Fatigue recognition method and fatigue recognition device based on forehead three-lead electroencephalogram signals

Abstract

The invention provides a fatigue recognition method and device based on forehead three-lead electroencephalogram signals. The fatigue recognition method comprises the steps that the forehead three-lead electroencephalogram signals of safety production personnel are obtained in real time; the forehead three-lead electroencephalogram signals are input into an electroencephalogram fatigue recognition model, the electroencephalogram fatigue recognition model is used for analyzing the forehead three-lead electroencephalogram signals, a fatigue state label value is obtained through output, and the fatigue state label value is used for representing the fatigue degree of the safety production personnel; under the condition that the fatigue state label value is larger than the fatigue early warning value, prompt information is sent out, and the prompt information is used for reminding corresponding safety production personnel to stop working. The problem that in the prior art, the cross-individual recognition effect is poor, and especially under the influence of individual differences, signal quality and task changes, an existing model is difficult to achieve efficient and stable fatigue recognition is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fatigue state recognition, and in particular to a fatigue recognition method based on forehead three-lead electroencephalogram signals, a fatigue recognition device, a computer-readable storage medium and a computer program product. Background Art

[0002] Fatigue state recognition has important application value in fields such as safe production operations and long-term monitoring. With the emphasis on safe production and work efficiency in modern society, fatigue recognition has become an important research topic. Traditional fatigue recognition methods usually rely on behavioral indicators and subjective reports, which have problems of poor real-time performance and strong subjectivity. For example, fatigue detection methods based on eye tracking and facial expression recognition, although they can provide certain reference information, have low recognition accuracy and robustness. In addition, subjective fatigue assessment methods based on self-reports are greatly affected by individual subjective factors, making it difficult to achieve real-time monitoring and accurate assessment.

[0003] As a bioelectric signal that directly reflects brain activity, electroencephalogram (EEG) has become an important means of fatigue recognition due to its objectivity and high temporal resolution. EEG signals contain rich information about brain activity and can reflect the cognitive and physiological changes of individuals in different states. By analyzing the characteristics of EEG signals, the fatigue state of individuals can be effectively identified. However, the practical application of EEG signal fatigue recognition is often limited by the complexity of multi-channel systems. Although EEG signals extracted from the frontal channel are more practical and portable, they may have higher noise and poorer signal quality compared to full-hat settings. Individual differences, signal quality, and task changes can significantly affect the performance of the recognition model, resulting in unstable recognition results. Summary of the invention

[0004] The main purpose of the present application is to provide a fatigue recognition method, a fatigue recognition device, a computer-readable storage medium and a computer program product based on three-lead forehead EEG signals, so as to at least solve the problem that the cross-individual recognition effect in the prior art is poor, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a fatigue recognition method based on three-lead forehead EEG signals is provided, comprising: acquiring three-lead forehead EEG signals of safety production personnel in real time; inputting the three-lead forehead EEG signals into an EEG fatigue recognition model, and using the EEG fatigue recognition model to analyze the three-lead forehead EEG signals, and outputting a fatigue state label value, wherein the EEG fatigue recognition model is a convergent model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, and each set of the training data includes sample three-lead forehead EEG data and sample The fatigue state label value, the EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the point convolution layer is used to extract the spatial domain features in the three-lead forehead EEG signal by using point convolution, the deep convolution layer is used to extract the time domain features in the three-lead forehead EEG signal by using deep convolution, the fatigue state label value is used to indicate the fatigue degree of the safety production personnel; when the fatigue state label value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding safety production personnel to suspend work.

[0006] Optionally, before inputting the three-lead forehead EEG signal into an EEG fatigue recognition model to analyze the three-lead forehead EEG signal using the EEG fatigue recognition model and outputting a fatigue state label value, the method further includes: obtaining original three-lead forehead EEG signals corresponding to multiple subjects, and recording the event occurrence time, reaction start time, and lane return center time of the corresponding subjects when a lane departure event occurs in an EEG data acquisition experiment, and aligning them with the EEG data timeline, wherein the EEG data acquisition experiment is an experiment in which a lane departure event is randomly triggered on the subject after the subject wears a helmet to obtain the subject's reaction time to the randomly occurring lane departure event; marking the corresponding original three-lead forehead EEG signal according to a local reaction time, wherein the local reaction time is the time interval between the event occurrence time and the reaction start time of the lane departure event, and the local reaction time is used to evaluate the fatigue level of the subject; performing a preprocessing operation on the original three-lead forehead EEG signal to obtain the training data, wherein the preprocessing operation at least includes bandpass filtering, independent component analysis, and baseline correction.

[0007] Optionally, before the three-lead forehead EEG signal is input into the EEG fatigue recognition model to analyze the three-lead forehead EEG signal using the EEG fatigue recognition model and output a fatigue state label value, the method further includes: a spatial processing step, using a plurality of point convolution nodes in the point convolution layer of the initial EEG fatigue recognition model to generate the same number of new signal channels, and using each of the point convolution nodes to perform spatial feature processing on the sample three-lead forehead EEG signal, and outputting the spatial domain feature through the corresponding new signal channel; a time processing step, in the initial In the deep convolution layer of the EEG fatigue recognition model, multiple deep convolution nodes are used to perform time feature processing on the spatial domain features to obtain the time domain features, and the number of the deep convolution nodes is twice the number of the point convolution nodes; in the feature processing step, the spatial domain features and the time domain features are sequentially input into the activation layer and the batch normalization layer of the initial EEG fatigue recognition model for feature processing to obtain a normalized feature map; in the attention weighting step, the normalized feature map is input into the attention mechanism layer of the initial EEG fatigue recognition model for attention weighting processing to obtain weighted features; a refinement step, inputting the weighted features into the point convolution layer of the initial EEG fatigue recognition model for refinement processing to obtain refined features, wherein the refinement processing is used to refine the correlation between different channels; a residual step, inputting the weighted features and the refined features into the activation layer and the attention mechanism layer in the form of residual connection in turn to obtain effective features; an output step, inputting the effective features into the output layer, and outputting the training fatigue state label value; a verification step, verifying the initial EEG fatigue recognition model according to the sample fatigue state label value to obtain model accuracy, wherein the model accuracy is used to represent the accuracy of the initial EEG fatigue recognition model; a repetition step, when the model accuracy does not meet the accuracy requirements, optimizing and adjusting the parameters of the initial EEG fatigue recognition model on the verification set, and re-executing the spatial processing step, the temporal processing step, the feature processing step, the attention weighting step, the refinement step, the residual step, the output step, the verification step and the repetition step in turn at least once until the model accuracy meets the accuracy requirements and the EEG fatigue recognition model is obtained.

[0008] Optionally, using each of the point convolution nodes to perform spatial feature processing on the sample forehead three-lead EEG signal, and outputting the spatial domain feature through the corresponding new signal channel, includes: calculating the spatial domain feature according to the weight and bias parameters of each of the point convolution nodes in combination with a first formula, and outputting the spatial domain feature through the corresponding new signal channel, wherein the first formula is in, represents the spatial characteristics, x p,jis the jth sampling point of the pth input channel of the sample three-lead forehead EEG signal, represents the weight corresponding to the p-th input channel of the i-th point convolution node, represents the bias parameter of the i-th point convolution node, i represents the number of the point convolution node, 1≤i≤N1, and m represents the total number of input channels.

