Intelligent driving fatigue relief method and system

Abstract

The application relates to an intelligent driving fatigue relieving method and system, which integrates fatigue monitoring, evaluation, early warning and intervention, takes a wearable device as a carrier, fuses physiological signal sensing, machine learning analysis and acupoint electric stimulation feedback and other advanced technologies, realizes early discovery and active relief of fatigue, and has the advantages of convenient wearing, real-time online, intelligent adaptation, effective relief and the like, and is expected to fundamentally reduce fatigue driving accidents, and make an important contribution to traffic safety and public health.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, in particular, to an intelligent driving fatigue relieving method and system. BACKGROUND

[0002] Driving fatigue is a global public safety problem. According to the statistics of the World Health Organization, more than 1.2 million people die in traffic accidents every year, and fatigue driving is one of the main causes. Fatigue driving not only endangers the safety of the driver himself, but also brings great loss to other traffic participants and society. Therefore, timely and effective monitoring and relieving of driving fatigue is of great significance to reduce the incidence of traffic accidents and protect the safety of public life and property.

[0003] Traditional fatigue detection methods mainly include subjective evaluation and objective measurement. Subjective evaluation usually uses self-rating scales such as Stanford Sleepiness Scale (SSS), Caroline Sleepiness Scale (CSS), etc., but is easily affected by individual perception differences and lacks objectivity. Objective measurement focuses on physiological and behavioral indicators such as electroencephalogram (EEG), electrocardiogram (ECG), eye movement (EOG), facial expression, head posture, etc., but most of them require complex laboratory environment and are difficult to realize continuous and undisturbed detection.

[0004] In recent years, with the rapid development of wearable sensing and artificial intelligence technology, some intelligent fatigue detection systems have emerged, such as Driver Safety system of Seeing Machines, intelligent fatigue reminder glasses of Vigo, etc. They use head-mounted or glasses-type devices to realize real-time fatigue monitoring, and through sound and light alarm to prompt the driver to pay attention to the state of inattention, which has achieved certain effect. However, these systems mostly only provide post-alarm, without active intervention measures, and have limited preventive effect on fatigue driving. SUMMARY

[0005] In order to overcome at least one deficiency in the prior art, the present application provides an intelligent driving fatigue relieving method and system.

[0006] In a first aspect, an intelligent driving fatigue relieving method is provided, comprising:

[0007] Obtaining multi-channel physiological signals of the driver; the multi-channel physiological signals include pulsatile signals, skin conductance at the wrist, skin temperature, and hand movement state signals;

[0008] The multi-channel physiological signals and auxiliary information are input into a fatigue detection network model to obtain the fatigue degree of the driver; the fatigue detection network model comprises an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer; the input layer is used for splicing a tensor corresponding to the multi-channel physiological signals and a vector corresponding to the auxiliary information to obtain a high-dimensional input representation; the preprocessing layer is used for preprocessing the high-dimensional input representation to obtain a preprocessed feature representation; the hybrid attention encoding layer is used for performing time domain, frequency domain, and channel three kinds of attention operations on the preprocessed feature representation to obtain a plurality of hybrid attention feature maps of different scales; the global fusion layer is used for fusing the plurality of hybrid attention feature maps to obtain a fused global feature representation; and the output layer is used for determining the fatigue degree based on the fused global feature representation;

[0009] If the fatigue degree meets a set condition, an electric stimulation instruction is sent to an electric stimulation acupoint glove; and the electric stimulation acupoint glove performs electric stimulation on the driver according to the electric stimulation instruction to relieve fatigue.

[0010] In one embodiment, the hybrid attention encoding layer comprises L hybrid attention modules, and each hybrid attention module comprises a time domain attention unit, a frequency domain attention unit, and a channel attention unit;

[0011] Each hybrid attention module is used to implement the following functions:

[0012] The preprocessed feature is divided into N local windows, and each local window is input in parallel to the time domain attention unit, the frequency domain attention unit, and the channel attention unit to obtain time domain attention features, frequency domain attention features, and channel attention features;

[0013] Based on the time domain attention features, the frequency domain attention features, and the channel attention features, a hybrid attention feature map corresponding to the local window is obtained;

[0014] The hybrid attention feature maps corresponding to adjacent local windows are shifted and spliced in the spatial dimension to obtain an updated feature representation;

[0015] The updated feature representation is input into an MLP, and is connected in residual with the preprocessed feature, and then is subjected to Layer Normalization to obtain a hybrid attention feature map output by the hybrid attention module.

[0016] In one embodiment, the global fusion layer is used to implement the following functions:

[0017] The plurality of hybrid attention feature maps are spliced in the channel dimension to obtain a high-dimensional feature representation;

[0018] The high-dimensional feature representation is subjected to attention pooling to obtain a compact feature vector;

[0019] The compact feature vector is input into a fully connected layer to map to a low-dimensional hidden space to obtain a fused global feature representation.

[0020] In one embodiment, the method further comprises:

[0021] acquiring a multi-channel physiological signal of the driver in real time;

[0022] calculating a stimulation effect index according to the multi-channel physiological information;

[0023] determining whether the stimulation effect index is within an ideal range, and if so, sending a command to stop electrical stimulation to the electrical stimulation acupoint glove, and if not, sending the stimulation effect index to the electrical stimulation acupoint glove.

[0024] In one embodiment, the method further comprises:

[0025] comparing the change in fatigue score before and after electrical stimulation based on a fatigue detection algorithm to calculate a physiological fatigue relief rate;

[0026] acquiring subjective feeling information of the driver before and after electrical stimulation, and calculating a subjective fatigue relief rate based on the subjective feeling information;

[0027] analyzing driving behavior data, comparing the change before and after electrical stimulation, and calculating a driving fatigue relief rate;

[0028] determining the effect of electrical stimulation according to the physiological fatigue relief rate, the subjective fatigue relief rate, and the driving fatigue relief rate;

[0029] sending the effect of electrical stimulation to the electrical stimulation acupoint glove.

