A multi-electrophysiological-based fall early warning method and system

Abstract

The application discloses a fall early warning method and system based on multi-electrophysiology, which is applied to the technical field of gait protection, and the method comprises the following steps: performing feature extraction on multi-modal electrophysiological signals to obtain multi-modal electrophysiological features; performing feature extraction on electromyographic signals and electroencephalogram signals respectively to obtain first single-point electrophysiological features and second single-point electrophysiological features; evaluating muscle states according to the electromyographic signals to obtain a first evaluation result; evaluating brain function states according to the electroencephalogram signals to obtain a second evaluation result; calculating reaction times of the electromyographic signals and the electroencephalogram signals to obtain a third evaluation result; inputting acquired individual information and evaluation results into a fall risk evaluation model to output a risk coefficient; inputting the multi-modal electrophysiological features, the single-point electrophysiological features and the risk coefficient into a fall early warning model for training; and predicting multi-modal electrophysiological signals of a target user based on the trained fall early warning model to output a fall early warning result.

Description

Technical Field

[0001] This invention relates to the field of gait protection technology, and in particular to a fall warning method and system based on multiple electrophysiological methods. Background Technology

[0002] Age-related functional decline and lower limb dysfunction significantly increase the risk of falls in the elderly. Therefore, the development of fall prevention measures for the elderly and those with lower limb dysfunction has become an urgent clinical and community need.

[0003] Currently, fall protection devices mainly fall into three categories: walking aids, follow-up fall arrestors, and wearable fall protection airbags. However, regardless of the method, accurate and timely prediction is crucial for effective operation. Current fall detection methods include millimeter-wave detection, video detection, and audio detection. These methods are prone to misjudgment due to their limited information, and the lag in information delivery leads to untimely warnings. Furthermore, they are difficult to use on a person and cannot be easily integrated with fall protection devices.

[0004] To overcome these shortcomings, this application proposes a fall warning method and system based on multiple electrophysiology. Summary of the Invention

[0005] The purpose of this application is to provide a fall warning method and system based on multiple electrophysiology, which aims to solve the problems of insufficient timeliness and easy misjudgment in existing fall warning systems.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] In a first aspect, this application provides a fall warning method based on multiple electrophysiology, including:

[0008] Acquire multimodal electrophysiological signals, including electromyographic signals and electroencephalographic signals;

[0009] Feature extraction is performed on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; feature extraction is performed on the electromyographic signals and the electroencephalographic signals respectively to obtain first single-point electrophysiological features and second single-point electrophysiological features;

[0010] The muscle state is assessed based on the electromyographic signals to obtain a first assessment result; the brain function state is assessed based on the electroencephalogram (EEG) signals to obtain a second assessment result; the reaction time of the electromyographic signals and the EEG signals is calculated to obtain a third assessment result.

[0011] The acquired individual information, the first assessment result, the second assessment result, and the third assessment result are input into the fall risk assessment model, and the risk coefficient is output.

[0012] The multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficient are input into the fall warning model for training to obtain a trained fall warning model.

[0013] The multimodal electrophysiological signals of the target user are acquired, and the multimodal electrophysiological signals of the target user are predicted based on the trained fall warning model, and the fall warning result is output.

[0014] Secondly, this application proposes a fall warning system based on multi-electrophysiology, comprising:

[0015] Acquisition module: Acquires multimodal electrophysiological signals, including electromyography (EMG) signals and electroencephalography (EEG) signals;

[0016] Feature extraction module: performs feature extraction on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; performs feature extraction on the electromyographic signals and the electroencephalogram signals respectively to obtain first single-point electrophysiological features and second single-point electrophysiological features;

[0017] Assessment module: Assess muscle state based on electromyography (EMG) signals to obtain a first assessment result; assess brain function state based on electroencephalography (EEG) signals to obtain a second assessment result; calculate the reaction time of the EMG and EEG signals to obtain a third assessment result; input the acquired individual information, the first assessment result, the second assessment result, and the third assessment result into the fall risk assessment model, and output the risk coefficient;

[0018] Model training module: The multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficient are input into the fall warning model for training to obtain a trained fall warning model;

[0019] Output module: Acquires the multimodal electrophysiological signals of the target user, predicts the multimodal electrophysiological signals of the target user based on the trained fall warning model, and outputs the fall warning result.

