Method and apparatus for determining sensitive brain regions and electroencephalographic features for fatigue detection

Abstract

The application discloses a method for determining sensitive brain areas and electroencephalogram features for fatigue detection. The method comprises the following steps: selecting multiple signal collection positions on the head and collecting wakefulness electroencephalogram signals and fatigue electroencephalogram signals; determining multiple wakefulness electroencephalogram features and multiple fatigue electroencephalogram features according to the wakefulness electroencephalogram signals and the fatigue electroencephalogram signals collected at each signal collection position; determining the difference degree of the same type of wakefulness electroencephalogram features and fatigue electroencephalogram features at each signal collection position and sorting to obtain a sorting result; and selecting sensitive brain areas and electroencephalogram features for fatigue judgment from the signal collection positions and electroencephalogram features within a preset high difference degree range according to the sorting result. The above method can select sensitive brain areas and electroencephalogram features that are more conducive to fatigue detection. The application also discloses a device for determining sensitive brain areas and electroencephalogram features for fatigue detection, a fatigue detection method, an electronic device and a computer readable storage medium.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, and in particular to a method for determining sensitive brain regions and EEG characteristics for fatigue detection, an apparatus for determining sensitive brain regions and EEG characteristics for fatigue detection, a fatigue detection method, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Due to prolonged, irregular, and nighttime flights, as well as cross-time zone changes, flight personnel are prone to sleep deprivation, fatigue, and altered circadian rhythms, severely impacting their work capacity and increasing the risk of flight safety hazards. The prevention and monitoring of flight fatigue is a crucial research topic in aerospace medicine. Approximately 75% of aviation accidents are related to human error, and the decline in work capacity due to flight fatigue is a significant contributing factor. Therefore, establishing a dynamic monitoring system for flight fatigue is of paramount importance for ensuring flight safety.

[0003] There are many methods for assessing flight fatigue, such as rating scales, objective observation, and physiological and biochemical methods, but the assessment system is not yet mature. Subjective methods such as fatigue rating scales cannot objectively reflect the actual functional state of fatigued personnel. Objective observation methods include head position sensing, mouth position detection, pupil measurement, visual-motor detection, and eyelid closure percentage detection. These techniques are designed for pilots of long-haul transport aircraft and civil aviation aircraft, who experience low workloads and prolonged periods of mental exertion, and are not applicable to military pilots who have high operational intensity and high sensitivity to fatigue.

[0004] EEG signals have advantages such as high temporal resolution and strong functional specificity, therefore they have become the gold standard for detecting fatigue based on physiological signals.

[0005] However, the following problems still exist in the current technology for fatigue detection using EEG signals.

[0006] There is no unified standard for fatigue detection indicators using electroencephalogram (EEG) signals. A 2009 study used EEG signals to assess driver fatigue levels, finding in simulated driving experiments that a higher ratio of EEG power spectrum (α+θ) / β correlated with a higher fatigue level. In 2013, researchers analyzed the EEG characteristics of actual long-distance bus drivers during fatigue, collecting data during breaks in driving, concluding that a higher θ / β ratio correlated with a higher fatigue level. In 2021, researchers found that the Rayleigh entropy of EEG increases continuously with accumulated fatigue. Also in 2021, researchers identified the prefrontal cortex as a sensitive brain region for fatigue, with brain functional connectivity indicators contributing most significantly to fatigue recognition models. Therefore, it is evident that there is no consensus or definitive conclusion regarding fatigue detection indicators and methods.

[0007] There is a trade-off between the accuracy and ease of use in EEG fatigue detection. High-precision EEG fatigue detection typically requires complex equipment and multiple electrodes to capture the brain's electrical activity. For example, some studies have used EEG acquisition systems with up to 64 leads to simultaneously acquire subjects' EEG signals. While such complex setups can achieve high accuracy in a laboratory environment, their ease of use may be reduced in practical applications, such as driving or the workplace, due to the complexity and lack of portability of the equipment.

[0008] To improve detection accuracy, researchers typically extract a large number of features from EEG signals and use complex algorithms for feature selection. For example, some studies have proposed using deep networks for multi-level feature generation, combining one-dimensional binary pattern recognition (BP) and statistical features with one-dimensional discrete wavelet transform (1D-DWT) to create multi-layer features. While these advanced feature extraction and selection methods can improve accuracy, they also increase the complexity of the system, hindering real-time and convenient use.