[0009] Optionally, in the deep convolution layer of the initial EEG fatigue recognition model, multiple deep convolution nodes are used to perform time feature processing on the spatial features to obtain the time domain features, including: calculating the time domain features according to the spatial features, the weights and bias parameters of the deep convolution nodes combined with a second formula, wherein the second formula is represents the time domain feature, represents the weight of the kth depth convolution node when the convolution kernel length is r, k represents the number of the depth convolution node, 1≤k≤2N1, represents the bias parameter of the kth depth convolution node, Indicates The spatial domain features of the deep convolution nodes at the j+r-1th sampling point, Indicates The spatial domain features of the deep convolution nodes at the j+r-1th sampling point.

[0010] Optionally, the normalized feature map is input into the attention mechanism layer of the initial EEG fatigue recognition model for recalibration to obtain calibrated features, including: using a global average pooling operation to compress each of the new signal channels of the normalized feature map into a corresponding scalar; recalculating the channel weight of each new signal channel through the fully connected layer; applying the calculated channel weight to each new signal channel corresponding to the normalized feature map, and adjusting the channel response of the normalized feature map by element-by-element multiplication to obtain the calibrated features.

[0011] Optionally, optimizing and adjusting the parameters of the initial EEG fatigue recognition model on the validation set includes: using the validation set to verify the initial EEG fatigue recognition model and calculating the loss value of the initial fatigue recognition model according to a loss function, wherein the loss function is θ=θ-η▽ θ L(θ;X,y), where η is the learning rate, ▽ θL is the gradient of the loss function with respect to the parameters, θ represents the loss value, y is the label value of the initial EEG fatigue recognition model in the current iteration, the validation set is a data set divided from the sample data set, and the validation set includes multiple groups of validation data, each group of the validation data includes the sample forehead three-lead EEG data and the sample fatigue state label value; the parameters of the initial EEG fatigue recognition model are optimized and adjusted according to the loss value.

[0012] According to another aspect of the present application, a fatigue recognition device based on a three-lead forehead EEG signal is provided, the device comprising: an acquisition unit, used to acquire the three-lead forehead EEG signals of the safety production personnel in real time; an identification unit, used to input the three-lead forehead EEG signals into an EEG fatigue recognition model, so as to analyze the three-lead forehead EEG signals using the EEG fatigue recognition model and output a fatigue state label value, wherein the EEG fatigue recognition model is a convergent model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, each set of the training data comprising sample three-lead forehead EEG data and The sample fatigue state label value, the EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the point convolution layer is used to extract the spatial domain features in the three-lead forehead EEG signal by using point convolution, the deep convolution layer is used to extract the time domain features in the three-lead forehead EEG signal by using deep convolution, the fatigue state label value is used to indicate the fatigue degree of the safety production personnel; a prompt unit is used to issue a prompt message when the fatigue state label value is greater than the fatigue warning value, and the prompt message is used to remind the corresponding safety production personnel to suspend work.

[0013] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0014] According to another aspect of the present application, a computer program product is provided, comprising computer instructions, wherein when the computer instructions are executed by a processor, any one of the methods described above is implemented.

[0015] By applying the technical solution of the present application, in a fatigue recognition method based on three-lead forehead EEG signals, first, the three-lead forehead EEG signals of the safety production personnel are obtained in real time; then, the three-lead forehead EEG signals are input into an EEG fatigue recognition model, so that the three-lead forehead EEG signals are analyzed by using the EEG fatigue recognition model, and a fatigue state label value is output. The EEG fatigue recognition model is a convergent model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, and each set of the training data includes sample three-lead forehead EEG data and sample fatigue state label values. Signature value, the above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The above-mentioned point convolution layer is used to extract the spatial features in the above-mentioned three-lead EEG signal of the forehead by point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead EEG signal of the forehead by deep convolution. The above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel; finally, when the above-mentioned fatigue state label value is greater than the fatigue warning value, a prompt message is issued, and the above-mentioned prompt message is used to remind the corresponding above-mentioned safety production personnel to suspend work. The present invention obtains an EEG fatigue recognition model by designing a multi-layer convolutional neural network, combining multi-step preprocessing and feature extraction technology training, and realizes high-accuracy fatigue recognition between different individuals by using only the EEG data of the forehead area (i.e., the above-mentioned three-lead EEG data of the forehead). The method of the present invention not only improves the accuracy and robustness of fatigue recognition, but also has good generalization ability, and can be widely used in practical scenarios such as driver monitoring and worker fatigue detection. The present application solves the problem that the existing technology has poor cross-individual recognition effect, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A hardware structure block diagram of a mobile terminal for executing a fatigue identification method based on forehead three-lead EEG signals provided in an embodiment of the present application is shown;

[0017] Figure 2 A schematic flow chart of a fatigue identification method based on forehead three-lead EEG signals provided according to an embodiment of the present application is shown;

[0018] Figure 3 A schematic flow chart of a fatigue identification method based on forehead three-lead EEG signals provided according to an embodiment of the present application is shown;

[0019] Figure 4 A schematic diagram of a framework of an EEG fatigue recognition model provided according to an embodiment of the present application is shown;

[0020] Figure 5A structural block diagram of a fatigue identification device based on forehead three-lead EEG signals provided according to an embodiment of the present application is shown.

[0021] The above drawings include the following reference numerals:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0023] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] As introduced in the background technology, although the EEG signals extracted from the forehead channel in the prior art are more practical and portable, they may have higher noise and worse signal quality compared with the full hat setting. Individual differences, signal quality and task changes will significantly affect the performance of the recognition model, resulting in unstable recognition results. In order to solve the problem that the cross-individual recognition effect in the prior art is poor, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition. The embodiments of the present application provide a fatigue recognition method based on forehead three-lead EEG signals, a fatigue recognition device, a computer-readable storage medium and a computer program product.

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a fatigue identification method based on forehead three-lead EEG signals according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is for illustration only and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0029] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the fatigue identification method based on the forehead three-lead EEG signal in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above network include but are not limited to the Internet, corporate intranets, local area networks, mobile communication networks and combinations thereof. The transmission device 106 is used to receive or send data via a network. The above network specific examples may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] In this embodiment, a fatigue identification method based on three-lead forehead EEG signals running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] Figure 2 It is a flowchart of a fatigue identification method based on forehead three-lead EEG signals according to an embodiment of the present application.

[0032] like Figure 2 As shown, the method comprises the following steps:

[0033] Step S201, obtaining three-lead forehead EEG signals of production safety personnel in real time;

[0034] Specifically, the three-lead EEG signals of the forehead of the safety production personnel are collected in real time through the EEG equipment. By obtaining the three-lead EEG signals of the forehead in real time, the system can monitor the fatigue state of the safety production personnel in real time and detect fatigue in time.

[0035] Step S202, input the above-mentioned three-lead forehead EEG signal into the EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal using the above-mentioned EEG fatigue recognition model, and output a fatigue state label value. The above-mentioned EEG fatigue recognition model is a convergence model obtained by iterative training by inputting multiple groups of training data into a predetermined model, and each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values. The above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution. The above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel.

[0036] Specifically, the collected three-lead forehead EEG signals are input into a pre-trained EEG fatigue recognition model, such as Figure 3 As shown in the figure, the EEG fatigue recognition model uses convolution kernels of different sizes to extract spatial and temporal features. Two sizes of convolution kernels are used. The first is a size of (ch, 1), where ch refers to the number of input filtered lead channels, which is used to extract spatial features; then a convolution kernel of size (1, 32) is used to extract temporal features. The EEG fatigue recognition model analyzes the input EEG signal and outputs the fatigue status label value of the safety production personnel.