[0020] Thirdly, this application provides a fall warning device, the fall warning device including a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing a fall warning method based on multiple electrophysiology; the processor is used to execute the program instructions stored in the memory to implement a fall warning method based on multiple electrophysiology.

[0021] Fourthly, this application provides a storage medium storing processor-executable program instructions for executing a fall warning method based on multiple electrophysiology.

[0022] This application provides a fall early warning method and system based on multi-electrophysiology, which has the following beneficial effects:

[0023] Fall warning is achieved by using multi-site electromyography (EMG) signals and 16-channel electroencephalography (EEG) signals triggered by human stress response. It captures physiological and motor characteristics prior to a fall from multiple angles and employs a multimodal electrophysiological sensor network with time synchronization to acquire multimodal electrophysiological signals at the same moment. The system utilizes both single-point and multimodal electrophysiological feature extraction, considering both irrelevant and relevant features, reducing the false alarm rate and improving the predictability of the warning. The trained fall warning model can accurately predict falls, providing earlier warnings and reaction time compared to traditional post-fall detection. Furthermore, it can be adjusted according to individual differences to be applicable to different user groups. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a fall warning method based on multiple electrophysiology according to Embodiment 1 of this application;

[0025] Figure 2 This is a schematic diagram of the structure of the convolutional neural network for feature extraction of multimodal electrophysiological signals according to Embodiment 1 of this application;

[0026] Figure 3 This is a schematic diagram of the deep neural network structure for feature extraction of electromyography signals according to Embodiment 1 of this application;

[0027] Figure 4 This is a schematic diagram of the structure of the convolutional neural network for feature extraction of electroencephalogram (EEG) signals according to Embodiment 1 of this application;

[0028] Figure 5 This is a schematic diagram of the recurrent neural network structure of Embodiment 1 of this application;

[0029] Figure 6 The target network model of Embodiment 1 of this application is composed of a deep neural network, a convolutional neural network, and a recurrent neural network;

[0030] Figure 7 This is a schematic diagram of the structure of a fall warning system based on multiple electrophysiology according to Embodiment 2 of this application;

[0031] Figure 8 This is a design block diagram of a fall warning system based on multiple electrophysiology, according to Embodiment 2 of this application.

[0032] Figure 9 This is a schematic diagram of the fall warning device structure according to Embodiment 3 of this application;

[0033] Figure 10 This is a schematic diagram of the storage medium structure of Embodiment 4 of this application. Detailed Implementation

[0034] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0035] The following analysis, based on relevant technologies, examines existing solutions.

[0036] Invention patent CN118731940A discloses a fall detection method and system based on millimeter-wave radar, relating to the field of millimeter-wave radar technology. It transmits linear frequency modulated (LFM) wave signals to a target human body via millimeter-wave radar and receives the electromagnetic wave signals reflected back from the target human body. It employs deep learning-based artificial intelligence technology to analyze the time-frequency characteristics of the LFM wave signal and the reflected electromagnetic wave signal, and then achieves intelligent recognition of the target human body's posture based on the differences in their characteristics. However, this patent's detection method, based on physical information such as millimeter waves and video, relies on environmental sensors, requiring pre-positioned sensor points in the environment for detection. Furthermore, it requires remote connection when linked with protective equipment, making real-time performance uncertain.

[0037] Chinese invention patent CN118609215A discloses a fall warning method and system based on deep learning, belonging to the field of fall detection and image recognition technology. It includes inputting the body parameters of the monitored target person, obtaining the actual centroid coordinates based on the body parameters, and then using a first radar fall estimation model to perform indoor fall detection and a second video stream iterative detection model to perform outdoor fall detection. The system continuously updates the training sample set to further update the training of the second video stream iterative detection model. However, this method can only be applied to environments with pre-positioned camera points, has low real-time performance, and cannot be integrated with safety equipment.

[0038] Invention patent CN118506807A discloses a fall detection method for the elderly based on deep learning and audio recognition. The method includes: S1, constructing a fall audio dataset; S2, performing data preprocessing to obtain a Mel spectrogram with a uniform format and containing audio time, frequency, and volume feature information; S3, constructing a fall audio recognition model; S4, defining evaluation metrics for model training; and S5, testing the model accuracy. However, detection methods based on physical information such as audio are greatly affected by noise and mostly rely on single pieces of information for detection, making them prone to misjudgment.