[0030] In a second aspect, an intelligent driving fatigue relief device is provided, comprising:

[0031] a signal acquisition module for acquiring a multi-channel physiological signal of the driver; the multi-channel physiological signal includes a pulsatile signal, a skin conductance at the wrist, a skin temperature, and a hand movement state signal;

[0032] The fatigue detection module is configured to input the multi-channel physiological signals and the auxiliary information into a fatigue detection network model to obtain a fatigue degree of the driver. The fatigue detection network model comprises an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer. The input layer is configured to splice a tensor corresponding to the multi-channel physiological signals and a vector corresponding to the auxiliary information to obtain a high-dimensional input representation. The preprocessing layer is configured to preprocess the high-dimensional input representation to obtain a preprocessed feature representation. The hybrid attention encoding layer is configured to perform time domain, frequency domain, and channel three kinds of attention operations on the preprocessed feature representation to obtain a plurality of hybrid attention feature maps of different scales. The global fusion layer is configured to fuse the plurality of hybrid attention feature maps to obtain a fused global feature representation. The output layer is configured to determine the fatigue degree based on the fused global feature representation.

[0033] The instruction sending module is configured to send an electric stimulation instruction to the electric stimulation acupoint glove if the fatigue degree meets a set condition. The electric stimulation acupoint glove performs electric stimulation on the driver according to the electric stimulation instruction to relieve fatigue.

[0034] In a third aspect, an intelligent driving fatigue relief system is provided, comprising: a portable bracelet, an electric stimulation acupoint glove, and a mobile device end.

[0035] The portable bracelet is configured to collect multi-channel physiological signals of the driver. The multi-channel physiological signals comprise a pulsatile signal, a skin conductance at the wrist, a skin temperature, and a hand movement state signal.

[0036] The electric stimulation acupoint glove is configured to perform electric stimulation on the driver according to the received electric stimulation instruction to relieve fatigue.

[0037] The mobile device end is configured to implement the intelligent driving fatigue relief method described above.

[0038] In an embodiment, the portable bracelet is provided with a PPG sensor, an EDA sensor, a skin temperature sensor, and a motion sensor. The PPG sensor is configured to collect the pulsatile signal, the EDA sensor is configured to collect the skin conductance at the wrist, the skin temperature sensor is configured to collect the skin temperature, and the motion sensor is configured to collect the hand movement state signal.

[0039] In an embodiment, the electric stimulation acupoint glove is further configured to:

[0040] determine optimal electric stimulation parameters based on a deep reinforcement learning algorithm. The electric stimulation parameters comprise current intensity, frequency, and pulse width. In the deep reinforcement learning algorithm, the received electric stimulation effect is taken as a reward function.

[0041] perform electric stimulation on the target acupoint according to the optimal electric stimulation parameters and the fatigue degree.

[0042] In one embodiment, the electric stimulation acupoint glove is also used for:

[0043] Adjusting the electric stimulation parameters according to the received stimulation effect index.

[0044] In one embodiment, the electric stimulation acupoint glove further comprises a flexible electrode array, an electric stimulation generator, and a biofeedback sensor; the flexible electrode array is used to act electric current on acupoints; the electric stimulation generator is used to generate electric stimulation current according to received electric stimulation instructions; and the biofeedback sensor is used to monitor the contact state of the electrode and the skin and local temperature changes.

[0045] In one embodiment, a wireless communication module is arranged in both the portable bracelet and the electric stimulation acupoint glove, so as to realize the synchronization of the portable bracelet and the electric stimulation acupoint glove.

[0046] Compared with the prior art, the intelligent driving fatigue relief method and system of the present application has the following beneficial effects: the intelligent driving fatigue relief method and system of the present application integrates fatigue monitoring, evaluation, early warning and intervention as a whole, takes wearable devices as carriers, and fuses physiological signal sensing, machine learning analysis and acupoint electric stimulation feedback and other advanced technologies, so as to realize early detection and active relief of fatigue; the system has the advantages of convenient wearing, real-time online, intelligent adaptation and effective relief, and is expected to fundamentally reduce fatigue driving accidents, and make important contributions to traffic safety and public health. BRIEF DESCRIPTION OF DRAWINGS

[0047] The present application can be better understood by referring to the description given below in conjunction with the accompanying drawings, which are included in the specification and form a part of the specification. In the drawings:

[0048] Figure 1 A schematic diagram of an intelligent driving fatigue relief system according to an embodiment of the present application is shown;

[0049] Figure 2 A principle diagram of a closed-loop electric stimulation fatigue relief strategy based on biofeedback and reinforcement learning is shown;

[0050] Figure 3 A structural schematic diagram of a fatigue detection network model is shown;

[0051] Figure 4 A structural schematic diagram of a hybrid attention module is shown;

[0052] Figure 5 A structural block diagram of an intelligent driving fatigue relief device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0053] The exemplary embodiments of this application will be described hereinafter with reference to the accompanying drawings. For the purpose of clarity and a concise description, all the features of the practical embodiments are not described in the specification. It should be appreciated, however, that many embodiment-specific decisions can be made in the process of developing any such practical embodiments in order to achieve the specific goals of the developers, and these decisions can vary from embodiment to embodiment.

[0054] It should also be noted here that, in order to avoid obscuring the application with unnecessary details, only the device structures closely related to the scheme according to the application are shown in the drawings, and other details not closely related to the application are omitted.

[0055] It should be understood that the application is not limited to the described embodiments by virtue of the following description with reference to the drawings. In this context, the embodiments can be combined with each other, features can be replaced or borrowed between different embodiments, and one or more features can be omitted in one embodiment.

[0056] The application provides an intelligent driving fatigue relief system, Figure 1 A schematic diagram of the intelligent driving fatigue relief system according to an embodiment of the application is shown, referring to Figure 1 The system comprises a portable bracelet, an electric stimulation acupoint glove, and a mobile device end. The specific implementation functions of each module are described in detail below.

[0057] The portable bracelet is used to collect the multi-channel physiological signals of the driver; the multi-channel physiological signals include pulsatile signals, skin conductance at the wrist, skin temperature, and hand movement state signals.

[0058] Specifically, the portable bracelet is provided with a PPG sensor, an EDA sensor, a skin temperature sensor, and a motion sensor; the PPG sensor is used to collect the pulsatile signals, the EDA sensor is used to collect the skin conductance at the wrist, the skin temperature sensor is used to collect the skin temperature, and the motion sensor is used to collect the hand movement state signals.

[0059] The photoplethysmogram (PPG) sensor adopts a Texas Instruments AFE4400 EVK platform, contains a green light LED and a high-sensitivity photodiode, collects the skin capillary volume pulsatile signals through reflection, and monitors key physiological indicators such as blood oxygen saturation (SpO2), heart rate (HR), and heart rate variability (HRV) in real time. The sampling rate of the sensor is 100 Hz, the dynamic range is 0-100%, and the accuracy is better than ±2%.