[0037] Step S203: When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0038] Specifically, a prompt message is issued: when the fatigue status tag value is greater than the fatigue warning value, the system automatically issues a prompt message to remind the corresponding safety production personnel to suspend work. When the system detects that the safety production personnel are in a fatigue state, prompt messages are issued in a timely manner, which can effectively reduce the occurrence of work safety accidents caused by fatigue. By promptly reminding fatigued safety production personnel to suspend work, the problem of reduced work efficiency and reduced production quality due to fatigue can be avoided. At the same time, it helps to protect the health of safety production personnel.

[0039] In this embodiment, in the fatigue recognition method based on the three-lead forehead EEG signal, first, the three-lead forehead EEG signal of the safety production personnel is obtained in real time; then, the three-lead forehead EEG signal is input into the EEG fatigue recognition model, so that the three-lead forehead EEG signal is analyzed by using the EEG fatigue recognition model, and the fatigue state label value is output. The EEG fatigue recognition model is a convergent model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, and each group of the training data includes sample three-lead forehead EEG data and sample fatigue state label value. The above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The above-mentioned point convolution layer is used to extract the spatial features in the above-mentioned three-lead EEG signal of the forehead by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead EEG signal of the forehead by using deep convolution. The above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel; finally, when the above-mentioned fatigue state label value is greater than the fatigue warning value, a prompt message is issued, and the above-mentioned prompt message is used to remind the corresponding above-mentioned safety production personnel to suspend work. The present invention obtains an EEG fatigue recognition model by designing a multi-layer convolutional neural network, combining multi-step preprocessing and feature extraction technology training, and realizes high-accuracy fatigue recognition between different individuals when only the EEG data of the forehead area (i.e., the above-mentioned three-lead EEG data of the forehead) is used. The method of the present invention not only improves the accuracy and robustness of fatigue recognition, but also has good generalization ability, and can be widely used in practical scenarios such as driver monitoring and worker fatigue detection. The present application solves the problem that the existing technology has poor cross-individual recognition effect, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0040] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the fatigue recognition method based on three-lead forehead EEG signals of the present application will be described in detail below in combination with specific embodiments.

[0041] In order to ensure the accuracy and reliability of the experimental data, in an optional implementation manner, before the above step S202, the above method further includes:

[0042] Step S301, obtaining original forehead three-lead EEG signals corresponding to multiple subjects, and recording the event occurrence time, reaction start time, lane return center time of the corresponding subjects when the lane departure event occurs in the EEG data acquisition experiment, and aligning them with the EEG data timeline. The EEG data acquisition experiment is a test in which a lane departure event is randomly triggered on the subjects after the subjects wear helmets to obtain the reaction time of the subjects to the randomly occurring lane departure event;

[0043] Step S302, marking the corresponding original forehead three-lead EEG signal according to the local reaction time, wherein the local reaction time is the time interval between the event occurrence time of the lane departure event and the reaction start time, and the local reaction time is used to evaluate the fatigue level of the subject;

[0044] Step S303, preprocessing the original three-lead forehead EEG signal to obtain the training data, wherein the preprocessing operation at least includes bandpass filtering, independent component analysis and baseline correction.

[0045] In the above embodiment, in the raw data acquisition experiment, the subject is required to concentrate as much as possible and respond immediately to the randomly occurring lane departure events. In the experiment, the time when the lane departure event occurs, the reaction occurs, and the lane returns to the center is recorded respectively, and aligned with the EEG data timeline. The local reaction time is used to evaluate the fatigue level, which is defined as the time interval between the start of the event and the start of the reaction of each lane departure event, because a long time interval may mean a slower reaction speed, which may be caused by fatigue. Marking the local reaction time can help researchers better understand the subject's reaction to the lane departure event. In the experiment, EEG data of the frontal area, including three channels of Fp1, Fp2, and Fz (forehead lead electrodes in the 10-20 international standard lead system), were used to downsample the EEG signal to 128Hz, and a 3-second long sample was extracted before each lane departure event to form a vector. Subsequently, the extracted data was preprocessed using physiological knowledge and existing conclusions on fatigue recognition, including bandpass filtering, independent component analysis (ICA) to remove artifacts, and baseline correction. Bandpass filtering is used to remove high-frequency noise and low-frequency drift. Bandpass filtering can remove noise from the signal and improve the quality of the signal; independent component analysis can identify and remove components that are not related to the signal of interest; baseline correction can make the data of different subjects comparable, making subsequent data analysis more accurate. Through the above steps, accurate EEG data can be obtained, and effective preprocessing operations can be performed to provide a reliable data basis for subsequent data analysis and fatigue level assessment. The accuracy and reliability of the experimental data are ensured, and an effective means is provided to assess the fatigue level of the subjects. Finally, the processed data set is cut into a training set, a test set, and a validation set. The training set is used for training the convolutional neural network, the test set is used to evaluate the model effect, and the validation set is used to adjust the model parameters and improve the generalization ability.

[0046] In order to improve the accuracy of the EEG fatigue recognition model, in an optional implementation, before the above step S202, the method further includes:

[0047] Step S401, a spatial processing step, using a plurality of point convolution nodes in the point convolution layer of the initial EEG fatigue recognition model to generate the same number of new signal channels, and using each of the point convolution nodes to perform spatial feature processing on the sample forehead three-lead EEG signal, and outputting the spatial domain features through the corresponding new signal channels;

[0048] Step S402, a time processing step, in which a plurality of deep convolution nodes are used in the deep convolution layer of the initial EEG fatigue recognition model to perform time feature processing on the spatial features to obtain the time domain features, wherein the number of the deep convolution nodes is twice the number of the point convolution nodes;

[0049] Step S403, a feature processing step, inputting the spatial domain features and the time domain features into the activation layer and the batch normalization layer of the initial EEG fatigue recognition model in sequence for feature processing to obtain a normalized feature map;

[0050] Step S404, an attention weighting step, inputting the normalized feature map into the attention mechanism layer of the initial EEG fatigue recognition model for attention weighting processing to obtain weighted features;

[0051] Step S405, a refinement step, performing refinement processing on the weighted features input to the point convolution layer of the initial EEG fatigue recognition model to obtain refined features, wherein the refinement processing is used to refine the correlation between different channels;

[0052] Step S406, residual step, inputting the weighted features and the refined features into the activation layer and the attention mechanism layer in the form of residual connection in sequence to obtain effective features;

[0053] Step S407, an output step, inputting the above-mentioned effective features into the above-mentioned output layer, and outputting a training fatigue state label value;

[0054] Step S408, a verification step, verifying the initial EEG fatigue recognition model according to the sample fatigue state label value to obtain model accuracy, and the model accuracy is used to represent the accuracy of the initial EEG fatigue recognition model;

[0055] Step S409, repeat the steps, when the accuracy of the above-mentioned model does not meet the accuracy requirements, optimize and adjust the parameters of the above-mentioned initial EEG fatigue recognition model on the verification set, and re-execute the above-mentioned spatial processing step, the above-mentioned time processing step, the above-mentioned feature processing step, the above-mentioned attention weighting step, the above-mentioned refinement step, the above-mentioned residual step, the above-mentioned output step, the above-mentioned verification step and the above-mentioned repetition step at least once in sequence, until the accuracy of the above-mentioned model meets the above-mentioned accuracy requirements, and the above-mentioned EEG fatigue recognition model is obtained.