[0039] Therefore, this application proposes a fall warning method and system based on multiple electrophysiological methods. It is well known that when a person is about to fall, their muscles will react in an emergency, and they will experience tension. By accurately detecting the muscle's emergency response through electromyography (EMG) signals and judging the tension through electroencephalography (EEG) signals, a timely warning can be issued, allowing protective equipment to take protective actions.

[0040] 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 a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0041] Example 1

[0042] Please see Figure 1 This is a flowchart illustrating a fall warning method based on multi-electrophysiology according to Embodiment 1 of this application; the steps include:

[0043] S1: Acquire multimodal electrophysiological signals, including electromyography (EMG) signals and electroencephalography (EEG) signals.

[0044] In this embodiment, the acquisition locations of the electromyography (EMG) signals include, but are not limited to, the following: lower limb muscles, upper limb muscles, gluteal muscles, core muscles, back muscles, and neck muscles; the acquisition of the electroencephalogram (EEG) signals uses 16 channels. The EMG acquisitions from different locations and the 16-channel EEG acquisition are networked using Bluetooth to obtain a multimodal electrophysiological human sensor network. The various electrophysiological acquisition points within the multimodal electrophysiological human sensor network have time synchronization capabilities, enabling the acquisition of multimodal electrophysiological signals at the same time. The specific implementation method is as follows:

[0045] First, Bluetooth networking was established for EMG and EEG acquisition, and their respective locations and acquisition types were determined. Each Bluetooth module has its own MAC address, which is then bound to the acquisition location and type of its respective electrophysiological signal. Nodes can be added or removed depending on the user's needs; after networking, nodes can communicate with each other. An implementation example is shown in the table below:

[0046]

[0047] To synchronize timestamps between nodes, a reference node is first selected, whose timestamp is designated as the reference timestamp Tref. Then, the synchronization is performed based on the reference timestamp and the timestamps of each node. Get the time offset of each node. ( (This refers to the number of nodes). In subsequent data collection, the data from each node is time-shifted accordingly. Specifically:

[0048] ,

[0049] ,

[0050] in For the first Data synchronized across all nodes; For the first The original data of each node; This is a synchronization function, achieved by shifting the sampling points.

[0051] The synchronized data undergoes preprocessing, primarily involving bandpass filtering to extract the frequency range of interest from the electrophysiological signals. For electromyography (EMG), the range is 20–500 Hz, and for electroencephalography (EEG), it is 0.5–100 Hz. Then, a notch filter is used to remove power frequency noise (50 Hz or 60 Hz).

[0052] S2: Perform feature extraction on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; perform feature extraction on the electromyographic signals and the electroencephalographic signals respectively to obtain the first single-point electrophysiological features and the second single-point electrophysiological features.

[0053] In this embodiment, feature extraction mainly includes two parts: single-point electrophysiological feature extraction and multimodal electrophysiological feature extraction. Single-point electrophysiological feature extraction can accurately extract different single-point electrophysiological features without considering other electrophysiological factors, while multimodal electrophysiological feature extraction can extract correlation features between electromyography signals at different locations and 16-channel electroencephalography signals.

[0054] Please see Figure 2 This is a schematic diagram of the convolutional neural network structure for feature extraction from multimodal electrophysiological signals according to Embodiment 1 of this application. Due to the large size of multimodal electrophysiological signal data, the network design is more complex. It mainly includes a multi-location electromyography convolutional pooling layer with a 4x4 kernel, a multi-channel electroencephalography convolutional pooling layer with a 4x4 kernel, a comprehensive convolutional pooling layer with a 3x3 kernel, and three fully connected layers with 64, 256, and 9 nodes respectively. The output is the coordination level of muscles and brain at eight locations and the current stability level.

[0055] The convolutional neural network is pre-trained using electromyography (EMG) signals from various locations and multi-channel electroencephalography (EEG) signals to output multimodal electrophysiological features, which are the degree of coordination between muscles and the brain at various locations and the current level of stability.

