Method for monitoring fatigue degree of operating personnel of individual soldier communication rescue system
By embedded the extreme learning machine model in the individual communication rescue system, using EEG signal data to classify and alert fatigue degrees, the problem of difficulty in quickly, accurately and comprehensively monitoring the fatigue degree of operators in the existing technology is solved, and the problem of rapid, accurate and comprehensive monitoring of the fatigue degree of operators is achieved. Efficient monitoring and warning of fatigue status of workers.
Patent Information
- Application Number
- CN202411653259.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-02-18
Smart Images
Figure CN119454051A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of vital sign monitoring of a single-soldier communication rescue system, in particular to a method for monitoring fatigue level of operators of a single-soldier communication rescue system. Background Art
[0002] The individual soldier communication rescue system is mainly composed of an embedded individual soldier system, a wireless network connection system and a command platform system. First, the individual soldier subsystem uses its rich sensors to collect various data, including audio and video data, human vital signs data, ambient temperature data, toxic and harmful and flammable and explosive gas concentration data, and GPS location data, and then sends these data to the command platform through radio, WIFI, 4G and other wireless networks; the command platform obtains various data sent by the individual soldier through the wireless network, displays or plays, saves, and feedbacks these data, and displays the location of the command platform and the individual soldier on the map. The command platform can replay the saved data. The command platform can send text data, graphic image data, and voice data to the individual soldier.
[0003] With the rapid development of mobile Internet technology, foreign individual soldier vital signs intelligent systems have been put into service. How to improve the intelligence level of our army's individual soldier vital signs equipment is a huge challenge and new topic. Summary of the invention
[0004] The technical problem to be solved by the present invention is to address the deficiencies in the prior art and to provide a method for monitoring the fatigue level of operators in an individual communication rescue system that can monitor the fatigue level of operators through wireless collection of electroencephalogram signals, thereby achieving fast, accurate and comprehensive monitoring and alarm functions for the fatigue level of operators.
[0005] The technical problem to be solved by the present invention is achieved through the following extreme learning machine model. The present invention is a method for monitoring the fatigue level of operators in a single-soldier communication rescue system. The method is embedded in the extreme learning machine model module of the single-soldier communication rescue system and comprises the following steps:
[0006] (1) Obtaining EEG signal data at each time step;
[0007] (2) An extreme learning machine model with fast training speed and high classification accuracy is used to classify the fatigue degree of the selected EEG signal feature values, monitor the fatigue degree of the operators and issue an alarm;
[0008] Among them, the classic power spectrum is used as the method to generate EEG signal features, and the time domain signal is converted into the frequency domain using fast Fourier transform, and it is divided into four frequency bands: δ (0.5Hz~4Hz), θ (4Hz~8Hz), α (8Hz~13Hz), and β (13Hz~30Hz). The average spectral energy of these four frequency bands and their spectral energy ratios (θ+α) / β and β / α are calculated respectively, which are used as features to characterize the degree of fatigue.
[0009] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for monitoring the fatigue degree of operators of the individual communication rescue system, in step (2), the process of establishing the extreme learning machine model is specifically as follows:
[0010] Collect EEG signal data in four frequency bands: δ (0.5 Hz ~ 4 Hz), θ (4 Hz ~ 8 Hz), α (8 Hz ~ 13 Hz), and β (13 Hz ~ 30 Hz) in advance;
[0011] Simulate the data collected during the rescue process of a single soldier, perform EEG frequency domain analysis and calculation on the acquired EEG signals, generate (θ+α) / β, β / α features representing the degree of fatigue, and form an input sequence;
[0012] The data is labeled into two states, namely high alertness and fatigue, to form an output sequence;
[0013] An extreme learning machine model with input sequence as input and output sequence as output is established. Based on the model and collected EEG signal data, extreme learning machine fatigue alarm is realized.
[0014] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for monitoring the fatigue degree of operators of the individual communication rescue system, in step (2), the power spectrum is calculated by using the EEG frequency domain analysis method, specifically:
[0015] (2.1) Fourier transform
[0016] Fourier transform can transform the original time domain signal to the frequency domain for analysis. The collected EEG signal is a series of discrete signals. The Fourier transform of the discrete time series x(1), x(2), ..., x(N)) is expressed as:
[0017]
[0018] in:
[0019] F(ω) is a function in the frequency domain, which represents the representation of the original function x(n) in the frequency domain;
[0020] x(n) is a function in the time domain (or space domain), which represents the original function to be transformed;
[0021] ω is the angular frequency, which represents the variable in the frequency domain;
[0022] j is an imaginary unit, satisfying j^2=-1;
[0023] ∫ represents the integral operation, which integrates the time t;
[0024] (2.2) Power spectral density
[0025] The power spectral density is a relationship between the power spectral density and the frequency value, which can be expressed as a curve graph of amplitude and frequency. R(k) represents the autocorrelation function of a random signal x(n), and the power spectral density function can be expressed as:
[0026]
[0027] where R(k) = E[x(n)x * (n+k)].
