a resume interrupted upload function is set, that is, when the upload process is interrupted due to network interruption or device restart, it can continue to upload from the breakpoint. By configuring the data retransmission parameters, if packet loss or transmission failure occurs during upload, the motion function data that has not been successfully uploaded is automatically resent.

[0120] According to the data upload frequency and data upload priority, use a wireless network to upload the compressed motion function data to the cloud server.

[0121] Specifically, higher priorities are set for key real-time motion data (such as heart rate, acceleration, etc.), and lower priorities are set for non-real-time data (such as body temperature, blood oxygen, etc.). According to the upload frequency and priority, the upload time point and the type of data to be uploaded are automatically determined according to the network condition, ensuring that key data is uploaded in a timely manner, while other low-priority data is uploaded when the bandwidth is sufficient.

[0122] The cloud server receives the uploaded compressed motion function data, decompresses it, extracts the key values of the motion function data from the JSON, and stores them in the database.

[0123] Specifically, after the cloud server receives the compressed motion function data, it first uses the LZW decompression algorithm to restore the motion function data. The decompressed motion function data is stored in JSON format, which is convenient for subsequent parsing and data processing. Each key value (such as heart rate, acceleration, blood pressure, etc.) in the JSON data is extracted and stored in the MySQL cloud database for further analysis and processing.

[0124] Preferably, the LZW lossless compression algorithm is used to effectively reduce the volume of the motion function data, save bandwidth and storage space, and ensure data integrity. The MQTT protocol is used for data transmission, which has advantages such as low bandwidth consumption and strong real-time performance. The reliability of the upload is improved through data chunking upload, breakpoint resumption, and retransmission functions. According to the data upload frequency and priority, the upload strategy is dynamically adjusted to ensure the real-time transmission of key data such as heart rate and acceleration, while optimizing the use of network resources. The cloud server receives and decompresses the uploaded data, stores it in JSON format, which is convenient for subsequent analysis and processing, improves the efficiency of data upload and storage, and ensures the accuracy and integrity of the data.

[0125] The prediction analysis module, based on the motion function data in the cloud database, is trained and predicted through a convolutional neural network to generate an early prediction result of personalized motion function disorders.

[0126] Specifically, it includes the following steps:

[0127] Obtain historical motion function data from the database and align the timestamps.

[0128] Specifically, use SQL statements to extract historical motion function data from the MySQL cloud database to ensure that each piece of data contains a timestamp. Align the timestamps of the historical motion function data to ensure that the time order of all historical motion function data is consistent, avoiding errors during training due to time confusion.

[0129] Extract time-frequency features from the motion function data, normalize them to the same range, and obtain a time-frequency feature vector set.

[0130] Specifically, extract time-domain features: calculate statistics such as mean, variance, maximum value, minimum value, peak-to-peak value, etc.

[0131] Extract frequency-domain features: calculate spectral energy, main frequency, etc. through the fast Fourier transform (FFT).

[0132] Extract spatio-temporal features: Combine accelerometer and gyroscope data to calculate gait parameters (such as step length, step frequency) or other motion pattern features.

[0133] Use the Min-Max normalization method to map all feature values to the range of [0, 1] to avoid the influence of dimensional differences between features on model training.

[0134] Arrange all the extracted features in chronological order to form a time-frequency feature vector set, where each vector represents the motion state at a certain moment.

[0135] Based on the time-frequency feature vector set, construct the input layer, adaptive convolutional layer, and adaptive pooling layer of the convolutional neural network. The expression of the adaptive convolutional kernel is:

[0136]

[0137] Among them, f out (t) is the feature map output after the convolution operation, n is the size of the convolutional kernel, i is the index coefficient of the convolutional kernel size, w i (t) is the dynamic convolutional kernel weight at time t, t is the time index, x(t - i) is the input data value at time t - i, b i (t) is the convolutional kernel bias term at time t.

[0138] Introduce a gating mechanism to control the dynamic change of weights. The expression is:

[0139] w i (t) = σ(W g ·[x(t - i), h(t - 1)])·W c ;

[0140] Among them, σ is the sigmoid function, W g is the gating weight matrix, h(t - 1) is the hidden state at the previous moment, W c is the basic weight of the static convolutional kernel.