[0056] Please see Figure 3 This is a schematic diagram of the deep neural network structure for feature extraction of electromyography (EMG) signals according to Embodiment 1 of this application. Since EMG signals are not very complex, the network structure does not need to be too complex. A deep neural network is used to extract features from EMG signals. The deep neural network consists entirely of fully connected layers (FCL), divided into 4 layers, with 64, 256, 128, and 3 nodes in each layer, respectively. The outputs are the degree of muscle contraction, the degree of fatigue, and the degree of lactic acid accumulation, which yield the first single-point electrophysiological feature.

[0057] Please see Figure 4 This is a schematic diagram of the convolutional neural network structure for feature extraction of EEG signals according to Embodiment 1 of this application. Since EEG data is multi-channel data, a convolutional neural network is used for feature extraction. The convolutional neural network consists of four layers. The first two layers are convolutional pooling layers, which mainly reduce the dimensionality of the 16-channel EEG signal twice. The first convolutional pooling layer uses a 4x4 kernel, and the second convolutional pooling layer uses a 3x3 kernel. The last two layers are fully connected layers (FCL), with 128 and 3 nodes respectively, outputting the level of emotional tension, the level of emotional relaxation, and the current level of attention. This yields the second single-point electrophysiological feature.

[0058] S3: Assess muscle state based on the electromyographic signal to obtain a first assessment result; assess brain function state based on the electroencephalogram (EEG) signal to obtain a second assessment result; calculate the reaction time of the electromyographic signal and the EEG signal to obtain a third assessment result.

[0059] In this embodiment, since muscle state, brain function state, fall risk coefficient and type do not change mutally, periodic assessment is performed, that is, an assessment is performed after acquiring electromyography and electroencephalography signals for one week, and the assessment results are stored.

[0060] The electromyographic (EMG) signals are input into a support vector machine (SVM) to assess muscle status, and the first assessment result is output. This includes calculation methods based on root mean square (RMS), integrated EMG values, and median frequency (MF). As muscle function weakens, the RMS gradually decreases, the integrated EMG value (IEMG) gradually decreases, and the MF gradually decreases.

[0061] The formula for calculating the root mean square is: ,

[0062] in For each point of the electromyographic signal, This represents the number of sampling points for the electromyographic signal;

[0063] The formula for calculating the integral electromyographic value is as follows: ,

[0064] When calculating the median frequency, the power spectral density needs to be calculated using FFT: ,

[0065] in For power spectral density, For frequency, This is a time series of electromyographic signals, and FFT stands for Fast Fourier Transform.

[0066] The median frequency is calculated using the power spectral density, and the formula is as follows: .

[0067] Furthermore, a random forest is used to assess brain functional status, outputting a second assessment result. Specifically:

[0068] The intensity of different waves is obtained based on the power spectral density, including Wave, Wave, Wave, Wave. Among them Waves are associated with basic neural activity and postural stability; Waves are associated with attention and postural control; Waves reflect brain inhibition and sensor integration; Waves are associated with active movement and posture regulation.

[0069] The power spectral density of the wave is expressed as: ,

[0070] The power spectral density of the wave is expressed as: ,

[0071] The power spectral density of the wave is expressed as: ,

[0072] The power spectral density of the wave is expressed as: ,

[0073] in This is a time series of EEG signals, and FFT stands for Fast Fourier Transform.

[0074] Simultaneously, the phase-locked value (PLV) is used to calculate the phase synchronization between brain regions and assess the functional connectivity of the brain regions.

[0075] ,

[0076] in The number of sampling points for the electroencephalogram (EEG) signal. The instantaneous phase of the first EEG signal at time point n. Let n be the instantaneous phase of the second signal at time point n. It is a natural constant. It is the imaginary unit.

[0077] Furthermore, in addition to calculating the RMS, IEMG, and MF of electromyographic signals, and the four-wave power spectrum and PLV of electroencephalogram (EEG) signals, it is also necessary to calculate the reaction time of electromyographic signals. and EEG reaction time Specific stimuli are provided during the assessment. The calculation formula is:

[0078] ,

[0079] ,

[0080] in This refers to muscle reaction time; This refers to the muscle stimulation time. This refers to the brainwave response time, specifically the time when the P300 wave peaks. The duration of brain region stimulation.