[0028] The power spectral density function can be defined as:
[0029]
[0030] When the autocorrelation function satisfies The above two equations are equivalent. The power spectrum density can be used to directly observe the distribution of signal power with frequency.
[0031] (2.3) Power spectrum estimation
[0032] The periodogram method of classical power spectrum estimation is used. The collected EEG signal is firstly a digital signal obtained by discretization. When calculating the power spectrum, the discrete Fourier transform is first used to convert the time domain signal into a frequency domain signal, that is, the signal x(n) is subjected to FFT transformation, and then its amplitude is squared. After dividing by N, the power spectrum estimation of the signal x(n) is obtained:
[0033]
[0034]
[0035] Therefore, specifically for a specific frequency band f1 to f2, the energy between the two can be expressed as:
[0036]
[0037] The power spectrum studies the signal from the perspective of energy, and describes the distribution of signal power in different frequency bands. It is simple and convenient to use the power spectrum of EEG signals to characterize the fatigue state, and the distribution and changes of each rhythm in the EEG signal can be directly observed. In the analysis and research of mental fatigue, the characteristic quantities related to the power spectrum, such as the band energy, the normalized band energy, and the band energy ratio, are widely used in many studies. The power spectrum is calculated using the fast Fourier transform, and these results are used as the main features of the EEG signal.
[0038] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for monitoring the fatigue degree of operators of the individual communication rescue system, in step (2), an extreme learning machine model is established, specifically:
[0039] Given a training sample:
[0040] P={(x i ,y)|x i ∈R d ,y∈R,i=1,2,…,n}x i =[x i1 ,x i2 ,…,x im ]∈Ry=[y1,y2,…,y m ]∈R
[0041] The network structure diagram of the extreme learning machine classification model consists of an input layer, a hidden layer, and an output layer: the input layer has n neurons, corresponding to n input variables; the hidden layer has a nodes; let the output layer be 1 output variable, and both the input and output variables have m sets of data.
[0042] The network structure of the extreme learning machine is:
[0043] Assume w i =[w i1 ,w i2 ,…,w ia ] T is the connection threshold between the ith neuron in the input layer and each neuron in the hidden layer; let b = [b1, b2, ..., b a ] T is the threshold of the hidden layer neuron, and the connection weight between the hidden layer neuron and the output neuron is β k =[β1,β2,…,β a ] T , g(x) is the activation function g(x) of the hidden layer neuron. Then the decision function of the extreme learning machine model with a hidden layer nodes can be expressed as:
[0044]
[0045] This formula can also be expressed as: Hβ=T', where H is the output matrix of the hidden layer of the neural network. The specific form is as follows:
[0046]
[0047] Among them, two SLFN theorems were given by Huang et al., proving that the input weight w of the extreme learning machine model and the threshold b of the hidden layer neurons can be randomly selected.
[0048] Given an activation function that is infinitely differentiable in any interval, when the number of neurons in the hidden layer is equal to the number of training samples, for any input weights and thresholds, the network output can approximate the training samples with zero error, that is,
[0049]
[0050] Given an activation function that is infinitely differentiable in any interval, when the number of training samples is large, in order to reduce the amount of calculation, the number of neurons in the hidden layer K is usually set to be less than the number of samples, but for any input weights and thresholds, there is still K such that the training error of the network output can approach an arbitrary ε>0 (ε is an arbitrarily small error value), that is,
[0051]
[0052] In summary, when g(x) is infinitely differentiable, there is no need to adjust all the parameters of the extreme learning machine, that is, the input connection weights and hidden layer node thresholds can be randomly selected at the beginning of training and remain unchanged during the training process. It is only necessary to set the number of hidden layer nodes and the activation function before training, and β can be obtained by solving the least squares solution:
[0053]
[0054] The solution β is unique, which can minimize the training error of the network and make it have better generalization performance.
[0055] The activation function selected in this article is the Sigmoid function:
[0056]
[0057] Among them, a and b are uniformly distributed between [0, 1].