[0056] In the above embodiment, multiple point convolution nodes are used to generate a new signal channel, and then the EEG signal is processed with spatial features, and the spatial features are output through the generated new signal channel. The spatial features of the EEG signal are extracted so that the subsequent processing process can better identify the EEG fatigue state. On the basis of the spatial features, the time features are further extracted to improve the recognition accuracy of the EEG fatigue state. The spatial features and the time domain features are input into the activation layer and the batch normalization layer for feature processing to obtain a normalized feature map. The role of this step is to further process the features through the activation layer and the batch normalization layer, so that the features are more distinguishable and representative. The normalized feature map is input into the attention mechanism layer for attention weighted processing to obtain weighted features, and the features are weighted through the attention mechanism to improve the degree of attention to key features, so as to better identify the EEG fatigue state. The weighted features are input into the point convolution layer for refinement processing to obtain refined features, which are used to refine the correlation between different channels. The role of this step is to refine the features through the point convolution layer to improve the correlation and representativeness of the features. The re-weighted features are then passed through the point convolution layer to further refine the correlation between different channels. Subsequently, the output features of the attention mechanism layer and the point convolution layer are input to the activation layer in the form of residual connections. By linearly combining the information of different channels, the feature representation capability is enhanced. The activated features are once again processed by the attention mechanism layer to weight the features; finally, after operations such as pooling, dropout, full connection, and softmax, the results of fatigue monitoring are output, and the valid features are input to the output layer, and the training fatigue state label value is output. The purpose of this step is to classify the valid features through the output layer to obtain the label value of the training fatigue state. The initial model is verified according to the training fatigue state label value to obtain the model accuracy, which is used to indicate the accuracy of the model. The purpose of this step is to evaluate the recognition accuracy of the model by verifying the model accuracy. If the model accuracy does not meet the requirements, the model parameters will be optimized and adjusted on the verification set, and the above steps will be re-executed until the model accuracy meets the requirements. By continuously optimizing and adjusting the model until the recognition accuracy that meets the requirements is achieved, the final EEG fatigue recognition model is obtained. The framework diagram of the EEG fatigue recognition model is shown in the figure. Figure 4 shown.

[0057] In order to improve the richness of data and promote model convergence from the perspective of spatial features, in an optional implementation, the above step S401 includes:

[0058] Step S4011, calculate the above-mentioned spatial domain features according to the weight and bias parameters of each of the above-mentioned point convolution nodes in combination with the first formula, and output the above-mentioned spatial domain features through the corresponding above-mentioned new signal channel. The above-mentioned first formula is in, represents the above spatial characteristics, x p,jis the jth sampling point of the pth input channel of the above sample three-lead forehead EEG signal, represents the weight corresponding to the p-th input channel of the i-th point convolution node, Represents the bias parameter of the i-th point convolution node, i represents the number of the point convolution node, 1≤i≤N1, and m represents the total number of input channels.

[0059] In the above embodiment, N1 point convolution nodes are first used to generate N1 new signal channels of the same length. For a given input sample X (m，n) , which contains m one-dimensional signals with input channel length n. The i-th (1≤i≤N1) node has m weights and 1 bias parameter Finally, the spatial features are extracted N1 is set to 8, which is approximately twice the number of input channels, to improve data richness and promote model convergence.

[0060] In order to improve the richness of data and promote model convergence from the perspective of time characteristics, in an optional implementation, the above step S402 includes:

[0061] Step S4021, calculate the time domain feature according to the spatial feature, the weight and bias parameter of the deep convolution node and the second formula, the second formula is: Represents the above time domain characteristics, represents the weight of the kth depth convolution node when the convolution kernel length is r, k represents the number of the depth convolution node, 1≤k≤2N1, represents the above bias parameter of the kth depth convolution node, Indicates The above spatial features of the above deep convolution nodes at the j+r-1th sampling point, Indicates The above-mentioned spatial domain features of the above-mentioned deep convolution nodes at the j+r-1th sampling point.

[0062] In the above embodiment, deep convolution is then used to extract features from the output signals of the obtained N1 new signal channels. Specifically, each new signal Convolution is performed with the two nodes of the second layer. Since the first layer has N1 channels of new signal output, there are a total of 2N1 depth convolution nodes. Assuming that the length of the convolution kernel is l, the kth (1≤k≤2N1) depth node has l weights and 1 bias The final extracted time domain feature output is: The length of the convolution kernel is set to one-fourth of the sampling frequency (128 Hz), that is, 32, and the final output feature size is (2N1, n-l+1), that is, (16,353).

[0063] In order to improve the expression ability and classification performance of the overall features, in an optional implementation, the above step S404 includes:

[0064] Step S4041, compressing each of the new signal channels of the normalized feature map into a corresponding scalar using a global average pooling operation;

[0065] Step S4042, recalculating the channel weight of each of the new signal channels through the fully connected layer;

[0066] Step S4043, applying the calculated channel weight to each of the new signal channels corresponding to the normalized feature map, and adjusting the channel response of the normalized feature map by element-by-element multiplication to obtain the calibration feature.

[0067] In the above embodiment, the attention mechanism layer aims to make the model pay more attention to the features useful for the task by adaptively weighting the channel weights of the input feature map, thereby improving the overall feature expression ability and classification performance. The attention mechanism layer is mainly divided into three steps: (1) Squeeze: Through the global average pooling operation, each channel of the input feature map is compressed into a scalar. This step captures the global spatial information and condenses it into a channel-level description. Specifically, given an input feature map H∈R C×N×T , where C is the number of RGB channels, N = 16 is the number of channels (electrodes) after expansion, T = 353 is the length of the time series, and the vector z∈R obtained by global average pooling N It can be expressed as: (2) Excitation: A fully connected two-layer neural network is used to recalculate the weight of each feature channel. This part of the network first reduces the feature dimension and then restores it to the original dimension. This "bottleneck" design captures the dependencies between nonlinear channels. The specific formula is s N =σ(W2·ReLU(W1·z N )),in and is the weight of the two fully connected layers, r is the scaling factor for adjusting the dimension. After two layers of fully connected neural network, σ is the sigmoid activation function (3) Scale: Finally, the channel weight s generated by the excitation is N Applied to each channel of the original feature map, the channel response of the feature map is adjusted by element-by-element multiplication. The specific operation is H i ',j,N =H i,j,N ·s N .

[0068] In order to further improve the ability of the model to extract features, in an optional implementation, the above step S409 includes:

[0069] Step S4091, using the above verification set to verify the above initial EEG fatigue recognition model and calculating the loss value of the above initial fatigue recognition model according to the loss function, the above loss function is Where η is the learning rate, is the gradient of the loss function with respect to the parameters, θ represents the above loss value, y is the label value of the above initial EEG fatigue recognition model in the current iteration, the above validation set is a data set divided from the sample data set, the above validation set includes multiple groups of validation data, each group of the above validation data includes the above sample forehead three-lead EEG data and the above sample fatigue state label value;

[0070] Step S4092, optimizing and adjusting the parameters of the initial EEG fatigue recognition model according to the loss value.

[0071] In the above embodiment, the model parameters are optimized and adjusted on the validation set. The specific method is as follows: By dynamically adjusting the model parameters during the training of each subject, the model's ability to extract features is further improved.

[0072] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0073] The embodiment of the present application also provides a fatigue identification device based on three-lead forehead EEG signals. It should be noted that the fatigue identification device based on three-lead forehead EEG signals in the embodiment of the present application can be used to execute the fatigue identification method based on three-lead forehead EEG signals provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0074] The following is an introduction to the fatigue identification device based on forehead three-lead EEG signals provided in an embodiment of the present application.

[0075] Figure 5is a structural block diagram of a fatigue identification device based on forehead three-lead EEG signals according to an embodiment of the present application. Figure 5 As shown, the device comprises:

[0076] An acquisition unit 10 is used to acquire a forehead three-lead EEG signal of a safety production personnel in real time;

[0077] Specifically, the three-lead EEG signals of the forehead of the safety production personnel are collected in real time through the EEG equipment. By obtaining the three-lead EEG signals of the forehead in real time, the system can monitor the fatigue state of the safety production personnel in real time and detect fatigue in time.