[0081] S4: Input the acquired individual information, the first assessment result, the second assessment result, and the third assessment result into the fall risk assessment model, and output the risk coefficient.

[0082] In this embodiment, the first assessment result, the second assessment result, the third assessment result, and the acquired individual information are input into a fall risk assessment model for fall risk assessment. The fall risk assessment model is a deep neural network. The deep neural network consists of four fully connected layers (FCL) with 128, 256, 64, and 4 nodes respectively, and outputs the risk coefficients for forward, backward, left, and right falls.

[0083] S5: Input the multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficient into the fall warning model for training to obtain a trained fall warning model.

[0084] In this embodiment, the fall warning model is a recurrent neural network. The parameters of the risk coefficient, the convolutional neural network for feature extraction of the multimodal electrophysiological signals, the deep neural network for feature extraction of the electromyographic signals, and the convolutional neural network for feature extraction of the electroencephalogram signals are locked and then connected to the fall warning model. After connection, the fall warning model is trained to obtain a trained fall warning model.

[0085] Please see Figure 5 This is a schematic diagram of the recurrent neural network structure according to Embodiment 1 of this application. The recurrent neural network includes: a fully connected layer with 128 nodes, three LSTM layers with 64, 64 and 32 nodes respectively, and an output fully connected layer with 1 node.

[0086] Please see Figure 6This is the target network model of Embodiment 1 of this application, composed of a deep neural network, a convolutional neural network, and a recurrent neural network. This target network model is composed of DNN, CNN, and RNN, and continuously collects data from the normal state to the fall process to retrain the target network model. At this time, no parameters need to be locked; the target network model is directly trained in real-time for the task. Further optimization of the weights and biases is performed during training. The specific formula is:

[0087]

[0088]

[0089] in The optimized weights, For the pre-trained weights, For the optimized weight value, For the optimized bias, For the pre-training bias, This is the bias optimization value.

[0090] S6: Acquire the multimodal electrophysiological signals of the target user, predict the multimodal electrophysiological signals of the target user based on the trained fall warning model, and output the fall warning result.

[0091] In this embodiment, the trained fall warning model can be adjusted according to the individual differences of users, making it suitable for different user groups.

[0092] In summary, this application provides fall warnings by utilizing multi-site electromyographic signals and 16-channel electroencephalogram (EEG) signals triggered by human stress responses, capturing physiological and motor characteristics prior to a fall from multiple angles. It employs a multimodal electrophysiological sensor network with time synchronization to acquire multimodal electrophysiological signals simultaneously. Furthermore, it utilizes both single-point and multimodal electrophysiological feature extraction, considering both irrelevant and correlated features, thus reducing the false alarm rate and improving the predictability of the warning. The signal acquisition adds real-time warning capabilities to the human body's personal signal acquisition system, enabling early warning and providing reaction time. Moreover, the fall warning model can self-train and adjust according to individual differences, making it suitable for different user groups.

[0093] Example 2

[0094] Please see Figure 7 This is a schematic diagram of the structure of a fall warning system based on multi-electrophysiology according to Embodiment 2 of this application; the specific content includes:

[0095] Acquisition module: Acquires multimodal electrophysiological signals, including electromyography (EMG) signals and electroencephalography (EEG) signals;

[0096] Feature extraction module: performs feature extraction on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; performs feature extraction on the electromyographic signals and the electroencephalogram signals respectively to obtain first single-point electrophysiological features and second single-point electrophysiological features;

[0097] Assessment module: Assess muscle state based on electromyography (EMG) signals to obtain a first assessment result; assess brain function state based on electroencephalography (EEG) signals to obtain a second assessment result; calculate the reaction time of the EMG and EEG signals to obtain a third assessment result; input the acquired individual information, the first assessment result, the second assessment result, and the third assessment result into the fall risk assessment model, and output the risk coefficient;

[0098] Model training module: The multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficient are input into the fall warning model for training to obtain a trained fall warning model;

[0099] Output module: Acquires the multimodal electrophysiological signals of the target user, predicts the multimodal electrophysiological signals of the target user based on the trained fall warning model, and outputs the fall warning result.