[0058] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for monitoring the fatigue degree of operators of the individual communication rescue system, in step (2), the extreme learning machine fatigue alarm is specifically:
[0059] Before the extreme learning machine is trained, the connection weights between the input layer neurons and the hidden layer neurons and the thresholds of the hidden layer neurons can be randomly generated. The connection weights between the hidden layer and the output layer can be calculated by simply setting the number of hidden layer neurons and an infinitely differentiable activation function of the hidden layer neurons.
[0060] The test data comes from the data collected during the simulation of a single-soldier rescue. The data is marked into two states, namely high alertness and fatigue. During the test, different features of the same channel are selected on the mobile phone for classification. 100 groups of training data, 20 groups of test data, and 10 hidden layer nodes are randomly selected. The results show that the classification accuracy of different algorithm combinations is different. Among them, the tangent rates of the four feature classifications of α, β, (θ+α) / β, and β / α are relatively high, indicating that these parameters also change significantly when the fatigue state changes.
[0061] Compared with the prior art, the method for monitoring the fatigue degree of operators through a single-soldier communication rescue system with wireless EEG signal acquisition is implemented by establishing an extreme learning machine model, collecting EEG spectrum signal changes according to an EEG signal wireless acquisition device, and utilizing fast Fourier transform to extract the average spectrum energy of each part from the four frequency bands of δ (0.5 Hz to 4 Hz), θ (4 Hz to 8 Hz), α (8 Hz to 13 Hz), and β (13 Hz to 30 Hz) as well as the ratios of their spectrum energies (θ+α) / β and β / α as features characterizing the fatigue degree, thereby realizing fast, accurate and comprehensive operator fatigue degree monitoring and alarm functions. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 is a system flow chart of the present invention;
[0063] Figure 2 is a flow chart of the method of the present invention;
[0064] Figure 3 is a flow chart of the establishment process of the extreme learning machine of the present invention;
[0065] Figure 4 This is a network structure diagram of the extreme learning machine of the present invention. DETAILED DESCRIPTION
[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0067] A method for monitoring fatigue of personnel in a single-soldier communication rescue system, wherein the fatigue of personnel is monitored by a single-soldier communication rescue system with wireless collection of EEG signals. The system flow chart of the present invention is as follows: Figure 1 As shown, the following steps are included:
[0068] (1) Obtaining the EEG signal data of the operator at each time step;
[0069] Specifically: the individual soldier communication rescue system uses its rich sensors to collect various data, including audio and video data, vital signs data of human workers, environmental temperature data, toxic and harmful and flammable and explosive gas concentration data, GPS location data and brain wave signals; these data are sent to the command platform through radio, WIFI, 4G and other wireless networks; the command platform obtains various data sent by individual soldiers through wireless networks, displays or plays, saves and feedbacks these data, and displays the location of the command platform and individual soldiers on the map;
[0070] (2) The extreme learning machine model is used to monitor the fatigue level of the operators and issue alarms based on the acquired EEG signal data. The classic power spectrum is used as the method to generate EEG signal features. The time domain signal is converted into the frequency domain using fast Fourier transform and divided into four frequency bands: δ (0.5 Hz to 4 Hz), θ (4 Hz to 8 Hz), α (8 Hz to 13 Hz), and β (13 Hz to 30 Hz). The average spectral energy of these four frequency bands and their spectral energy ratios (θ+α) / β and β / α are calculated respectively. These are used as features to characterize the fatigue level, and then different levels of alarm prompts are issued to the operators. The alarm information is uploaded to the command center, thus achieving the purpose of monitoring the fatigue level of operators in the individual communication rescue system.
[0071] A method for monitoring fatigue level of personnel in a single-soldier communication rescue system, embedded in Figure 1 In the extreme learning machine model module, the specific method flow chart is as follows Figure 2 As shown, the following steps are included:
[0072] Collect EEG signal data in four frequency bands: δ (0.5 Hz ~ 4 Hz), θ (4 Hz ~ 8 Hz), α (8 Hz ~ 13 Hz), and β (13 Hz ~ 30 Hz) in advance;
[0073] Simulate the data collected during the rescue process of a single soldier, perform EEG frequency domain analysis and calculation on the acquired EEG signal data, generate (θ+α) / β, β / α features representing the degree of fatigue, and form an input sequence;
[0074] The data is labeled into two states, namely high alertness and fatigue, to form an output sequence;
[0075] An extreme learning machine model with input sequence as input and output sequence as output is established, and fatigue alarm of the extreme learning machine is performed based on the model and collected EEG signal data.