[0060] Skin conductance (EDA) sensor, using Thought Technology's SA9309M sensor, reflects the degree of sympathetic nervous system excitation by measuring the change of skin conductance at the wrist. The working current of the sensor is less than 10μA, the sensitivity can reach 0.01μS, and the frequency response is 0-10Hz.

[0061] Skin temperature sensor, using Measurement Specialties' 10K3A542 NTC thermistor, evaluates the body temperature regulation function by measuring the change of skin surface temperature. The resistance of the sensor is 10kΩ, the B value is 3435K, the temperature measurement range is 0-60℃, and the accuracy can reach ±0.1℃.

[0062] Motion sensor, using InvenSense's MPU-6050 six-axis motion processing unit, integrates a three-axis MEMS gyroscope and a three-axis MEMS accelerometer, which can monitor the motion state and attitude change of the hand in real time. The gyroscope range of the sensor is ±2000dps, the accelerometer range is ±16g, and the digital output interface is I2C.

[0063] In addition, the portable bracelet also includes a microprocessor, a wireless communication module, a storage module and a power supply module.

[0064] The microprocessor uses STM32F405RGT6 single-chip microcomputer with ARM Cortex-M4 core, with a main frequency of 168MHz, 1MB Flash and 192KB SRAM built-in, and rich peripheral interfaces, which can meet the needs of multi-sensor data acquisition, fusion and processing.

[0065] The wireless communication module uses Nordic nRF51822 Bluetooth 4.0 SoC chip, supports BLE protocol stack, ultra-low power consumption, and good compatibility with mobile phone APP. The chip has a built-in 2.4GHz transceiver, Flash and RAM are 256KB and 16KB respectively, and peripheral interfaces include SPI, I2C, UART, etc.

[0066] The storage module uses Micron's MT29F4G08ABBDAH4 NAND Flash, with a capacity of 4GB, supporting SLC mode, read and write speeds of 40MB / s and 12MB / s, and can store at least a week of continuous monitoring data.

[0067] The power supply module uses Texas Instruments' bq27421-G1 power meter and bq24074 battery charging management chip, with a built-in 500mAh polymer lithium battery, supporting USB charging. The average power consumption of the system is less than 5mW, and the endurance time can reach more than 7 days.

[0068] The mobile device end is used to acquire the multi-channel physiological signals of the driver, and determine the fatigue degree of the driver according to the multi-channel physiological signals by using a fatigue detection network model; and send an electric stimulation instruction to the electric stimulation acupoint gloves according to the fatigue degree. Here, the mobile device end can be a smart phone APP. The fatigue detection network model is a model obtained after training and saved in the cloud, and its specific training process is prior art and will not be described here. The communication between the mobile device end and the portable bracelet and the electric stimulation acupoint gloves is based on the BLE protocol, and the data transmission adopts encryption measures to ensure privacy and security.

[0069] The electric stimulation acupoint gloves are used to stimulate the driver according to the received electric stimulation instruction to relieve fatigue.

[0070] Specifically, the electric stimulation acupoint gloves include a flexible electrode array, an electric stimulation generator, and a biofeedback sensor; the flexible electrode array is used to apply electric current to the acupoint; the electric stimulation generator is used to generate electric stimulation current according to the received electric stimulation instruction; and the biofeedback sensor is used to monitor the contact state of the electrode and the skin and the local temperature change.

[0071] Flexible electrode array: a 10x10 electrode array is made of silver / silver chloride conductive fabric, with an electrode spacing of 5mm, excellent conductivity, and strong oxidation resistance. The back of the electrode is coated with conductive silicone, which tightly adheres to the skin and reduces contact impedance.

[0072] Electric stimulation generator: a programmable constant current stimulation circuit is designed, which is composed of STM32L151 single-chip microcomputer, DAC8562 digital-to-analog converter, OPA2340 operational amplifier and other devices. The circuit can output a double-phase square wave pulse with an amplitude of 0-10mA and a pulse width of 50-500us in a frequency range of 5-200Hz, realizing flexible and controllable electric stimulation parameter modulation.

[0073] Biofeedback sensor: an array of pressure sensors (Interlink FSR402) and temperature sensors (Semitec 103JT) is arranged around the electrode to monitor the contact state of the electrode and the skin and the local temperature change in real time, avoiding discomfort caused by excessive stimulation.

[0074] The electric stimulation acupoint gloves are also provided with a power supply module, which is built-in with a 200mAh polymer lithium battery and adopts a wireless charging method (Qi standard) with a coil and a rectifier filter circuit. The average power consumption of the gloves is about 10mW, and the endurance time can reach 24 hours.

[0075] Further, the portable bracelet and the electric stimulation acupoint gloves are both provided with a wireless communication module for synchronizing the portable bracelet and the electric stimulation acupoint gloves.

[0076] In one embodiment, the electric stimulation acupoint glove determines the electric stimulation parameters after receiving the fatigue degree, specifically, determines the optimal electric stimulation parameters based on a deep reinforcement learning algorithm; the electric stimulation parameters include current intensity, frequency and pulse width; in the deep reinforcement learning algorithm, the received electric stimulation effect is taken as a reward function, the electric stimulation parameters are taken as an action space, and the fatigue state representation vector is taken as a state space, and through continuous trial and error and policy iteration, the optimal control sequence is gradually approached.

[0077] Then, electric stimulation is applied to the target acupoint according to the optimal electric stimulation parameters and the fatigue degree.

[0078] Here, for the complex physiological mechanism of fatigue, this embodiment proposes a multi-channel collaborative electric stimulation scheme, which comprehensively considers the functional characteristics of different acupoints and produces more comprehensive and effective fatigue relief effects through the synergistic effect of multiple electrodes.

[0079] Specifically, according to the meridian theory and modern acupuncture research, the following groups of acupoints are selected for collaborative stimulation:

[0080] · The sober acupoint group includes Hegu (LI4), Neiguan (PC6), Waiguan (TE5), etc., and mainly plays a role in refreshing the mind, restoring consciousness;

[0081] · The calming acupoint group includes Shenmen (HT7), Daling (PC7), Shao Hai (HT3), etc., and mainly plays a role in calming the mind, relieving emotions;

[0082] · The balance acupoint group includes Laogong (PC8), Zhongchong (TE3), Cunshu (SI4), etc., and mainly plays a role in regulating yin and yang, restoring physical strength.