[0009] High-precision EEG fatigue detection often requires complex algorithms to process and analyze the data. For example, convolutional neural networks (CNNs) are used to automatically extract features from EEG signals and classify states. While these algorithms perform well in classification accuracy, they typically require significant computational resources, which may limit their real-time application on mobile or low-power devices.

[0010] To obtain high-quality EEG signals, users typically need to wear multiple electrodes, which can cause discomfort, especially during prolonged use. Some studies have proposed using forehead EEG signals to design portable driver fatigue detection frames to reduce discomfort and improve ease of use. However, reducing the number of electrodes may affect signal quality, thus impacting detection accuracy.

[0011] High-precision EEG fatigue detection equipment is often expensive, which limits its widespread adoption. On the other hand, low-cost equipment may not provide sufficient accuracy, thus limiting its effectiveness in practical applications.

[0012] Therefore, it is evident that a major challenge is to select sensitive brain regions for fatigue detection, select EEG indicators with high sensitivity and good consistency, and minimize the number of selected sensitive brain regions and EEG indicators while making them easy to calculate. Summary of the Invention

[0013] This invention proposes a method for determining sensitive brain regions and EEG features for fatigue detection, in order to solve the above-mentioned problems.

[0014] In a first aspect, embodiments of the present invention disclose a method for determining sensitive brain regions and electroencephalogram (EEG) features for fatigue detection. The sensitive brain regions are areas of the head where EEG features are collected. The determination method includes:

[0015] Multiple signal acquisition locations were selected on the head, and corresponding awake and fatigued EEG signals were acquired at each location; where awake EEG signals are EEG signals in an awake state, and fatigued EEG signals are EEG signals in a fatigued state.

[0016] Based on the awake and fatigued EEG signals collected at each signal acquisition location, we determine multiple awake EEG characteristics corresponding to the awake EEG signals and multiple fatigued EEG characteristics corresponding to the fatigued EEG signals.

[0017] The degree of difference between the same type of awake EEG features and fatigue EEG features at each signal acquisition location is determined, and the signal acquisition locations and corresponding EEG features are sorted according to the degree of difference. The sorting results are obtained, and the final signal acquisition location and final EEG features are selected from the signal acquisition locations and EEG features located in the preset high degree of difference range as sensitive brain regions and EEG features used for fatigue judgment. The preset high degree of difference range is the range with relatively high degree of difference in the sorting.

[0018] Using the above technical solution, the method for determining sensitive brain regions and EEG features for fatigue detection in this invention can compare the degree of difference between the same type of awake EEG features and fatigue EEG features collected from different signal acquisition locations, and can also compare the sensitivity of different types of EEG features to fatigue. Then, by sorting, signal acquisition locations and EEG features located within a preset high degree of difference range are selected, and then sensitive brain regions and EEG features used for fatigue judgment are selected. This method can accurately select sensitive brain regions and EEG features suitable for fatigue detection, and can also make the number of selected sensitive brain regions relatively small and the selected EEG features relatively easy to calculate.

[0019] According to another specific embodiment of the present invention, the types of awake EEG features and fatigue EEG features include aperiodic offset features and aperiodic scaling transform features, and the aperiodic offset features and aperiodic scaling transform features of the EEG signals are determined by the following methods:

[0020] Determine the logarithmic power spectrum of the EEG signal;

[0021] The corresponding fitting curve is obtained by performing nonlinear fitting on the logarithmic power spectrum using the following formula 1:

[0022] L(f)=ab×log(P(f))(Formula 1)

[0023] Where L(f) is the fitted curve, P(f) is the logarithmic power spectrum, a is the offset parameter of the fitted curve, b is the scaling transformation parameter of the fitted curve, a and b are determined by least squares fitting, and f is the frequency of the EEG signal, with the range of f being 2-5Hz and 14-30Hz.

[0024] The offset parameter 'a' in Formula 1 is determined to be an aperiodic offset feature, and the scaling transformation parameter 'b' in Formula 1 is determined to be an aperiodic scaling transformation feature.