[0076] Among them, the spectrum analysis calculation adopts the EEG frequency domain analysis method, specifically:
[0077] (1) Fourier transform
[0078] Fourier transform can transform the original time domain signal to the frequency domain for analysis. The collected EEG signal is a series of discrete signals. The Fourier transform of the discrete time series x(1), x(2), ..., x(N)) is expressed as:
[0079]
[0080] in:
[0081] F(ω) is a function in the frequency domain, which represents the representation of the original function x(n) in the frequency domain;
[0082] x(n) is a function in the time domain (or space domain), which represents the original function to be transformed;
[0083] ω is the angular frequency, which represents the variable in the frequency domain;
[0084] j is an imaginary unit, satisfying j^2=-1;
[0085] ∫ represents the integration operation, which integrates the time t.
[0086] (2) Power spectral density
[0087] The power spectral density is a relationship between the power spectral density and the frequency value, which can be expressed as a curve graph of amplitude and frequency. R(k) represents the autocorrelation function of a random signal x(n), and the power spectral density function can be expressed as:
[0088]
[0089] where R(k) = E[x(n)x * (n+k)].
[0090] The power spectral density function can be defined as:
[0091]
[0092] When the autocorrelation function satisfies The above two equations are equivalent. The power spectrum density can be used to directly observe the distribution of signal power with frequency.
[0093] (3) Power spectrum estimation
[0094] The periodogram method of classical power spectrum estimation is used. The collected EEG signal is firstly a digital signal obtained by discretization. When calculating the power spectrum, the discrete Fourier transform is first used to convert the time domain signal into a frequency domain signal, that is, the signal x(n) is subjected to FFT transformation, and then its amplitude is squared. After dividing by N, the power spectrum estimation of the signal x(n) is obtained:
[0095]
[0096]
[0097] Therefore, specifically for a specific frequency band f1 to f2, the energy between the two can be expressed as:
[0098]
[0099] The power spectrum studies the signal from the perspective of energy, and describes the distribution of signal power in different frequency bands. It is simple and convenient to use the power spectrum of EEG signals to characterize the fatigue state, and the distribution and changes of each rhythm in the EEG signal can be directly observed. In the analysis and research of mental fatigue, the characteristic quantities related to the power spectrum, such as the band energy, the normalized band energy, and the band energy ratio, are widely used in many studies. The power spectrum is calculated using the fast Fourier transform, and these results are used as the main features of the EEG signal.
[0100] like Figure 3 As shown in Figure 1, the establishment process of the extreme learning machine model is as follows:
[0101] Step 1: Build an Extreme Learning Machine Model
[0102] Given a training sample:
[0103] P={(x i ,y)|x i ∈R d ,y∈R,i=1,2,…,n}x i =[x i1 ,x i2 ,…,x im ]∈Ry=[y1,y2,…,y m ]∈R
[0104] The network structure diagram of the extreme learning machine classification model consists of an input layer, a hidden layer, and an output layer: the input layer has n neurons, corresponding to n input variables; the hidden layer has a nodes; let the output layer be 1 output variable, and both the input and output variables have m sets of data.
[0105] The network structure of the extreme learning machine is as follows Figure 4 As shown;
[0106] Assume w i =[w i1 ,w i2 ,…,w ia ] T is the connection threshold between the ith neuron in the input layer and each neuron in the hidden layer; let b = [b1, b2, ..., b a ] T is the threshold of the hidden layer neuron, and the connection weight between the hidden layer neuron and the output neuron is β k =[β1,β2,…,β a ] T , g(x) is the activation function g(x) of the hidden layer neuron. Then the decision function of the extreme learning machine model with a hidden layer nodes can be expressed as:
[0107]
[0108] This formula can also be expressed as: Hβ=T', where H is the output matrix of the hidden layer of the neural network. The specific form is as follows:
[0109]
[0110] Among them, two SLFN theorems were given by Huang et al., proving that the input weight w of the extreme learning machine model and the threshold b of the hidden layer neurons can be randomly selected.
[0111] Given an activation function that is infinitely differentiable in any interval, when the number of neurons in the hidden layer is equal to the number of training samples, for any input weights and thresholds, the network output can approximate the training samples with zero error, that is,
[0112]
[0113] Given an activation function that is infinitely differentiable in any interval, when the number of training samples is large, in order to reduce the amount of calculation, the number of neurons in the hidden layer K is usually set to be less than the number of samples, but for any input weights and thresholds, there is still K such that the training error of the network output can approach an arbitrary ε>0 (ε is an arbitrarily small error value), that is,
[0114]
[0115] In summary, when g(x) is infinitely differentiable, there is no need to adjust all the parameters of the extreme learning machine, that is, the input connection weights and hidden layer node thresholds can be randomly selected at the beginning of training and remain unchanged during the training process. It is only necessary to set the number of hidden layer nodes and the activation function before training, and β can be obtained by solving the least squares solution:
[0116]
[0117] The solution β is unique, which can minimize the training error of the network and make it have better generalization performance.