[0078] The recognition unit 20 is used to input the above-mentioned three-lead forehead EEG signal into the EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal using the above-mentioned EEG fatigue recognition model, and output a fatigue state label value. The above-mentioned EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, and each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values. The above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution. The above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel.

[0079] Specifically, the collected three-lead forehead EEG signals are input into a pre-trained EEG fatigue recognition model, such as Figure 3 As shown in the figure, the EEG fatigue recognition model uses convolution kernels of different sizes to extract spatial and temporal features. Two sizes of convolution kernels are used. The first is a size of (ch, 1), where ch refers to the number of input filtered lead channels, which is used to extract spatial features; then a convolution kernel of size (1, 32) is used to extract temporal features. The EEG fatigue recognition model analyzes the input EEG signal and outputs the fatigue status label value of the safety production personnel.

[0080] The prompt unit 30 is used to issue a prompt message when the fatigue state label value is greater than the fatigue warning value, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0081] Specifically, a prompt message is issued: when the fatigue status tag value is greater than the fatigue warning value, the system automatically issues a prompt message to remind the corresponding safety production personnel to suspend work. When the system detects that the safety production personnel are in a fatigue state, prompt messages are issued in a timely manner, which can effectively reduce the occurrence of work safety accidents caused by fatigue. By promptly reminding fatigued safety production personnel to suspend work, the problem of reduced work efficiency and reduced production quality due to fatigue can be avoided. At the same time, it helps to protect the health of safety production personnel.

[0082] In this embodiment, the acquisition unit is used to acquire the forehead three-lead EEG signals of the safety production personnel in real time; the recognition unit is used to input the above-mentioned forehead three-lead EEG signals into the EEG fatigue recognition model, so as to analyze the above-mentioned forehead three-lead EEG signals using the above-mentioned EEG fatigue recognition model, and output a fatigue state label value. The above-mentioned EEG fatigue recognition model is a convergent model obtained by iterative training by inputting multiple groups of training data into a predetermined model, each group of the above-mentioned training data includes sample forehead three-lead EEG data and sample fatigue state label values, and the ... The model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The point convolution layer is used to extract the spatial features in the three-lead EEG signal of the forehead by using point convolution. The deep convolution layer is used to extract the time domain features in the three-lead EEG signal of the forehead by using deep convolution. The fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel; the prompt unit is used to issue a prompt message when the fatigue state label value is greater than the fatigue warning value. The prompt message is used to remind the corresponding safety production personnel to suspend work. The present invention obtains an EEG fatigue recognition model by designing a multi-layer convolutional neural network, combining multi-step preprocessing and feature extraction technology training, and realizes high-accuracy fatigue recognition between different individuals by using only the EEG data of the forehead area (i.e., the three-lead EEG data of the forehead). The method of the present invention not only improves the accuracy and robustness of fatigue recognition, but also has good generalization ability, and can be widely used in practical scenarios such as driver monitoring and worker fatigue detection. The present application solves the problem that the existing technology has poor cross-individual recognition effect, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0083] In order to ensure the accuracy and reliability of the experimental data, in an optional implementation manner, the above-mentioned device further includes:

[0084] a recording unit for obtaining original three-lead frontal EEG signals corresponding to a plurality of subjects before inputting the three-lead frontal EEG signals into an EEG fatigue recognition model to analyze the three-lead frontal EEG signals using the EEG fatigue recognition model and outputting fatigue state label values, and recording event occurrence time, reaction start time, lane return center time corresponding to the subjects when lane departure events occur in an EEG data acquisition experiment, and aligning them with the EEG data timeline, wherein the EEG data acquisition experiment is an experiment in which a lane departure event is randomly triggered on the subjects after the subjects wear helmets to obtain reaction time of the subjects to the randomly occurring lane departure events;

[0085] a marking unit, for marking the corresponding original three-lead forehead electroencephalogram signal according to a local reaction time, wherein the local reaction time is a time interval between the event occurrence time of the lane departure event and the reaction start time, and the local reaction time is used to evaluate the fatigue level of the subject;

[0086] The preprocessing unit is used to perform preprocessing operations on the original three-lead forehead EEG signals to obtain the training data. The preprocessing operations at least include bandpass filtering, independent component analysis and baseline correction.

[0087] In the above embodiment, in the raw data acquisition experiment, the subject is required to concentrate as much as possible and respond immediately to the randomly occurring lane departure events. In the experiment, the time when the lane departure event occurs, the reaction occurs, and the lane returns to the center is recorded respectively, and aligned with the EEG data timeline. The local reaction time is used to evaluate the fatigue level, which is defined as the time interval between the start of the event and the start of the reaction of each lane departure event, because a long time interval may mean a slower reaction speed, which may be caused by fatigue. Marking the local reaction time can help researchers better understand the subject's reaction to the lane departure event. In the experiment, EEG data of the frontal area, including three channels of Fp1, Fp2, and Fz (forehead lead electrodes in the 10-20 international standard lead system), were used to downsample the EEG signal to 128Hz, and a 3-second long sample was extracted before each lane departure event to form a vector. Subsequently, the extracted data was preprocessed using physiological knowledge and existing conclusions on fatigue recognition, including bandpass filtering, independent component analysis (ICA) to remove artifacts, and baseline correction. Bandpass filtering is used to remove high-frequency noise and low-frequency drift. Bandpass filtering can remove noise from the signal and improve the quality of the signal; independent component analysis can identify and remove components that are not related to the signal of interest; baseline correction can make the data of different subjects comparable, making subsequent data analysis more accurate. Through the above steps, accurate EEG data can be obtained, and effective preprocessing operations can be performed to provide a reliable data basis for subsequent data analysis and fatigue level assessment. The accuracy and reliability of the experimental data are ensured, and an effective means is provided to assess the fatigue level of the subjects. Finally, the processed data set is cut into a training set, a test set, and a validation set. The training set is used for training the convolutional neural network, the test set is used to evaluate the model effect, and the validation set is used to adjust the model parameters and improve the generalization ability.

[0088] In order to improve the accuracy of the EEG fatigue recognition model, in an optional implementation, the device further includes:

[0089] A spatial feature extraction unit is used to input the above-mentioned three-lead forehead EEG signal into the EEG fatigue recognition model to analyze the above-mentioned three-lead forehead EEG signal using the above-mentioned EEG fatigue recognition model, and before outputting the fatigue state label value, perform a spatial processing step, generate the same number of new signal channels using multiple point convolution nodes in the above-mentioned point convolution layer of the initial EEG fatigue recognition model, and use each of the above-mentioned point convolution nodes to perform spatial feature processing on the above-mentioned sample three-lead forehead EEG signal, and output the above-mentioned spatial domain features through the corresponding above-mentioned new signal channel;

[0090] A time feature extraction unit, used to execute the time processing step, using a plurality of deep convolution nodes in the deep convolution layer of the initial EEG fatigue recognition model to perform time feature processing on the spatial features to obtain the time domain features, wherein the number of the deep convolution nodes is twice the number of the point convolution nodes;

[0091] A feature processing unit, used to perform a feature processing step, inputting the spatial domain features and the time domain features into the activation layer and the batch normalization layer of the initial EEG fatigue recognition model in sequence for feature processing, and obtaining a normalized feature map;

[0092] An attention weighting unit, used to perform an attention weighting step, input the normalized feature map into the attention mechanism layer of the initial EEG fatigue recognition model for attention weighting processing, and obtain weighted features;