[0083] In the actual stimulation process, the stimulation intensity and time proportion of the three groups of acupoints are dynamically adjusted according to different stages of fatigue state. For example, in the moderate fatigue stage, the stimulation intensity and frequency of the sober acupoint group are appropriately increased; in the severe fatigue stage, the stimulation intensity and duration of the calming acupoint group are appropriately increased; and in the fatigue recovery stage, the balance acupoint group is mainly stimulated. The synergistic effect of the three groups of acupoints can relieve fatigue symptoms from multiple aspects, achieving a multiplier effect.

[0084] In this embodiment, electric stimulation achieves the effect of refreshing the mind by regulating the nervous system, especially the sympathetic / parasympathetic balance. However, improper setting of electric stimulation parameters (such as current intensity, frequency, pulse width, etc.) not only fails to achieve the ideal effect, but also may cause discomfort. Therefore, it is necessary to optimize the electric stimulation parameters according to individual differences and fatigue degree. This embodiment uses a deep reinforcement learning (DRL) algorithm to establish a mapping relationship between the electric stimulation parameters and the fatigue relief effect.

[0085] In one embodiment, in order to further improve the pertinence and real-time performance of the electric stimulation, a biofeedback control mechanism is introduced in this embodiment. This mechanism dynamically adjusts the electric stimulation parameter SEI by monitoring the physiological response of the driver in real time, so as to realize the synchronous optimization of the stimulation effect and the physiological state. The electric stimulation acupoint glove is also used to adjust the electric stimulation parameter according to the received stimulation effect index SEI. Figure 2 The principle diagram of the closed-loop electric stimulation fatigue relief strategy based on biofeedback and reinforcement learning is shown.

[0086] Specifically, the basic principle of adjustment is: when SEI is lower than the first set value (the first set value can be 3.5), the stimulation intensity is appropriately increased, such as increasing the current amplitude or pulse frequency; when SEI is higher than the second set value (the second set value can be 4.5), the stimulation intensity is appropriately reduced; when SEI is in the ideal interval (such as between 3.5 and 4.5), the current parameters are maintained unchanged.

[0087] The determination process of the stimulation effect index SEI is implemented in the mobile device end, and the specific implementation process can be referred to the following embodiments.

[0088] The embodiment of the present application also provides an intelligent driving fatigue relief method, which is applied to the mobile device end. The specific process of the intelligent driving fatigue relief method is specifically introduced below, and the method comprises the following steps:

[0089] Step S1, obtaining the multi-channel physiological signals of the driver; the multi-channel physiological signals comprise the pulsatile signal, the skin conductance at the wrist, the skin temperature, and the hand movement state signal.

[0090] Step S2, inputting the multi-channel physiological signals and auxiliary information into a fatigue detection network model (HST model) to obtain the fatigue degree of the driver; here, the auxiliary information comprises age, height, time, environment, and physiological information.

[0091] Figure 3 The structure diagram of the fatigue detection network model is shown, which is shown in Figure 3 The fatigue detection network model comprises an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer; the input layer is used to splice the tensor corresponding to the multi-channel physiological signals and the vector corresponding to the auxiliary information to obtain a high-dimensional input representation; the preprocessing layer is used to preprocess the high-dimensional input representation to obtain a preprocessed feature representation; the hybrid attention encoding layer is used to perform time domain, frequency domain, and channel three kinds of attention operations on the preprocessed feature representation to obtain a plurality of hybrid attention feature maps of different scales; the global fusion layer is used to fuse the plurality of hybrid attention feature maps to obtain a fused global feature representation; and the output layer is used to determine the fatigue degree based on the fused global feature representation.

[0092] Step S3, if the fatigue degree meets the set condition, send an electric stimulation instruction to the electric stimulation acupoint glove; the electric stimulation acupoint glove stimulates the driver according to the electric stimulation instruction to relieve fatigue.

[0093] Here, the fatigue degree can include severe fatigue, moderate fatigue and mild fatigue, and the set condition can be, for example, that the fatigue degree is moderate fatigue or severe fatigue.

[0094] In the above embodiment, the fatigue detection network model is a deep learning framework, which combines the self-attention mechanism of Transformer and the local modeling ability of CNN to build an end-to-end spatio-temporal sequence classifier. The model inherits the successful experience of Swin Transformer in the field of computer vision, and through the introduction of window-based local self-attention mechanism and cross-window information interaction, it reduces the computational complexity while improving the richness of feature representation. At the same time, the model also integrates attention mechanisms of three different granularities in time domain, frequency domain and channel, which can adaptively focus on significant features in different modalities and different scales, thereby greatly improving the accuracy of fatigue state discrimination.

[0095] In one embodiment, the input data contains 4 channels, each with 3000 sampling points, corresponding to about 5 minutes of measurement time; at the same time, there are 4-dimensional auxiliary information. The model contains 4 mixed attention encoding layers, each with 10 local windows, the hidden layer dimension is 64, and the output is 3 fatigue degrees.

[0096] The input layer obtains the tensor X ∈ R C×T , C × T is the dimension of the tensor, and the auxiliary information vector A ∈ R D , D is the dimension of the vector, and the channel dimension is spliced to obtain a high-dimensional input representation Z0 ∈ R (C+D)×T :

[0097] Z0 = Concat (X, A)

[0098] This layer does not introduce any learnable parameters.

[0099] The pre-processing layer pre-processes the high-dimensional input representation to obtain a pre-processed feature representation Z1; the pre-processing operation can include data normalization, noise filtering, data enhancement, etc., according to the specific task and data characteristics. This layer usually only introduces a small number of fixed parameters (such as normalization coefficients) and does not participate in model training.

[0100] The hybrid attention encoding layer consists of L hybrid attention modules, each of which includes a time-domain attention unit, a frequency-domain attention unit, and a channel-domain attention unit. The hybrid attention encoding layer is the core component of the HST model and is composed of multiple stacked hybrid attention modules. Each module contains three parallel attention sub-layers, modeling the intrinsic dependencies of the sequence at three different granularities: time, frequency, and channel.