[0118] The activation function selected in this article is the Sigmoid function:
[0119]
[0120] Among them, a and b are uniformly distributed between [0, 1].
[0121] Step 2: Extreme Learning Machine Fatigue Alarm
[0122] Based on the above model, an EEG signal acquisition sensor can be installed on the head, and the preprocessed feature vector data is used as the input of the extreme learning machine model. The selected feature vectors are classified using the extreme learning machine algorithm with a faster training speed. The data is marked as two states, namely high alertness and fatigue. Before the extreme learning machine is trained, the connection weights between the input layer neurons and the hidden layer neurons and the thresholds of the hidden layer neurons can be randomly generated. Only the number of hidden layer neurons and an infinitely differentiable activation function of the hidden layer neurons need to be set to calculate the connection weights between the hidden layer and the output layer. During the test, the system selects different features of the same channel for classification. 100 groups of training data, 20 groups of test data, and 10 hidden layer nodes are randomly selected. The results show that the classification accuracy of different algorithm combinations is different.
[0123] From the above content, it is not difficult to find that the method for monitoring the fatigue degree of operators through the individual communication rescue system based on wireless collection of EEG signals, on the one hand, according to the changes in EEG signals, the average spectral energy of each part and the ratio of its spectral energy (θ+α) / β, β / α are extracted as features to characterize the fatigue degree, among which α, β, (θ+α) / β, β / α, the four feature classifications have a high accuracy rate, indicating that when the fatigue state changes, the eigenvalue parameters change accordingly, and there is a strong correlation between the fatigue state and the eigenvalue, which further shows the effectiveness of EEG signal preprocessing and eigenvalue selection; on the other hand, by establishing an extreme learning machine model, the extreme learning machine (ELM) algorithm with fast training speed, good generalization ability and high accuracy is used to classify the fatigue degree of the selected eigenvalues, thereby improving the recognition accuracy of the fatigue degree and realizing fast, accurate and comprehensive recognition of the life status of operators.
Claims
1. A method for monitoring fatigue level of personnel in a single-soldier communication rescue system, characterized in that: The method comprises the following steps: (1) Obtaining EEG signal data at each time step; (2) An extreme learning machine model with fast training speed and high classification accuracy is used to classify the fatigue degree of the selected EEG signal feature values, monitor the fatigue degree of the operators and issue an alarm; Among them, the classic power spectrum is used as the method to generate EEG signal features, and the fast Fourier transform is used to convert the time domain signal into the frequency domain, and it is divided into four frequency bands: δ, θ, α, and β. The average spectral energy of these four frequency bands and the ratios of their spectral energies (θ+α) / β and β / α are calculated respectively, which are used as features to characterize the degree of fatigue. The frequency band of δ is 0.5Hz~4Hz, the frequency band of θ is 4Hz~8Hz, the frequency band of α is 8Hz~13Hz, and the frequency band of β is 13Hz~30Hz.
2. The method for monitoring fatigue level of personnel in a single-soldier communication rescue system according to claim 1 is characterized in that: In step (2), the process of establishing the extreme learning machine model is: Collect EEG signal data of four frequency bands: δ, θ, α, and β in advance; Simulate the data collected during the rescue process of a single soldier, perform EEG frequency domain analysis and calculation on the acquired EEG signals, generate (θ+α) / β, β / α features representing the degree of fatigue, and form an input sequence; The data is labeled into two states, namely high alertness and fatigue, to form an output sequence; An extreme learning machine model with input sequence as input and output sequence as output is established, and fatigue alarm of the extreme learning machine is performed based on the model and collected EEG signal data.