[0093] A refinement unit, used to perform a refinement step, to refine the weighted features input to the point convolution layer of the initial EEG fatigue recognition model to obtain refined features, wherein the refinement is used to refine the correlation between different channels;

[0094] A residual unit is used to perform a residual step, and input the weighted features and the refined features into the activation layer and the attention mechanism layer in the form of residual connections to obtain effective features;

[0095] An output unit, used to execute the output step, input the above-mentioned effective features into the above-mentioned output layer, and output a training fatigue state label value;

[0096] A verification unit, used to perform a verification step, verify the initial EEG fatigue recognition model according to the fatigue state label value of the sample, and obtain a model accuracy, wherein the model accuracy is used to represent the accuracy of the initial EEG fatigue recognition model;

[0097] A repeating unit is used to execute the repetitive steps. When the accuracy of the above-mentioned model does not meet the accuracy requirements, the parameters of the above-mentioned initial EEG fatigue recognition model are optimized and adjusted on the verification set, and the above-mentioned spatial processing step, the above-mentioned temporal processing step, the above-mentioned feature processing step, the above-mentioned attention weighting step, the above-mentioned refinement step, the above-mentioned residual step, the above-mentioned output step, the above-mentioned verification step and the above-mentioned repetitive step are re-executed in sequence at least once, until the accuracy of the above-mentioned model meets the above-mentioned accuracy requirements, thereby obtaining the above-mentioned EEG fatigue recognition model.

[0098] In the above embodiment, multiple point convolution nodes are used to generate a new signal channel, and then the EEG signal is processed with spatial features, and the spatial features are output through the generated new signal channel. The spatial features of the EEG signal are extracted so that the subsequent processing process can better identify the EEG fatigue state. On the basis of the spatial features, the time features are further extracted to improve the recognition accuracy of the EEG fatigue state. The spatial features and the time domain features are input into the activation layer and the batch normalization layer for feature processing to obtain a normalized feature map. The role of this step is to further process the features through the activation layer and the batch normalization layer, so that the features are more distinguishable and representative. The normalized feature map is input into the attention mechanism layer for attention weighted processing to obtain weighted features, and the features are weighted through the attention mechanism to improve the degree of attention to key features, so as to better identify the EEG fatigue state. The weighted features are input into the point convolution layer for refinement processing to obtain refined features, which are used to refine the correlation between different channels. The role of this step is to refine the features through the point convolution layer to improve the correlation and representativeness of the features. The re-weighted features are then passed through the point convolution layer to further refine the correlation between different channels. Subsequently, the output features of the attention mechanism layer and the point convolution layer are input to the activation layer in the form of residual connections. By linearly combining the information of different channels, the feature representation capability is enhanced. The activated features are once again processed by the attention mechanism layer to weight the features; finally, after operations such as pooling, dropout, full connection, and softmax, the results of fatigue monitoring are output, and the valid features are input to the output layer, and the training fatigue state label value is output. The purpose of this step is to classify the valid features through the output layer to obtain the label value of the training fatigue state. The initial model is verified according to the training fatigue state label value to obtain the model accuracy, which is used to indicate the accuracy of the model. The purpose of this step is to evaluate the recognition accuracy of the model by verifying the model accuracy. If the model accuracy does not meet the requirements, the model parameters will be optimized and adjusted on the verification set, and the above steps will be re-executed until the model accuracy meets the requirements. By continuously optimizing and adjusting the model until the recognition accuracy that meets the requirements is achieved, the final EEG fatigue recognition model is obtained. The framework diagram of the EEG fatigue recognition model is shown in the figure. Figure 4 shown.

[0099] In order to improve the richness of data and promote model convergence from the perspective of spatial features, in an optional implementation, the spatial feature extraction unit includes:

[0100] The first calculation module is used to calculate the above-mentioned spatial domain features according to the weights and bias parameters of each of the above-mentioned point convolution nodes in combination with the first formula, and output the above-mentioned spatial domain features through the corresponding above-mentioned new signal channel. The above-mentioned first formula is in, represents the above spatial characteristics, x p,jis the jth sampling point of the pth input channel of the above sample three-lead forehead EEG signal, represents the weight corresponding to the p-th input channel of the i-th point convolution node, Represents the bias parameter of the i-th point convolution node, i represents the number of the point convolution node, 1≤i≤N1, and m represents the total number of input channels.

[0101] In the above embodiment, N1 point convolution nodes are first used to generate N1 new signal channels of the same length. For a given input sample X (m，n) , which contains m one-dimensional signals with input channel length n. The i-th (1≤i≤N1) node has m weights and 1 bias parameter Finally, the spatial features are extracted N1 is set to 8, which is approximately twice the number of input channels, to improve data richness and promote model convergence.

[0102] In order to improve the richness of data and promote model convergence from the perspective of time features, in an optional implementation, the time feature extraction unit includes:

[0103] The second calculation module is used to calculate the time domain features according to the above-mentioned spatial domain features, the weights and bias parameters of the above-mentioned deep convolution nodes combined with the second formula. The above-mentioned second formula is Represents the above time domain characteristics, represents the weight of the kth depth convolution node when the convolution kernel length is r, k represents the number of the depth convolution node, 1≤k≤2N1, represents the above bias parameter of the kth depth convolution node, Indicates The above spatial features of the above deep convolution nodes at the j+r-1th sampling point, Indicates The above-mentioned spatial domain features of the above-mentioned deep convolution nodes at the j+r-1th sampling point.

[0104] In the above embodiment, deep convolution is then used to extract features from the output signals of the obtained N1 new signal channels. Specifically, each new signal Convolution is performed with the two nodes of the second layer. Since the first layer has N1 channels of new signal output, there are a total of 2N1 depth convolution nodes. Assuming that the length of the convolution kernel is l, the kth (1≤k≤2N1) depth node has l weights and 1 bias The final extracted time domain feature output is: The length of the convolution kernel is set to one-fourth of the sampling frequency (128 Hz), that is, 32, and the final output feature size is (2N1, n-l+1), that is, (16,353).

[0105] In order to improve the expression ability and classification performance of the overall features, in an optional implementation, the attention weighting unit includes:

[0106] A compression module, used for compressing each of the new signal channels of the normalized feature map into a corresponding scalar using a global average pooling operation;

[0107] A third calculation module, used to recalculate the channel weight of each of the new signal channels through the fully connected layer;

[0108] The scaling module is used to apply the calculated channel weight to each of the new signal channels corresponding to the normalized feature map, and adjust the channel response of the normalized feature map by element-by-element multiplication to obtain the calibration feature.

[0109] In the above embodiment, the attention mechanism layer aims to make the model pay more attention to the features useful for the task by adaptively weighting the channel weights of the input feature map, thereby improving the overall feature expression ability and classification performance. The attention mechanism layer is mainly divided into three steps: (1) Squeeze: Through the global average pooling operation, each channel of the input feature map is compressed into a scalar. This step captures the global spatial information and condenses it into a channel-level description. Specifically, given an input feature map H∈R C×N×T , where C is the number of RGB channels, N = 16 is the number of channels (electrodes) after expansion, T = 353 is the length of the time series, and the vector z∈R obtained by global average pooling N It can be expressed as: (2) Excitation: A fully connected two-layer neural network is used to recalculate the weight of each feature channel. This part of the network first reduces the feature dimension and then restores it to the original dimension. This "bottleneck" design captures the dependencies between nonlinear channels. The specific formula is s N =σ(W2·ReLU(W1·z N )),in and is the weight of the two fully connected layers, r is the scaling factor for adjusting the dimension. After two layers of fully connected neural network, σ is the sigmoid activation function (3) Scale: Finally, the channel weight s generated by the excitation is N Applied to each channel of the original feature map, the channel response of the feature map is adjusted by element-by-element multiplication. The specific operation is H' i,j,N =Hi,j,N ·s N .