[0025] According to another specific embodiment of the present invention, the types of awake EEG characteristics and fatigue EEG characteristics further include alpha wave band energy characteristics and alpha wave band peak characteristics, which are determined by the following methods:

[0026] Based on the logarithmic power spectrum and the fitted curve, the periodic components of the α-wave spectrum are determined using the following formula 2:

[0027] P ′ (f)=P(f)-L(f)(Formula 2)

[0028] Among them, P ′ (f) represents the periodic component of the α-wave spectrum, L(f) is the fitted curve, P(f) is the logarithmic power spectrum, and f is the frequency of the EEG signal.

[0029] The periodic energy component of the α-wave spectrum is determined using the following formula 3 based on the periodic component of the α-wave spectrum:

[0030] P ″ (f) = 10 × 10 P′(f) (Formula 3)

[0031] Among them, P ′ (f) represents the periodic component of the α-wave spectrum, P ″ (f) represents the periodic energy component of the alpha wave spectrum, and f represents the frequency of the EEG signal;

[0032] The energy characteristics and peak characteristics of the α-wave band are determined based on the periodic energy components of the α-wave spectrum.

[0033] According to another specific embodiment of the present invention, corresponding awake EEG signals and fatigue EEG signals are collected at each of a plurality of signal acquisition locations on the head, specifically including:

[0034] At each of the multiple signal acquisition locations on the head, the corresponding raw awake EEG signal and raw fatigue EEG signal were acquired respectively.

[0035] The raw awake EEG signal and the raw fatigue EEG signal are preprocessed to obtain the awake EEG signal and the fatigue EEG signal, respectively. The preprocessing method includes at least one of the following: data segmentation, averaging, filtering, and windowing.

[0036] According to another specific embodiment of the present invention, the degree of difference between the same type of awake EEG features and fatigue EEG features is determined by a feature selection algorithm. Based on the ranking results, the final signal acquisition location and final EEG features are selected from signal acquisition locations and EEG features located within a preset high degree of difference range. Specifically, this includes:

[0037] Determine the frequency of occurrence of different signal acquisition locations and different EEG features within a preset range of high variability;

[0038] The signal acquisition locations and EEG features that appear relatively frequently are selected as the candidate datasets for the location and feature selection algorithm. The final signal acquisition locations and final EEG features are then selected from the candidate datasets.

[0039] According to another specific embodiment of the present invention, the feature selection algorithm comprises multiple algorithms, which select the final signal acquisition location and final EEG features from signal acquisition locations and EEG features located within a preset high degree of difference range based on the ranking result, and further includes:

[0040] Obtain the location and candidate feature dataset for each feature selection algorithm;

[0041] In the union of multiple candidate datasets for location and feature, the signal acquisition location and EEG feature that appear relatively frequently are selected as the final signal acquisition location and final EEG feature.

[0042] According to another specific embodiment of the present invention, the multiple feature selection algorithms include a machine learning-based feature selection algorithm and a statistical feature selection algorithm.

[0043] According to another specific embodiment of the present invention, multiple feature selection algorithms include the Minimum Redundancy Maximum Relevance (MRMR) algorithm, the Chi-square test algorithm, the ReliefF algorithm, the Kruskal-Wallis test algorithm, and the Support Vector Machine-based Recursive Feature Ranking Algorithm (SVM-REF).

[0044] According to another specific embodiment of the present invention, the final signal acquisition location and final EEG characteristics are selected from signal acquisition locations and EEG characteristics located within a preset high degree of difference range based on the sorting results, specifically including:

[0045] A fatigue detection model is acquired, which detects fatigue based on signal acquisition location and EEG characteristics.

[0046] Select the undetermined signal acquisition location and undetermined EEG features from a preset high degree of difference range, input the undetermined signal acquisition location and undetermined EEG features into the fatigue detection model and perform fatigue detection, and determine the accuracy of the fatigue detection model in judging fatigue.

[0047] The number and / or type of pending signal acquisition locations and pending EEG features are changed, and the changed pending signal acquisition locations and pending EEG features are input into the fatigue detection model to determine the accuracy of fatigue judgment, until the minimum number of pending signal acquisition locations and pending EEG features that meet the preset high accuracy conditions are obtained. The last changed pending signal acquisition location and pending EEG feature are determined as the final signal acquisition location and final EEG feature.