[0101] Figure 4 shows the structural diagram of the hybrid attention module, see Figure 4 , each hybrid attention module is used to implement the following functions, which are described by taking the lth hybrid attention module as an example:

[0102] The preprocessed features, i.e. the features input by the lth hybrid attention module, are denoted as Z l , divided into N local windows, denoted as The size of the local window is M;

[0103] The following three attentions are calculated for each local window, with the i-th local window Take the example to describe, that is, the local window is input into the time domain attention unit, the frequency domain attention unit and the channel attention unit in parallel to obtain the time domain attention feature, the frequency domain attention feature and the channel attention feature;

[0104] Temporal attention unit: Through the self-attention mechanism, it models the long-range dependencies between different time steps within the window and highlights local temporal patterns. It is described by the following formula:

[0105]

[0106] Among them, for the local window Perform linear transformation to obtain the query matrix of time domain attention Bond Matrix Value Matrix d is the hidden layer dimension, which is used to scale the dot product result. Softmax operation is normalized in the time dimension. Based on the above formula, the time domain attention feature is calculated

[0107] Frequency-domain attention unit: Through a similar self-attention mechanism, it models the dependencies between different frequency components within a window and highlights local frequency-domain features. It is described by the following formula:

[0108]

[0109] Among them, for the local window Perform Fourier transform and linear projection to obtain the query matrix of time domain attention Key matrix Value matrix Based on the above formula, the frequency domain attention feature is calculated

[0110] Channel attention unit: through the Squeeze-and-Excitation mechanism, the importance of different input channels is adaptively adjusted, and the cross-modal information interaction fusion is strengthened. The following formula is used to describe:

[0111]

[0112] Where, W1, W2 are two learnable weight vectors, vgPool and MaxPool represent the average pooling and maximum pooling operations in the time dimension respectively, used to compress the time dimension information. The sigmoid function performs nonlinear transformation and normalization on the channel weight. The channel attention feature is.

[0113] Based on the time domain attention feature, the frequency domain attention feature and the channel attention feature, the mixed attention feature map corresponding to the local window is obtained; the following formula is used to describe:

[0114]

[0115] Where, The mixed attention feature map corresponding to the i-th local window The mixed attention feature map corresponding to the i-th local window.

[0116] The mixed attention feature maps corresponding to adjacent local windows are shifted and spliced in the spatial dimension to obtain the updated feature representation Z' l ; here, the shift and splice use the Shifted Window strategy.

[0117] The updated feature representation Z' l is input into the MLP, and is connected with the preprocessed feature Z l , and then goes through the Layer Normalization to obtain the mixed attention feature map output by the mixed attention module, here, the mixed attention feature map output by the l-th mixed attention module is denoted by . Specifically, the following formula is used:

[0118]

[0119] Where, MLP is a two-layer feedforward neural network, used to enhance the representation ability of the feature.

[0120] The global fusion layer is used to realize the following functions:

[0121] The plurality of mixed attention feature maps are spliced in the channel dimension to obtain a high-dimensional feature representation G; the following formula is used:

[0122]

[0123] The high-dimensional feature representation G is obtained through attention pooling to obtain a compact feature vector g; the following formula is used:

[0124] g = Attention(G)

[0125] The compact feature vector g is input into a fully connected layer to map to a low-dimensional hidden space to obtain a fused global feature representation h. The following formula is used:

[0126] h = W h g + b h

[0127] where W h and b h are the weight matrix and bias matrix of the fully connected layer, respectively.

[0128] Output layer: based on the global feature h, a Softmax classifier is used to predict the fatigue degree corresponding to the current sampling window The following formula is used:

[0129]

[0130] where W o and b o are the weight matrix and bias vector of the classifier, respectively.

[0131] In the above embodiment, the hybrid attention mechanism is used to fully utilize the self-attention of the Transformer and the local receptive field of the CNN, to model the internal patterns of the physiological signal in the time domain, frequency domain and channel three granularities, and to mine the correlation between different modalities, so as to obtain a more comprehensive and more detailed fatigue state feature representation.

[0132] Local-global feature extraction is used to extract local and global dependencies of the spatio-temporal sequence in different scales through local window division and cross-window information interaction, to capture fine-grained regional patterns and also consider global context information, thereby enhancing the feature representation capability of the model.

[0133] An end-to-end learning paradigm is used to abandon traditional manual feature engineering, to realize direct modeling and classification of the original physiological signal through end-to-end joint optimization, to simplify the system design process, and to improve the generalization performance of the model.

[0134] The local window attention and parameter sharing mechanism are introduced, which greatly reduces the computational complexity and memory consumption of Self-Attention, enables the model to process long sequence input, and realizes more efficient and scalable training and inference.

[0135] The multi-granularity spatio-temporal modeling is adopted, the attention mechanism is applied in time domain, frequency domain and channel with three different granularities, the HST model can comprehensively depict the time dependence, spectral characteristics and cross-modal interaction of physiological signals, and construct a more accurate and robust fatigue state discrimination model.

[0136] The fatigue detection network model (HST model) of the embodiment has the following technical advantages:

[0137] 1. Compared with the traditional machine learning method based on artificial design features, the HST model adopts an end-to-end deep learning paradigm, which can automatically learn high-level semantic features in the data, reduce the dependence on expert experience, and improve the self-adaptation ability and generalization performance of the model.

[0138] 2. Compared with the method based on single neural network architecture such as CNN and RNN, the HST model introduces a hybrid attention mechanism and a local-global feature extraction strategy, which can better mine the long-range dependence and cross-modal correlation in physiological signals, and obtain a more comprehensive and fine-grained fatigue state representation.

[0139] 3. Compared with the original Transformer model, HST significantly reduces the computational complexity and memory consumption by introducing local window attention and parameter sharing, so that the model can process longer time series data, realize more efficient and scalable training and inference.

[0140] 4. The HST model provides a general spatio-temporal sequence modeling framework that can be flexibly adapted to different input data types and application scenarios. In addition to driver fatigue detection, this model can also be applied to other physiological signal analysis tasks such as emotion recognition, sleep monitoring, and seizure warning.

[0141] In summary, the HST model inherits the advantages of Transformer and CNN, and is improved and optimized for the characteristics of physiological signals, forming an end-to-end solution suitable for driver fatigue detection. The hybrid attention mechanism, local-global feature extraction, and end-to-end learning embodied in this model provide a new paradigm for spatio-temporal sequence data modeling and analysis, and are expected to have a wide impact in multiple fields.

[0142] In one embodiment, the intelligent driving fatigue relief method further comprises:

[0143] Real-time acquisition of multi-channel physiological signals of the driver;

[0144] calculating a stimulation effect index according to multi-channel physiological information;

[0145] determining whether the stimulation effect index is within an ideal range, if yes, sending a command to stop electrical stimulation to the electrical stimulation acupoint glove, if not, sending the stimulation effect index to the electrical stimulation acupoint glove.