[0110] In order to further improve the ability of the model to extract features, in an optional embodiment, the above-mentioned repeating unit includes:

[0111] The fourth calculation module is used to verify the initial EEG fatigue recognition model using the verification set and calculate the loss value of the initial fatigue recognition model according to the loss function, and the loss function is θ = θ - η▽ θ L(θ;X,y), where η is the learning rate, ▽ θ L is the gradient of the loss function with respect to the parameter, θ represents the above loss value, y is the label value of the above initial EEG fatigue recognition model in the current iteration, the above validation set is a data set divided from the sample data set, the above validation set includes multiple groups of validation data, and each group of the above validation data includes the above sample forehead three-lead EEG data and the above sample fatigue state label value;

[0112] The optimization module is used to optimize and adjust the parameters of the initial EEG fatigue recognition model according to the loss value.

[0113] In the above embodiment, the model parameters are optimized and adjusted on the validation set, and the specific method is θ = θ - η ▽ θ L(θ; X, y), further improves the model's ability to extract features by dynamically adjusting the model parameters during the training of each subject.

[0114] The fatigue identification device based on forehead three-lead EEG signals comprises a processor and a memory, and the above units are stored in the memory as program units, and the processor executes the above program units stored in the memory to realize the corresponding functions. The above modules are all located in the same processor; or, the above modules are located in different processors in any combination.

[0115] The processor includes a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problem that the cross-individual recognition effect in the existing technology is poor, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0116] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0117] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the fatigue identification method based on forehead three-lead EEG signals.

[0118] Specifically, the fatigue recognition method based on the forehead three-lead EEG signal includes:

[0119] Step S201, obtaining three-lead forehead EEG signals of production safety personnel in real time;

[0120] Step S202, inputting the above-mentioned three-lead forehead EEG signal into an EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal by using the above-mentioned EEG fatigue recognition model, and outputting a fatigue state label value, wherein the above-mentioned EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution, and the above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel;

[0121] Step S203: When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0122] An embodiment of the present invention provides a processor, which is used to run a program, wherein the fatigue identification method based on forehead three-lead EEG signals is executed when the program is run.

[0123] Specifically, the fatigue recognition method based on the forehead three-lead EEG signal includes:

[0124] Step S201, obtaining three-lead forehead EEG signals of production safety personnel in real time;

[0125] Step S202, inputting the above-mentioned three-lead forehead EEG signal into an EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal by using the above-mentioned EEG fatigue recognition model, and outputting a fatigue state label value, wherein the above-mentioned EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution, and the above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel;

[0126] Step S203: When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0127] The embodiment of the present invention provides a fatigue recognition system based on a forehead three-lead EEG signal. The fatigue recognition system based on a forehead three-lead EEG signal includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented:

[0128] Step S201, obtaining three-lead forehead EEG signals of production safety personnel in real time;

[0129] Step S202, inputting the above-mentioned three-lead forehead EEG signal into an EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal by using the above-mentioned EEG fatigue recognition model, and outputting a fatigue state label value, wherein the above-mentioned EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution, and the above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel;

[0130] Step S203: When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0131] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0132] Step S201, obtaining three-lead forehead EEG signals of production safety personnel in real time;

[0133] Step S202, inputting the above-mentioned three-lead forehead EEG signal into an EEG fatigue recognition model, so as to analyze the above-mentioned three-lead forehead EEG signal by using the above-mentioned EEG fatigue recognition model, and outputting a fatigue state label value, wherein the above-mentioned EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple groups of training data into a predetermined model, each group of the above-mentioned training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the above-mentioned point convolution layer is used to extract the spatial domain features in the above-mentioned three-lead forehead EEG signal by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead forehead EEG signal by using deep convolution, and the above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel;

[0134] Step S203: When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

[0135] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0136] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0137] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0140] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0141] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0142] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0143] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0144] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0145] 1) The fatigue recognition method based on the three-lead forehead EEG signal of the present application first obtains the three-lead forehead EEG signal of the safety production personnel in real time; then, the three-lead forehead EEG signal is input into the EEG fatigue recognition model, and the three-lead forehead EEG signal is analyzed by the EEG fatigue recognition model, and the fatigue state label value is output. The EEG fatigue recognition model is a convergent model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, and each set of the training data includes sample three-lead forehead EEG data and sample fatigue state label value. The above-mentioned EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer. The above-mentioned point convolution layer is used to extract the spatial features in the above-mentioned three-lead EEG signal of the forehead by using point convolution, and the above-mentioned deep convolution layer is used to extract the time domain features in the above-mentioned three-lead EEG signal of the forehead by using deep convolution. The above-mentioned fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel; finally, when the above-mentioned fatigue state label value is greater than the fatigue warning value, a prompt message is issued, and the above-mentioned prompt message is used to remind the corresponding above-mentioned safety production personnel to suspend work. The present invention obtains an EEG fatigue recognition model by designing a multi-layer convolutional neural network, combining multi-step preprocessing and feature extraction technology training, and realizes high-accuracy fatigue recognition between different individuals when only the EEG data of the forehead area (i.e., the above-mentioned three-lead EEG data of the forehead) is used. The method of the present invention not only improves the accuracy and robustness of fatigue recognition, but also has good generalization ability, and can be widely used in practical scenarios such as driver monitoring and worker fatigue detection. The present application solves the problem that the existing technology has poor cross-individual recognition effect, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0146] 2) The fatigue identification device based on the three-lead forehead EEG signal of the present application comprises an acquisition unit for acquiring the three-lead forehead EEG signal of the safety production personnel in real time; an identification unit for inputting the three-lead forehead EEG signal into the EEG fatigue identification model, so as to analyze the three-lead forehead EEG signal using the EEG fatigue identification model and output a fatigue state label value. The EEG fatigue identification model is a convergent model obtained by iteratively training multiple sets of training data input into a predetermined model, and each set of the training data includes sample three-lead forehead EEG data and sample fatigue state label. Value, the EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, the point convolution layer is used to extract the spatial features in the three-lead EEG signal of the forehead by point convolution, the deep convolution layer is used to extract the time domain features in the three-lead EEG signal of the forehead by deep convolution, and the fatigue state label value is used to indicate the fatigue degree of the above-mentioned safety production personnel; the prompt unit is used to send a prompt message when the fatigue state label value is greater than the fatigue warning value, and the prompt message is used to remind the corresponding safety production personnel to suspend work. The present invention designs a multi-layer convolutional neural network, combines multi-step preprocessing and feature extraction technology training to obtain an EEG fatigue recognition model, and realizes high-accuracy fatigue recognition between different individuals by using only the EEG data of the forehead area (i.e. the three-lead EEG data of the forehead). The method of the present invention not only improves the accuracy and robustness of fatigue recognition, but also has good generalization ability, and can be widely used in practical scenarios such as driver monitoring and worker fatigue detection. The present application solves the problem that the existing technology has poor cross-individual recognition effect, especially under the influence of individual differences, signal quality and task changes, the existing model is difficult to achieve efficient and stable fatigue recognition.