[0048] Secondly, embodiments of the present invention disclose a device for determining sensitive brain regions and electroencephalogram (EEG) features for fatigue detection. The sensitive brain region is the area of ​​the head where EEG features are collected. The device includes:

[0049] The signal acquisition module is used to select multiple signal acquisition locations on the head and acquire the corresponding awake and fatigued EEG signals for each signal acquisition location; wherein, the awake EEG signal is the EEG signal in the awake state, and the fatigued EEG signal is the EEG signal in the fatigued state.

[0050] The feature calculation module is used to determine multiple awake EEG features corresponding to awake EEG signals and multiple fatigue EEG features corresponding to fatigue EEG signals based on the awake EEG signals and fatigue EEG signals collected at each signal acquisition location.

[0051] The sensitive brain region and EEG feature determination module is used to determine the degree of difference between the same type of awake and fatigue EEG features at each signal acquisition location, and sort the signal acquisition locations and corresponding EEG features according to the degree of difference to obtain a sorting result. Based on the sorting result, the final signal acquisition location and final EEG feature are selected from the signal acquisition locations and EEG features located within a preset high degree of difference range as the sensitive brain region and the EEG feature used for fatigue judgment; wherein, the preset high degree of difference range is the range with relatively high degree of difference in the sorting.

[0052] Using the above technical solution, the device for determining sensitive brain regions and EEG features for fatigue detection of the present invention can compare the degree of difference between the same type of awake EEG features and fatigue EEG features collected from different signal acquisition locations, and can also compare the sensitivity of different types of EEG features to fatigue. Then, by sorting, signal acquisition locations and EEG features located within a preset high degree of difference range are selected, and then sensitive brain regions and EEG features used for fatigue judgment are selected. It can accurately select sensitive brain regions and EEG features suitable for fatigue detection, and can also make the number of selected sensitive brain regions relatively small and the selected EEG features relatively easy to calculate.

[0053] Thirdly, embodiments of the present invention disclose a fatigue detection method, which uses the sensitive brain regions and EEG characteristics determined by the method for determining sensitive brain regions and EEG characteristics for fatigue detection in any of the foregoing embodiments to perform fatigue detection.

[0054] By adopting the above technical solution, the fatigue detection method of the present invention can achieve accurate fatigue detection with as few sensitive brain regions and EEG features as possible, and can reduce the amount of computation.

[0055] According to another specific embodiment of the present invention, the identified sensitive brain regions include the F3, F4, O1, O2, and Oz electrode positions of the 10-20 international standard lead system, and the identified EEG characteristics include delta wave energy characteristics, theta wave energy characteristics, alpha wave energy characteristics, alpha wave peak characteristics, beta wave energy characteristics, aperiodic shift characteristics, aperiodic stretching transformation characteristics, alpha wave band energy characteristics, and alpha wave band peak characteristics.

[0056] Fourthly, embodiments of the present invention disclose an electronic device, including a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction, when executed by the processor, implements the method for determining sensitive brain regions and EEG features for fatigue detection in any of the foregoing embodiments, and / or implements the fatigue detection method in any of the foregoing embodiments.

[0057] Using the above technical solution, the electronic device can compare the differences between awake and fatigue EEG features acquired from different signal acquisition locations, and also compare the sensitivity of different types of EEG features to fatigue. It then selects signal acquisition locations and EEG features within a preset high-difference range through sorting, and further selects sensitive brain regions and EEG features for fatigue assessment. This allows for accurate selection of suitable sensitive brain regions and EEG features for fatigue detection, while minimizing the number of selected sensitive brain regions and making the selected EEG features easier to calculate. Furthermore, during fatigue detection, accurate fatigue detection can be achieved with as few sensitive brain regions and EEG features as possible, reducing computational load.

[0058] Fifthly, embodiments of the present invention disclose a computer-readable storage medium storing at least one instruction, which, when executed, implements the method for determining sensitive brain regions and EEG features for fatigue detection in any of the embodiments, and / or implements the fatigue detection method in any of the foregoing embodiments.