[0146] Here, in order to further improve the pertinence and real-time of electrical stimulation, the present application introduces a biofeedback control mechanism, which dynamically adjusts the electrical stimulation parameters by real-time monitoring of the physiological response of the driver, to realize the synchronous optimization of stimulation effect and physiological state.

[0147] Specifically, the system continuously monitors the following physiological indicators during electrical stimulation:

[0148] Instantaneous heart rate (HR): reflects the influence of electrical stimulation on the cardiovascular system, HR should increase slightly after stimulation, but should not exceed the safe range (such as 120 times / min);

[0149] Blood oxygen saturation (SpO2): reflects the influence of electrical stimulation on the respiratory system, SpO2 should not decrease significantly after stimulation, and the ideal situation is to maintain above 95%;

[0150] Skin conductance (EDA): reflects the influence of electrical stimulation on the sympathetic nervous system, EDA should decrease after stimulation, indicating a decrease in sympathetic nervous excitability;

[0151] Electroencephalogram (EEG): reflects the influence of electrical stimulation on the cerebral cortex, EEG should show an increase in alpha waves and a decrease in theta waves after stimulation, indicating an increase in alertness.

[0152] According to the changes of the above indicators, the system will calculate a "stimulation effect index (SEI)" in real time as the basis for optimizing the electrical stimulation parameters.

[0153] For example, for a fatigued driver, real-time monitoring of the following physiological indicators is performed:

[0154] Before intervention: HR=60 times / min, SpO2=98%, EDA=2μS, alpha wave ratio=30%, theta wave ratio=40%

[0155] After intervention: HR=70 times / min, SpO2=97%, EDA=1.5μS, alpha wave ratio=40%, theta wave ratio=30%

[0156] First, the change of each indicator needs to be scored. The basis for scoring is whether the change is in the direction of relieving fatigue and whether the amplitude of the change is within the safe and effective range. The following scoring rules can be set:

[0157] Heart rate (HR): HR increased by 10 beats / min after the intervention, which is within the safe range (no more than 120 beats / min). The increase is moderate, indicating that the stimulation intensity is appropriate. Score = 4 (5 is the full score)

[0158] Oxygen saturation (SpO2): SpO2 decreased by 1% after the intervention, which is a small change and still within the normal range (> 95%). Score = 5 (5 is the full score)

[0159] Electrodermal activity (EDA): EDA decreased by 0.5 μS after the intervention, indicating a slight decrease in sympathetic nervous excitability, which is consistent with the expected effect. However, the decrease is slightly small, and the stimulation intensity can be appropriately increased. Score = 3 (5 is the full score)

[0160] Electroencephalogram (EEG): The proportion of α waves increased by 10% and the proportion of θ waves decreased by 10% after the intervention, indicating an improvement in alertness. The change direction is correct, and the amplitude is moderate. Score = 4 (5 is the full score)

[0161] Second, the scores of the four indicators are weighted and averaged to obtain the preliminary "stimulation effect index (SEI)". The allocation of weights should consider the degree of influence of different indicators on fatigue status. Generally, it is believed that the weights of EEG and HR should be larger because brain electrical activity and heart rate variability are most closely related to cognitive fatigue; followed by EDA, reflecting the excitability of the sympathetic nervous system; and the weight of SpO2 can be smaller because blood oxygen changes are relatively lagging and have limited predictive effect on fatigue.

[0162] Assuming the allocated weights are: wEEG = 0.4, wHR = 0.3, wEDA = 0.2, wSpO2 = 0.1, then the calculation formula of SEI is:

[0163] SEI = wEEG × scoreEEG + wHR × scoreHR + wEDA × scoreEDA + wSpO2 × scoreSpO2

[0164] where wEEG, wHR, wEDA, and wSpO2 are the weights of EEG, HR, EDA, and SpO2, respectively, and scoreEEG, scoreHR, scoreEDA, scoreSpO2 are the scores of EEG, HR, EDA, and SpO2, respectively.

[0165] Substituting the above score results, we get:

[0166] SEI = 0.4 × 4 + 0.3 × 4 + 0.2 × 3 + 0.1 × 5 = 3.9

[0167] This result shows that the current electrical stimulation program is generally effective in alleviating fatigue, but the SEI has not yet reached the ideal level (4.0-4.5), and there is room for further optimization.

[0168] In the third step, the parameters of the next round of electrical stimulation need to be adjusted dynamically based on the SEI value. The basic principle of adjustment is: when SEI is low (e.g. <3.5), appropriately increase the stimulation intensity, such as increasing the current amplitude or pulse frequency; when SEI is too high (e.g. >4.5), appropriately reduce the stimulation intensity; when SEI is in the ideal interval, maintain the current parameters unchanged.

[0169] In this example, SEI = 3.9, close to the lower limit of the ideal interval, so the stimulation intensity can be slightly increased, such as increasing the current amplitude from 2mA to 2.2mA, or increasing the pulse frequency from 20Hz to 25Hz. This can further improve the fatigue relief effect while ensuring safety and comfort.

[0170] The above is the basic principle and operation process of real-time optimization of electrical stimulation parameters using biofeedback mechanism. The key is to establish a quantitative correlation between physiological indicator changes and fatigue state improvement, and design reasonable scoring rules and control strategies accordingly. Through this dynamic closed-loop control, the electrical stimulation intervention effect can be precisely matched with the physiological needs of the driver, thereby significantly improving the effectiveness and individual adaptability of fatigue relief. Through this biofeedback control strategy, the electrical stimulation process can be dynamically matched with the physiological state of the driver, neither affecting the fatigue relief effect due to insufficient stimulation, nor causing discomfort due to excessive stimulation, thereby achieving the purpose of safe, efficient and individualized intervention.

[0171] In one embodiment, the ultimate goal of electrical stimulation intervention is to alleviate the fatigue state of the driver, so it is necessary to objectively evaluate the stimulation effect. This embodiment designs a multi-dimensional fatigue relief index system to quantitatively evaluate the effect of electrical stimulation from the aspects of physiology, psychology and behavior. The method also includes:

[0172] Based on the fatigue detection algorithm, compare the changes in fatigue scores before and after electrical stimulation, and calculate the physiological fatigue relief rate; here, the fatigue detection algorithm is used to compare the changes in fatigue scores before and after stimulation, and a "physiological fatigue relief rate" is calculated. At the same time, the average HR, SpO2, EDA and other indicators during the stimulation process are counted to evaluate the improvement of physiological state.