[0141] Preferably, the fixed convolutional kernel of the traditional CNN is difficult to capture the time-varying features of motor dysfunction. The improved adaptive convolution can enhance the modeling ability for non-stationary signals such as gait mutations through dynamic weight adjustment.

[0142] Adopt adaptive pooling, and dynamically adjust the pooling window according to the size of the feature map and data changes. The expression is:

[0143]

[0144] Among them, p out (t) is the output after pooling, max is the identifier for taking the maximum value, [t, t + k] is the pooling window range, and k is the window size.

[0145] Introduce the self-attention mechanism, and the expression is:

[0146]

[0147] where a t is the value of the attention weight vector at time t, softmax is a non-linear activation function, Q t is the query vector representing the feature at the current time, K t is the key vector representing the feature at the historical time, T is the vector transpose operation, d is the dimension of the key vector, and V t is the value vector representing the weighted value of the current feature.

[0148] Adopt the weighted cross-entropy loss function to train the convolutional neural network model, and the expression is:

[0149]

[0150] where L is the weighted cross-entropy loss function, N is the number of samples, j is the sample number index, y j is the true label, is the predicted output of the convolutional neural network model, and u j is the importance weight of sample j.

[0151] Introduce the dynamic learning rate adjustment strategy to accelerate the convergence of the convolutional neural network model. The learning rate adjustment expression is:

[0152]

[0153] where η s+1 is the adjusted learning rate, s is the number of training steps, η s is the current learning rate, and α is the decay coefficient.

[0154] It should be noted that the time-frequency feature vector set is divided into a training set and a validation set. The convolutional neural network model is trained using the training set, and the performance of the convolutional neural network model is monitored using the validation set.

[0155] Use the trained convolutional neural network to predict the newly collected motion function data and generate early prediction results of personalized motion dysfunction.

[0156] Specifically, convert the prediction results into an interpretable form, such as probability values or classification labels, and generate a personalized report in combination with the motion function data.

[0157] Preferably, data consistency is ensured by aligning timestamps to avoid errors caused by time confusion; dynamic changes in motor function data are fully captured by extracting time domain, frequency domain, and spatiotemporal features, while using Min-Max normalization to eliminate dimensional differences between features and promote the convergence of the convolutional neural network model. The introduction of adaptive convolution and pooling mechanisms enhances the modeling capabilities of time-varying signals and gait mutations, and further improves the adaptability of the convolutional neural network model to complex motion data. The self-attention mechanism helps strengthen the modeling of long-term dependencies and enhances the dynamic attention to the importance of historical data. The weighted cross entropy loss function handles unbalanced data, and the dynamic learning rate adjustment accelerates the training of the convolutional neural network model, and provides accurate prediction results through personalized report generation.

[0158] The real-time feedback module generates personalized exercise rehabilitation suggestions based on the early prediction results of personalized motor dysfunction and feeds them back to the user through the user device.

[0159] Specifically, the following steps are included:

[0160] Based on the personalized early prediction results of motor dysfunction, areas requiring rehabilitation intervention are identified and the user's motor rehabilitation goals are set.

[0161] Specifically, based on the prediction results of motor dysfunction generated by the convolutional neural network, the user's motor function data is analyzed to identify key factors affecting motor function (such as gait instability, muscle fatigue, cardiopulmonary endurance, etc.). Rehabilitation goals are set in a targeted manner based on the user's health status, athletic ability, historical records, and personal needs. For example, if the user's gait is abnormal, the rehabilitation goal is to improve gait stability; if the prediction results show muscle fatigue, the goal is to increase muscle strength.

[0162] According to the user's sports rehabilitation goals, select the type of exercise and determine the training intensity, customize the training time and frequency, provide action guidance and precautions, and generate personalized sports rehabilitation suggestions.

[0163] Specifically, suitable exercise types are recommended according to the rehabilitation goals and the specific needs of the user. For example, for users with unstable gait, gait training, walking, running, etc. can be recommended; for users with muscle weakness, targeted strength training can be recommended. The intensity of the exercise is dynamically adjusted according to the user's physical condition and training goals. For example, in the initial stage, low-intensity exercises such as brisk walking or low-resistance strength training can be selected, and the intensity is gradually increased as the rehabilitation progresses. A training plan suitable for the user is customized to ensure that the training is neither excessive nor too light, considering factors such as the user's endurance and time arrangement. For example, low-intensity gait training three times a week for 20-30 minutes each time. Action guidance based on sports data analysis is provided, such as correct postures, paces, breathing methods, etc., to reduce the risk of sports injuries. For example, it is recommended that the user keep the knees slightly bent during gait training and avoid excessive exertion. The user can also be reminded to pay attention to diet, rest and adjust the training plan.