[0100] In this embodiment, after the training module is completed, the system enters the usage stage and provides an early warning for falls. The system can be linked with protective equipment.

[0101] Please see Figure 8 This is a design block diagram of a fall warning system based on multimodal electrophysiology, according to Embodiment 2 of this application. It mainly includes two stages. The first stage is an individual assessment stage, which includes muscle state assessment to evaluate muscle health, single-point brain function assessment to evaluate the user's brain function state, and fall risk assessment based on multimodal electrophysiology and individual circumstances, saving the risk coefficient and type. The second stage is the warning stage, which includes a multimodal sensor network composed of multi-site electromyography and multi-channel electroencephalography, single-point electrophysiological feature extraction, multimodal electrophysiological feature extraction, and the muscle state, brain function state, and fall risk coefficient and type saved in the first stage. These are input into the fall warning model for training, so that the trained model can provide real-time warnings.

[0102] Example 3

[0103] Please see Figure 9 This is a schematic diagram of the fall warning device structure according to Embodiment 3 of this application. The fall warning device 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0104] The memory 52 stores program instructions for implementing the aforementioned fall warning method based on multiple electrophysiology.

[0105] The processor 51 is used to execute program instructions stored in the memory 52 to implement a fall warning system based on multiple electrophysiology.

[0106] The processor 51 can also be referred to as a CPU (Central Processing Unit).

[0107] Processor 51 may be an integrated circuit chip with signal processing capabilities. Processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0108] Example 4

[0109] Please see Figure 10 This is a schematic diagram of the storage medium in Embodiment 4 of this application. The storage medium in this embodiment stores a program file 61 capable of implementing all the above methods. This program file 61 can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or devices such as computers, servers, mobile phones, and tablets.

[0110] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0111] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0112] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

[0113] Of course, the present invention may have many other embodiments. Based on this embodiment, other embodiments obtained by those skilled in the art without any creative effort are all within the scope of protection of the present invention.

Claims

1. A fall warning method based on multi-electrophysiology, characterized in that, include: Multimodal electrophysiological signals are acquired, including electromyography (EMG) signals and electroencephalography (EEG) signals. The EMG signals are acquired from the following locations: lower limb muscles, upper limb muscles, gluteal muscles, back muscles, and neck muscles. The EEG signals are acquired using 16 channels. The EMG signals from different locations are networked with the 16-channel EEG signals to form a multimodal electrophysiological human sensor network. The electrophysiological acquisition points between the multimodal electrophysiological human sensor network are time-stamped and synchronized. After preprocessing with bandpass filtering and notch filtering, the EMG signals and EEG signals at the same time are acquired. Feature extraction is performed on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; The first convolutional neural network is used to extract features from the multimodal electrophysiological signals; The first convolutional neural network includes: a multi-location electromyography (EMG) convolutional pooling layer, a multi-channel electroencephalography (EEG) convolutional pooling layer, a comprehensive convolutional pooling layer, and a fully connected layer. The first convolutional neural network is pre-trained using EMG signals from various locations and multi-channel EEG signals to output multimodal electrophysiological features, i.e., the coordination level and current stability of muscles and brain at each location. Feature extraction is performed on the EMG signals and the EEG signals respectively to obtain a first single-point electrophysiological feature and a second single-point electrophysiological feature. A deep neural network is used for feature extraction of the EMG signals. The system consists of four fully connected layers. The electromyographic (EMG) signals are input into the deep neural network for pre-training, outputting a first single-point electrophysiological feature, namely, the degree of muscle contraction, fatigue, and lactic acid accumulation. A second convolutional neural network is used to extract features from the EEG signals. This second convolutional neural network consists of two convolutional pooling layers and two fully connected layers, with the two convolutional pooling layers performing dimensionality reduction on the EEG signals twice. The EEG signals are then input into the second convolutional neural network for pre-training, outputting a second single-point electrophysiological feature, namely, the degree of emotional tension, the degree of emotional relaxation, and the current level of attention. The muscle state is assessed based on the electromyography (EMG) signals to obtain a first assessment result. The EMG signals are then input into a support vector machine (SVM) to assess the muscle state and output the first assessment result. This includes calculation methods based on root mean square (RMS), integral EMG values, and median frequency. The brain function state is assessed based on the electroencephalography (EEG) signals to obtain a second assessment result. The reaction times of the EMG and EEG signals are calculated to obtain a third assessment result. The brain function state is then assessed using a random forest and the second assessment result is output. The acquired individual information, the first assessment result, the second assessment result, and the third assessment result are input into the fall risk assessment model, and the risk coefficient is output. The multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficients are input into the fall warning model for training to obtain a trained fall warning model. The fall risk assessment model is a deep neural network. The acquired individual information, the first assessment result, the second assessment result, and the third assessment result are input into the fall risk assessment model to output the risk coefficients for forward, backward, left, and right falls. The fall warning model is a recurrent neural network. The parameters of the fall risk assessment model, the first convolutional neural network for feature extraction of the multimodal electrophysiological signals, the deep neural network for feature extraction of the electromyographic signals, and the second convolutional neural network for feature extraction of the electroencephalogram signals are locked and then connected to the fall warning model. After connection, the fall warning model is trained to obtain a trained fall warning model. The multimodal electrophysiological signals of the target user are acquired, and the multimodal electrophysiological signals of the target user are predicted based on the trained fall warning model, and the fall warning result is output.