3. The method for monitoring fatigue level of personnel in a single-soldier communication rescue system according to claim 2 is characterized in that: In step (2), the EEG frequency domain analysis method is used to perform spectrum analysis calculation, specifically: (2.1) Fourier transform Fourier transform can transform the original time domain signal to the frequency domain for analysis. The collected EEG signal is a series of discrete signals. The Fourier transform of the discrete time series x(1), x(2), ..., x(N)) is expressed as: in: F(ω) is a function in the frequency domain, which represents the representation of the original function x(n) in the frequency domain; x(n) is a function in the time domain or space domain, representing the original function to be transformed; ω is the angular frequency, which represents the variable in the frequency domain; j is an imaginary unit, satisfying j^2=-1; ∫ represents the integral operation, which integrates the time t; (2.2) Power spectral density The power spectrum density is a relationship between the power spectrum density and the frequency value, which is expressed as a curve graph of the amplitude and frequency. R(k) is used to represent the autocorrelation function of a random signal x(n), and the power spectrum density function can be expressed as: Where R(k) = E[x(n)x * (n+k)]; The power spectral density function can be defined as: When the autocorrelation function satisfies When , the above two equations are equivalent, and the power spectrum density can directly observe the distribution of signal power with frequency; (2.3) Power spectrum estimation The periodogram method of classical power spectrum estimation is used. The collected EEG signal is firstly a digital signal obtained by discretization. When calculating the power spectrum, the discrete Fourier transform is first used to convert the time domain signal into a frequency domain signal, that is, the signal x(n) is subjected to FFT transformation, and then its amplitude is squared. After dividing by N, the power spectrum estimation of the signal x(n) is obtained: Therefore, specifically for a specific frequency band f1 to f2, the energy between the two can be expressed as: Fast Fourier transform is used to calculate the power spectrum, and these results are used as the main features of the EEG signal.
4. The method for monitoring fatigue level of personnel in a single-soldier communication rescue system according to claim 2, characterized in that: Establish an extreme learning machine model, specifically: Given a training sample: P={(x i ,y)|x i ∈R d ,y∈R,i=1,2,…,n}x i =[x i1 ,x i2 ,…,x im ]∈Ry=[y1,y2,…,y m ]∈R The network structure diagram of the extreme learning machine classification model consists of an input layer, a hidden layer, and an output layer: the input layer has n neurons, corresponding to n input variables; the hidden layer has a nodes; let the output layer be 1 output variable, and both the input and output variables have m sets of data; The network structure of the extreme learning machine is: Assume w i =[w i1, w i2, …,w ia ] T is the connection threshold between the ith neuron in the input layer and each neuron in the hidden layer; let b = [b1, b2, ..., b a ] T is the threshold of the hidden layer neuron, and the connection weight between the hidden layer neuron and the output neuron is β k =[β1,β2,…,β a ] T , g(x) is the activation function of the hidden layer neurons; then the decision function of the extreme learning machine model with a hidden layer nodes is expressed as: Given an activation function that is infinitely differentiable in any interval, when the number of neurons in the hidden layer is equal to the number of training samples, for any input weights and thresholds, the network output can approximate the training samples with zero error, that is, Given an activation function that is infinitely differentiable in any interval, when the number of training samples is large, in order to reduce the amount of calculation, the number of neurons in the hidden layer K is set to be less than the number of samples, but for any input weights and thresholds, there is still K so that the training error of the network output can approach an arbitrary ε>0, where ε is an arbitrarily small error value, that is, When g(x) is infinitely differentiable, there is no need to adjust all the parameters of the extreme learning machine, that is, the input connection weights and hidden layer node thresholds can be randomly selected at the beginning of training and remain unchanged during the training process. It is only necessary to set the number of hidden layer nodes and the activation function before training, which can be obtained by solving the least squares solution: The solution β is unique, which can minimize the training error of the network and make it have better generalization performance; The selected activation function is the Sigmoid function: Among them, a and b are uniformly distributed between [0, 1].
5. The method for monitoring fatigue level of personnel in a single-soldier communication rescue system according to claim 2, characterized in that: Extreme learning machine fatigue alarm, specifically: Before the extreme learning machine is trained, the connection weights between the input layer neurons and the hidden layer neurons and the thresholds of the hidden layer neurons can be randomly generated. The connection weights between the hidden layer and the output layer can be calculated by simply setting the number of hidden layer neurons and an infinitely differentiable activation function of the hidden layer neurons. The test data comes from the data collected during the simulation of a single-soldier rescue. The data is marked into two states, namely high alertness and fatigue. During the test, the mobile phone selects different features of the same channel for classification. 100 groups of training data, 20 groups of test data, and 10 hidden layer nodes are randomly selected. The results show that the classification accuracy of different algorithm combinations is different; Among them, the tangent rates of the four feature classifications of α, β, (θ+α) / β, and β / α are relatively high, indicating that these parameters also change significantly when the fatigue state changes.