[0164] Through the user's device, personalized exercise rehabilitation suggestions are fed back to the user.

[0165] Specifically, through intelligent devices (such as smart bracelets, smart watches, mobile applications, etc.), customized exercise rehabilitation suggestions are transmitted to the user in real time. The device can help the user understand and execute the training plan through visual interfaces or voice prompts.

[0166] Preferably, by analyzing the user's movement function data and prediction results, areas that need intervention are identified, and personalized rehabilitation goals are set, such as improving gait stability or enhancing muscle strength. Appropriate exercise types, intensities, times and frequencies are selected according to the goals. At the same time, the training plan is dynamically adjusted to ensure that the training is neither excessive nor insufficient, and action guidance is provided to reduce the risk of sports injuries. Rehabilitation suggestions are transmitted in real time through intelligent devices (such as bracelets, mobile applications, etc.) to enhance the user's sense of participation and rehabilitation motivation, ensuring the scientific nature and safety during the rehabilitation process. In addition, the real-time feedback and health reminders on the device improve the user's rehabilitation effect and provide valuable data support for subsequent optimization, ensuring that the user receives comprehensive health management during the rehabilitation process.

[0167] In summary, in the present invention, the data acquisition module uses wearable devices and sensors to collect the user's motion state and physiological parameter data in real time, providing an accurate data basis for subsequent analysis. The cloud-edge collaboration module dynamically adjusts the data upload frequency and priority according to the battery power and network environment of the user device, ensuring the stable operation of the system under various conditions. The edge computing module preprocesses the data on the user device side, reducing the computing pressure on the cloud, improving the response speed, and monitoring the user's health status in real time. The data upload module uses data compression and an efficient transmission protocol to ensure the accurate transmission of data in an unstable network environment. Through the prediction analysis module, personalized prediction is carried out based on cloud big data and convolutional neural networks to discover potential risks of movement dysfunction in advance and provide early warnings for users. The real-time feedback module generates personalized exercise rehabilitation suggestions for users according to the prediction results and feeds them back to the users through the device side to help users implement targeted rehabilitation plans.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A cloud-based movement dysfunction monitoring system, characterized in that: include, The data acquisition module uses wearable devices and sensors to collect the user's motion status data and physiological parameter data; The cloud-edge collaboration module calculates the upload priority weight based on the adaptive entropy model, uses a dynamic adjustment algorithm to calculate the upload cycle, and adjusts the data upload frequency and data upload priority according to the power of the user device and the network environment; The edge computing module pre-processes the motion state data and physiological parameter data on the user device through edge computing and integrates them into motion function data; The data upload module compresses the motion function data, uploads the motion function data to the cloud using a wireless network according to the data upload frequency and data upload priority, and stores the data in a cloud database; The prediction analysis module generates personalized early prediction results of motor dysfunction based on the motor function data in the cloud database through training and prediction through convolutional neural networks; The real-time feedback module generates personalized exercise rehabilitation suggestions based on the early prediction results of personalized motor dysfunction and feeds them back to the user through the user device.

2. The cloud-based movement dysfunction monitoring system according to claim 1, characterized in that: The wearable devices include smart watches, bracelets, smart insoles and smart clothing. The sensors include accelerometers, gyroscopes, pressure sensors, electromyographic sensors, heart rate sensors, temperature sensors, blood oxygen sensors and blood pressure sensors. The motion state data include acceleration changes, angle changes, pressure distribution, electromyographic signals and heart rate change data during motion. The physiological parameter data include body temperature, blood oxygen saturation and blood pressure.

3. The cloud-based movement dysfunction monitoring system according to claim 2, characterized in that: The wearable device and the sensor are used to collect the user's motion state data and physiological parameter data. The specific steps are: Set the sampling frequency and collection period of wearable devices and sensors; According to the sampling frequency and acquisition period, the user's motion physical signal is converted into an electrical signal through the sensor integrated circuit; Through the digital-to-analog converter, the electrical signal is converted into a digital signal to obtain the user's motion status data and physiological parameter data.