[0173] Obtain the subjective feeling information of the driver before and after electrical stimulation, and calculate the subjective fatigue relief rate based on the subjective feeling information; here, through subjective questionnaire survey, the fatigue feeling, attention level, emotional state and other information of the driver before and after stimulation are obtained, and a "subjective fatigue relief rate" is calculated. At the same time, the satisfaction scores of the driver on the comfort, duration and other aspects of electrical stimulation are recorded.

[0174] analyze the driving behavior data, compare the changes before and after the electric stimulation, and calculate the driving fatigue relief rate; here, by analyzing the driving behavior data (such as vehicle speed, accelerator, brake, etc.), comparing the changes before and after the stimulation, and calculating a "driving fatigue relief rate". At the same time, the number of high-risk events such as lane deviation, sudden acceleration, sudden deceleration during the stimulation is counted, and the degree of reduction of driving risk is evaluated.

[0175] According to the physiological fatigue relief rate, the subjective fatigue relief rate and the driving fatigue relief rate, the effect of electric stimulation is determined; for example, the weighted average of the physiological fatigue relief rate and the subjective fatigue relief rate can be used as a quantitative indicator of the overall fatigue improvement level; the weighted average of the improvement degree of physiological state and the reduction degree of driving risk can be used as a quantitative indicator of the improvement level of driving safety. By setting reasonable weight coefficients, the quantitative indicators of the overall fatigue improvement level and the quantitative indicators of the improvement level of driving safety can be added to obtain a comprehensive "electric stimulation effect score", which directly reflects the overall effect of the intervention.

[0176] The electric stimulation effect is sent to the electric stimulation acupoint glove. After the electric stimulation acupoint glove receives the electric stimulation effect, it is used as a reward function in the reinforcement learning algorithm to optimize the electric stimulation parameters.

[0177] In this embodiment, such a multi-dimensional evaluation mechanism not only verifies the effectiveness of the present application, but also provides an important basis for subsequent system optimization. By statistically analyzing the effect data of different drivers, different fatigue stages and different stimulation modes, the best intervention strategy can be summarized, and the fatigue relief can be targeted and effective. At the same time, the evaluation results can also be presented as feedback to the driver to help them intuitively understand the improvement of their fatigue state, and adjust their driving behavior accordingly.

[0178] Further, after the glove receives the instruction to stop the electric stimulation, the portable bracelet continues to detect the physiological signals, and judges whether severe fatigue or physiological abnormalities occur according to the physiological signals, and when it is judged that severe fatigue or physiological abnormalities occur, the system timely warns and prompts.

[0179] Based on the same inventive concept as the intelligent driving fatigue relief method, the present embodiment also provides an intelligent driving fatigue relief device corresponding thereto, Figure 5 The structure block diagram of the intelligent driving fatigue relief device according to the embodiment of the present application is shown, which comprises:

[0180] The signal acquisition module 51 is used for acquiring the multi-channel physiological signals of the driver; the multi-channel physiological signals include pulsatile signals, skin conductance at the wrist, skin temperature, and hand movement state signals;

[0181] The fatigue detection module 52 is configured to input the multi-channel physiological signals and the auxiliary information into a fatigue detection network model to obtain the fatigue degree of the driver. The fatigue detection network model comprises an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer. The input layer is configured to splice the tensor corresponding to the multi-channel physiological signals and the vector corresponding to the auxiliary information to obtain a high-dimensional input representation. The preprocessing layer is configured to preprocess the high-dimensional input representation to obtain a preprocessed feature representation. The hybrid attention encoding layer is configured to perform time domain, frequency domain, and channel three kinds of attention operations on the preprocessed feature representation to obtain a plurality of hybrid attention feature maps of different scales. The global fusion layer is configured to fuse the plurality of hybrid attention feature maps to obtain a fused global feature representation. The output layer is configured to determine the fatigue degree based on the fused global feature representation.

[0182] The instruction sending module 53 is configured to send an electric stimulation instruction to the electric stimulation acupoint glove if the fatigue degree meets a set condition. The electric stimulation acupoint glove performs electric stimulation on the driver according to the electric stimulation instruction to relieve fatigue.

[0183] The intelligent driving fatigue relieving device of the embodiment has the same inventive concept as the intelligent driving fatigue relieving method described above, and therefore the specific embodiments of the device can be seen in the embodiment part of the intelligent driving fatigue relieving method described above, and the technical effects thereof correspond to those of the method described above, which will not be described here again.

[0184] In summary, the present application has the following technical effects:

[0185] 1. An intelligent wearable system integrating fatigue detection, evaluation, and relief is proposed, which ingeniously combines portable biosensing technology with electric stimulation neuroregulation technology, creating a new mode of active fatigue management.

[0186] 2. Aiming at the major traffic safety hazard of driving fatigue, an objective evaluation index system of multi-parameter fusion is constructed from the physiological mechanism, overcoming the limitations of traditional methods.

[0187] 3. A Hybrid-Swin-Transformer deep learning model (HST model) is innovatively proposed, which introduces a hybrid attention mechanism and hierarchical feature representation, achieving significant performance improvement in multi-modal time series analysis tasks and providing a new idea for intelligent discrimination of fatigue state.

[0188] 4. The deep reinforcement learning method is first introduced into the field of electric stimulation parameter optimization, and an adaptive closed-loop control strategy is proposed, realizing real-time feedback and dynamic adjustment of the electric stimulation intervention process, and significantly improving the individual adaptability of fatigue relief effect.

[0189] 5. The application innovatively uses acupoint electrical stimulation to relieve central fatigue, and designs a multi-channel collaborative stimulation scheme, which is more effective than the traditional single stimulation mode, and provides a new research direction for the modernization of fatigue intervention means.

[0190] 6. The application develops a flexible wearable electrode array, combined with pulse current regulation technology, to achieve precise stimulation of human acupoints, which has broad application prospects in the fields of intelligent medical electronics and human-computer interface.