[0147] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fatigue recognition method based on forehead three-lead EEG signals, characterized in that: include: Real-time acquisition of three-lead EEG signals from the forehead of production safety personnel; The three-lead forehead EEG signal is input into an EEG fatigue recognition model, so as to analyze the three-lead forehead EEG signal using the EEG fatigue recognition model, and output a fatigue state label value, wherein the EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, each set of the training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, wherein the point convolution layer is used to extract spatial domain features from the three-lead forehead EEG signal by using point convolution, and the deep convolution layer is used to extract time domain features from the three-lead forehead EEG signal by using deep convolution, and the fatigue state label value is used to indicate the fatigue degree of the safety production personnel; When the fatigue status tag value is greater than the fatigue warning value, a prompt message is issued, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

2. The method according to claim 1, characterized in that Before inputting the forehead three-lead EEG signal into the EEG fatigue recognition model to analyze the forehead three-lead EEG signal using the EEG fatigue recognition model and outputting a fatigue state label value, the method further includes: Obtaining original three-lead forehead EEG signals corresponding to multiple subjects, and recording the event occurrence time, reaction start time, and lane return center time of the subjects when a lane departure event occurs in an EEG data acquisition experiment, and aligning them with the EEG data timeline. The EEG data acquisition experiment is a test in which a lane departure event is randomly triggered on the subject after the subject wears a helmet to obtain the subject's reaction time to the randomly occurring lane departure event; Marking the corresponding original three-lead forehead electroencephalogram signal according to the local reaction time, wherein the local reaction time is the time interval between the event occurrence time of the lane departure event and the reaction start time, and the local reaction time is used to evaluate the fatigue level of the subject; The original three-lead forehead electroencephalogram signal is preprocessed to obtain the training data, wherein the preprocessing operation at least includes bandpass filtering, independent component analysis and baseline correction.

3. The method according to claim 1, characterized in that Before inputting the forehead three-lead EEG signal into the EEG fatigue recognition model to analyze the forehead three-lead EEG signal using the EEG fatigue recognition model and outputting a fatigue state label value, the method further includes: A spatial processing step, using a plurality of point convolution nodes in the point convolution layer of the initial EEG fatigue recognition model to generate the same number of new signal channels, and using each of the point convolution nodes to perform spatial feature processing on the sample forehead three-lead EEG signal, and outputting the spatial domain features through the corresponding new signal channel; A time processing step, in the deep convolution layer of the initial EEG fatigue recognition model, using a plurality of deep convolution nodes to perform time feature processing on the spatial features to obtain the time domain features, wherein the number of the deep convolution nodes is twice the number of the point convolution nodes; A feature processing step, inputting the spatial domain features and the time domain features into the activation layer and the batch normalization layer of the initial EEG fatigue recognition model in sequence for feature processing to obtain a normalized feature map; An attention weighting step, inputting the normalized feature map into the attention mechanism layer of the initial EEG fatigue recognition model for attention weighting processing to obtain weighted features; A refinement step, performing refinement processing on the weighted features input to the point convolution layer of the initial EEG fatigue recognition model to obtain refined features, wherein the refinement processing is used to refine the correlation between different channels; The residual step is to input the weighted features and the refined features into the activation layer and the attention mechanism layer in the form of residual connection in sequence to obtain effective features; An output step, inputting the effective features into the output layer, and outputting a training fatigue state label value; A verification step, verifying the initial EEG fatigue recognition model according to the sample fatigue state label value to obtain a model accuracy, wherein the model accuracy is used to represent the accuracy of the initial EEG fatigue recognition model; Repeating the steps, when the model accuracy does not meet the accuracy requirements, optimizing and adjusting the parameters of the initial EEG fatigue recognition model on the validation set, and sequentially re-executing the spatial processing step, the temporal processing step, the feature processing step, the attention weighting step, the refinement step, the residual step, the output step, the verification step and the repetition step at least once until the model accuracy meets the accuracy requirements, thereby obtaining the EEG fatigue recognition model.

4. The method according to claim 3, characterized in that Using each of the point convolution nodes to perform spatial feature processing on the sample forehead three-lead EEG signal, and outputting the spatial domain features through the corresponding new signal channel, including: The spatial domain feature is calculated according to the weight and bias parameters of each point convolution node in combination with the first formula, and the spatial domain feature is output through the corresponding new signal channel. The first formula is: in, represents the spatial characteristics, x p,j is the jth sampling point of the pth input channel of the sample three-lead forehead EEG signal, represents the weight corresponding to the p-th input channel of the i-th point convolution node, represents the bias parameter of the i-th point convolution node, i represents the number of the point convolution node, 1≤i≤N1, and m represents the total number of input channels.

5. The method according to claim 3, characterized in that: In the deep convolution layer of the initial EEG fatigue recognition model, a plurality of deep convolution nodes are used to perform time feature processing on the spatial features to obtain the time domain features, including: The time domain feature is calculated according to the spatial domain feature, the weight and bias parameter of the deep convolution node and the second formula, where the second formula is: represents the time domain feature, represents the weight of the kth depth convolution node when the convolution kernel length is r, k represents the number of the depth convolution node, 1≤k≤2N1, represents the bias parameter of the kth depth convolution node, Indicates The spatial domain features of the deep convolution nodes at the j+r-1th sampling point, Indicates The spatial domain features of the deep convolution nodes at the j+r-1th sampling point.

6. The method according to claim 3, characterized in that The normalized feature map is input into the attention mechanism layer of the initial EEG fatigue recognition model for recalibration to obtain calibration features, including: Compressing each of the new signal channels of the normalized feature map into a corresponding scalar using a global average pooling operation; Recalculating the channel weight of each of the new signal channels through the fully connected layer; The calculated channel weight is applied to each of the new signal channels corresponding to the normalized feature map, and the channel response of the normalized feature map is adjusted by element-by-element multiplication to obtain the calibration feature.

7. The method according to claim 3, characterized in that Optimizing and adjusting the parameters of the initial EEG fatigue recognition model on the validation set includes: The initial EEG fatigue recognition model is verified by using the verification set and the loss value of the initial fatigue recognition model is calculated according to the loss function, wherein the loss function is: Where η is the learning rate, is the gradient of the loss function with respect to the parameter, θ represents the loss value, y is the label value of the initial EEG fatigue recognition model in the current iteration, the validation set is a data set divided from the sample data set, the validation set includes multiple groups of validation data, each group of validation data includes the sample forehead three-lead EEG data and the sample fatigue state label value; The parameters of the initial EEG fatigue recognition model are optimized and adjusted according to the loss value.

8. A fatigue identification device based on forehead three-lead EEG signals, characterized in that: The device comprises: An acquisition unit, used for acquiring three-lead forehead EEG signals of safety production personnel in real time; A recognition unit, used for inputting the three-lead forehead EEG signal into an EEG fatigue recognition model, so as to analyze the three-lead forehead EEG signal by using the EEG fatigue recognition model, and outputting a fatigue state label value, wherein the EEG fatigue recognition model is a convergence model obtained by iteratively training by inputting multiple sets of training data into a predetermined model, each set of the training data includes sample three-lead forehead EEG data and sample fatigue state label values, and the EEG fatigue recognition model includes a point convolution layer, a deep convolution layer, an activation layer, a batch normalization layer, a fully connected layer and an output layer, wherein the point convolution layer is used for extracting spatial domain features from the three-lead forehead EEG signal by using point convolution, and the deep convolution layer is used for extracting time domain features from the three-lead forehead EEG signal by using deep convolution, and the fatigue state label value is used for indicating the fatigue degree of the safety production personnel; The prompt unit is used to issue a prompt message when the fatigue status tag value is greater than the fatigue warning value, and the prompt message is used to remind the corresponding production safety personnel to suspend work.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images