4. The cloud-based movement dysfunction monitoring system according to claim 3, characterized in that: The specific steps of calculating the upload priority weight based on the adaptive entropy model are as follows: Regularly obtain the remaining battery power and current network latency and bandwidth, and store them in the local cache; Adaptive entropy model is used to calculate the priority and frequency of uploaded data. Comprehensively calculate the upload priority.

5. The cloud-based movement dysfunction monitoring system according to claim 4, characterized in that: The specific steps of using the dynamic adjustment algorithm to calculate the upload period are as follows: Get the basic power upload cycle; Dynamically build a dynamic adjustment algorithm based on the basic power upload cycle and the current remaining battery power; The upload period is adjusted according to the dynamic adjustment algorithm.

6. The cloud-based movement dysfunction monitoring system according to claim 5, characterized in that: The specific steps of adjusting the data upload frequency and data upload priority according to the power of the user device and the network environment are as follows: Get the remaining battery power from the local cache and set the power threshold; Adjust the data upload frequency based on the power threshold and in combination with a dynamic adjustment algorithm; Detect the network environment data on the user's device and analyze the current network environment; The data upload frequency and data upload priority are further adjusted according to the network environment and in combination with the upload priority weight.

7. The cloud-based movement dysfunction monitoring system according to claim 6, characterized in that: The edge computing is used to pre-process the motion state data and physiological parameter data on the user device side and integrate them into motion function data. The specific steps are: The sensor noise of acceleration change and angle change is removed through low-pass filter; The baseline drift and noise in the heart rate variation data are removed through a high-pass filter, and the heartbeat fluctuation part is retained; Use waveform filters to remove external electromagnetic interference from EMG signals; Standardize the motion status data and physiological parameter data; Linear interpolation was used to fill missing values ​​in motion state data and physiological parameter data; The motion status data and physiological parameter data are time-windowed and integrated into motion function data in JSON format.

8. The cloud-based movement dysfunction monitoring system according to claim 7, characterized in that: The specific steps of compressing the motion function data, uploading the motion function data to the cloud using a wireless network according to the data upload frequency and data upload priority, and storing the data in a cloud database are as follows: Use LZW lossless compression algorithm to compress the motor function data; Use the MQTT protocol to configure data block upload, breakpoint resume and data retransmission parameters; Upload the compressed motion function data to the cloud server using a wireless network according to the data upload frequency and data upload priority; The cloud server receives the uploaded compressed motion function data, decompresses it, extracts the key value of the motion function data from JSON and stores it in the database.

9. The cloud-based movement dysfunction monitoring system according to claim 8, characterized in that: The motor function data in the cloud database is trained and predicted by a convolutional neural network to generate personalized early prediction results of motor dysfunction, and the specific steps are: Obtain historical motion function data from the database and align timestamps; Extract time-frequency features from the motor function data, normalize them to the same range, and obtain a time-frequency feature vector set; Based on the time-frequency feature vector set, the input layer, adaptive convolution layer and adaptive pooling layer of the convolutional neural network are constructed; Introduce a gating mechanism to control the dynamic changes of weights; Adaptive pooling is used to dynamically adjust the pooling window according to the size of the feature map and data changes; Introducing the self-attention mechanism; The weighted cross entropy loss function is used to train the convolutional neural network model; Introducing a dynamic learning rate adjustment strategy to accelerate the convergence of the convolutional neural network model; Use the trained convolutional neural network to predict the newly collected motor function data and generate personalized early prediction results of motor dysfunction.

10. The cloud-based movement dysfunction monitoring system according to claim 9, characterized in that: The method generates personalized exercise rehabilitation suggestions based on the personalized motor dysfunction early prediction results and feeds back to the user through the user device. The specific steps are: Based on the personalized early prediction results of motor dysfunction, the areas that require rehabilitation intervention are identified and the user's motor rehabilitation goals are set; According to the user's sports rehabilitation goals, select the type of exercise and determine the training intensity, customize the training time and frequency, provide action guidance and precautions, and generate personalized sports rehabilitation suggestions; Personalized exercise rehabilitation suggestions are fed back to users through their devices.

Images