[0191] The above is only various embodiments of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the application, which should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. An intelligent driving fatigue relief method, characterized in that: include: Acquire multi-channel physiological signals of the driver; The multi-channel physiological signals include pulsation signals, wrist skin conductance, skin temperature, and hand movement state signals; Inputting the multi-channel physiological signals and auxiliary information into a fatigue detection network model to obtain the driver's fatigue level; The fatigue detection network model includes an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer; the input layer is used to concatenate the tensor corresponding to the multi-channel physiological signal and the vector corresponding to the auxiliary information to obtain a high-dimensional input representation; the preprocessing layer is used to preprocess the high-dimensional input representation to obtain a preprocessed feature representation; the hybrid attention encoding layer is used to perform three attention operations in the time domain, frequency domain, and channel on the preprocessed feature representation to obtain multiple hybrid attention feature maps of different scales; The global fusion layer is used to fuse the multiple mixed attention feature maps to obtain a fused global feature representation; The output layer is used to determine the fatigue level based on the fused global feature representation; the auxiliary information includes age, height, time, and environment; If the fatigue level meets the set conditions, an electrical stimulation instruction is sent to the electrical stimulation acupoint gloves; the electrical stimulation acupoint gloves electrically stimulate the driver according to the electrical stimulation instruction to relieve fatigue.

2. The method according to claim 1, wherein The hybrid attention encoding layer includes L hybrid attention modules, each of which includes a time domain attention unit, a frequency domain attention unit and a channel attention unit; Each of the hybrid attention modules is used to implement the following functions: Divide the preprocessed features into N local windows, and input each local window into the time domain attention unit, the frequency domain attention unit, and the channel attention unit in parallel to obtain the time domain attention feature, the frequency domain attention feature, and the channel attention feature; Based on the time domain attention feature, the frequency domain attention feature and the channel attention feature, obtaining a mixed attention feature map corresponding to the local window; Shift and concatenate the hybrid attention feature maps corresponding to adjacent local windows in the spatial dimension to obtain the updated feature representation; The updated feature representation is input into the MLP and is residually connected with the preprocessed feature, and then undergoes Layer Normalization to obtain the mixed attention feature map output by the mixed attention module.

3. The method according to claim 1, wherein The global fusion layer is used to implement the following functions: Concatenate multiple hybrid attention feature maps in the channel dimension to obtain a high-dimensional feature representation; The high-dimensional feature representation is obtained by attention pooling to obtain a compact feature vector; The compact feature vector is input into the fully connected layer and mapped to a low-dimensional latent space to obtain a fused global feature representation.

4. The method according to claim 1, wherein The method further comprises: Acquire multi-channel physiological signals of the driver in real time; Calculating a stimulation effect index according to the multi-channel physiological signals; Determine whether the stimulation effect index is within the ideal range. If so, send an instruction to stop electrical stimulation to the electrical stimulation acupoint gloves. If not, send the stimulation effect index to the electrical stimulation acupoint gloves.

5. The method according to claim 1, wherein The method further comprises: Based on the fatigue detection algorithm, the changes in fatigue scores before and after electrical stimulation were compared and the physiological fatigue relief rate was calculated; obtaining the driver's subjective feeling information before and after the electrical stimulation, and calculating the subjective fatigue relief rate based on the subjective feeling information; Analyze driving behavior data, compare changes before and after electrical stimulation, and calculate driving fatigue relief rate; determining an electrical stimulation effect according to the physiological fatigue relief rate, the subjective fatigue relief rate, and the driving fatigue relief rate; The electrical stimulation effect is sent to the electrical stimulation acupoint glove.

6. An intelligent driving fatigue relief device, characterized in that: include: A signal acquisition module, used to acquire the driver's multi-channel physiological signals; The multi-channel physiological signals include pulsation signals, wrist skin conductance, skin temperature, and hand movement state signals; a fatigue detection module, configured to input the multi-channel physiological signals and auxiliary information into a fatigue detection network model to obtain the driver's fatigue level; The fatigue detection network model includes an input layer, a preprocessing layer, a hybrid attention encoding layer, a global fusion layer, and an output layer; the input layer is used to concatenate the tensor corresponding to the multi-channel physiological signal and the vector corresponding to the auxiliary information to obtain a high-dimensional input representation; the preprocessing layer is used to preprocess the high-dimensional input representation to obtain a preprocessed feature representation; the hybrid attention encoding layer is used to perform three attention operations in the time domain, frequency domain, and channel on the preprocessed feature representation to obtain multiple hybrid attention feature maps of different scales; The global fusion layer is used to fuse the multiple mixed attention feature maps to obtain a fused global feature representation; The output layer is used to determine the fatigue level based on the fused global feature representation; the auxiliary information includes age, height, time, and environment; The instruction sending module is used to send an electrical stimulation instruction to the electrical stimulation acupoint gloves if the fatigue level meets the set conditions; the electrical stimulation acupoint gloves electrically stimulate the driver according to the electrical stimulation instruction to relieve fatigue.

7. An intelligent driving fatigue relief system, characterized in that: include: Portable wristbands, electrical acupoint stimulation gloves and mobile devices; The portable wristband is used to collect multi-channel physiological signals of the driver; The multi-channel physiological signals include pulsation signals, wrist skin conductance, skin temperature, and hand movement state signals; The electrical acupoint stimulation gloves are used to electrically stimulate the driver according to the received electrical stimulation instructions to relieve fatigue; The mobile device is used to implement the intelligent driving fatigue relief method described in any one of claims 1 to 5.

8. The system according to claim 7, wherein: The portable bracelet is provided with a PPG sensor, an EDA sensor, a skin temperature sensor and a motion sensor; the PPG sensor is used to collect pulsation signals, the EDA sensor is used to collect skin conductance at the wrist, the skin temperature sensor is used to collect skin temperature, and the motion sensor is used to collect hand motion state signals.

9. The system according to claim 7, wherein: The electrical acupoint stimulation gloves are also used for: Determining optimal electrical stimulation parameters based on a deep reinforcement learning algorithm; the electrical stimulation parameters include current intensity, frequency, and pulse width; and using the received electrical stimulation effect as a reward function in the deep reinforcement learning algorithm; Electrical stimulation is applied to the target acupuncture point according to the optimal electrical stimulation parameters and the fatigue level.

10. The system according to claim 7, wherein: The electrical acupoint stimulation gloves are also used for: Adjust the electrical stimulation parameters according to the received stimulation effect index.

11. The system according to claim 7, wherein: The electrical stimulation acupoint gloves also include a flexible electrode array, an electrical stimulation generator, and a biofeedback sensor; the flexible electrode array is used to apply current to acupoints; the electrical stimulation generator is used to generate electrical stimulation current according to the received electrical stimulation instructions; the biofeedback sensor is used to monitor the contact status between the electrode and the skin and local temperature changes.

12. The system according to claim 7, wherein: The portable wristband and the electrical acupoint stimulation gloves are both provided with wireless communication modules for achieving synchronization between the portable wristband and the electrical acupoint stimulation gloves.

Images