2. The fall warning method based on multi-electrophysiology according to claim 1, characterized in that, The formula for calculating the root mean square is: , in For each point of the electromyographic signal, This represents the number of sampling points for the electromyographic signal; The formula for calculating the integral electromyographic value is as follows: , Calculate the power spectral density using FFT: , in For power spectral density, For frequency, This is a time series of electromyographic signals, and FFT stands for Fast Fourier Transform. The median frequency is calculated using the power spectral density, and the formula is as follows: .

3. The fall warning method based on multi-electrophysiology according to claim 1, characterized in that, The brain functional state is assessed using random forest, and the second assessment results are output, including: The intensity of different waves is obtained based on the power spectral density, including Wave, Wave, Wave, Wave; The The power spectral density of the wave is expressed as: , The The power spectral density of the wave is expressed as: , The The power spectral density of the wave is expressed as: , The The power spectral density of the wave is expressed as: , in This is a time series of EEG signals, where FFT stands for Fast Fourier Transform. Phase synchronicity between brain regions was calculated using phase lock value (PLV) to assess functional connectivity of brain regions. ， in The number of sampling points for the electroencephalogram (EEG) signal. The instantaneous phase of the first EEG signal at time point n. The instantaneous phase of the second EEG signal at time point n. It is a natural constant. It is the imaginary unit.

4. A fall warning system for implementing the fall warning method based on multi-electrophysiology according to claim 1, characterized in that, include: Acquisition module: Acquires multimodal electrophysiological signals, including electromyography (EMG) signals and electroencephalography (EEG) signals; Feature extraction module: performs feature extraction on the multimodal electrophysiological signals to obtain multimodal electrophysiological features; Feature extraction is performed on the electromyography signal and the electroencephalography signal respectively to obtain the first single-point electrophysiological feature and the second single-point electrophysiological feature; Assessment module: Assess muscle state based on the electromyographic signals to obtain a first assessment result; The brain function status is assessed based on the electroencephalogram (EEG) signals to obtain a second assessment result. The reaction time of the electromyographic signal and the electroencephalogram signal is calculated to obtain a third assessment result; the acquired individual information, the first assessment result, the second assessment result, and the third assessment result are input into the fall risk assessment model to output the risk coefficient; Model training module: The multimodal electrophysiological features, the first single-point electrophysiological features, the second single-point electrophysiological features, and the risk coefficient are input into the fall warning model for training to obtain a trained fall warning model; Output module: Acquires the multimodal electrophysiological signals of the target user, predicts the multimodal electrophysiological signals of the target user based on the trained fall warning model, and outputs the fall warning result.

5. A fall warning device, characterized in that, The fall warning device includes a processor and a memory coupled to the processor, wherein the memory stores program instructions; the processor is used to execute the program instructions stored in the memory to implement a fall warning method based on multiple electrophysiology according to any one of claims 1-3.

6. A storage medium, characterized in that, The device stores processor-executable program instructions, which are used to execute a fall warning method based on multiple electrophysiology as described in any one of claims 1-3.

Images