[0059] Using the above technical solution, the computer-readable storage medium can compare the differences between awake and fatigue EEG features acquired from different signal acquisition locations, and also compare the sensitivity of different types of EEG features to fatigue. Furthermore, by sorting and selecting signal acquisition locations and EEG features within a preset high-difference range, it can further select sensitive brain regions and EEG features for fatigue assessment. This allows for accurate selection of suitable sensitive brain regions and EEG features for fatigue detection, while minimizing the number of selected sensitive brain regions and making the selected EEG features easier to compute. Moreover, during fatigue detection, accurate fatigue detection can be achieved with as few sensitive brain regions and EEG features as possible, reducing computational load. Attached Figure Description

[0060] Figure 1 This diagram illustrates a method for determining sensitive brain regions and EEG features for fatigue detection in an embodiment of the present invention. Figure 1 ;

[0061] Figure 2 This diagram illustrates a method for determining sensitive brain regions and EEG features for fatigue detection in an embodiment of the present invention. Figure 2 ;

[0062] Figure 3 This diagram illustrates a method for determining sensitive brain regions and EEG features for fatigue detection in an embodiment of the present invention. Figure 3 ;

[0063] Figure 4 This diagram illustrates the structure of a device for determining sensitive brain regions and EEG features for fatigue detection in an embodiment of the present invention.

[0064] Figure 5 A schematic diagram of the structure of the electronic device in an embodiment of the present invention is shown;

[0065] Figure 6 A schematic diagram of the logarithmic power spectrum and fitting curve in an embodiment of the present invention is shown;

[0066] Figure 7 A schematic diagram of the periodic energy components of the α-wave spectrum in an embodiment of the present invention is shown;

[0067] Figure 8A A brain topography map showing the statistical p-values ​​corresponding to the δ-wave energy characteristics in an embodiment of the present invention is shown.

[0068] Figure 8B A brain topography map showing the statistical t-values ​​corresponding to the delta wave energy characteristics in an embodiment of the present invention is shown.

[0069] Figure 9A A brain topography map showing the statistical p-values ​​corresponding to the theta wave energy characteristics in an embodiment of the present invention is shown.

[0070] Figure 9B A brain topography map showing the statistical t-values ​​corresponding to the theta wave energy characteristics in an embodiment of the present invention is shown.

[0071] Figure 10A A brain topography map showing the statistical p-values ​​corresponding to the α-wave energy characteristics in an embodiment of the present invention is shown.

[0072] Figure 10B A brain topography map showing the statistical t-values ​​corresponding to the α-wave energy characteristics in an embodiment of the present invention is shown.

[0073] Figure 11A A brain topography map showing the statistical p-values ​​corresponding to the peak energy characteristics of α waves in an embodiment of the present invention is shown.

[0074] Figure 11B A brain topography map showing the statistical t-values ​​corresponding to the peak energy characteristics of the α-wave in an embodiment of the present invention is shown.

[0075] Figure 12A A brain topography map showing the statistical p-values ​​corresponding to the β-wave energy characteristics in an embodiment of the present invention is shown.

[0076] Figure 12B Brain topography maps showing the statistical t-values ​​corresponding to the β-wave energy characteristics in embodiments of the present invention are shown.

[0077] Figure 13A A brain topography map showing the statistical p-values ​​corresponding to the non-periodic scaling transformation features in an embodiment of the present invention is shown.

[0078] Figure 13B Brain topography maps showing the statistical t-values ​​corresponding to the non-periodic scaling transformation features in embodiments of the present invention are shown.

[0079] Figure 14A This invention presents a brain topography map showing the statistical p-values ​​corresponding to the non-periodic shift features in an embodiment of the invention.

[0080] Figure 14B This invention presents a brain topography map showing the statistical t-values ​​corresponding to the non-periodic offset features in an embodiment of the invention.

[0081] Figure 15A This invention presents a brain topography map showing the statistical p-values ​​corresponding to the α-wave band energy characteristics in an embodiment of the invention.

[0082] Figure 15BThis invention presents a brain topography map showing the statistical t-values ​​corresponding to the α-wave band energy characteristics in an embodiment of the invention.

[0083] Figure 16A This invention presents a brain topography map showing the statistical p-values ​​corresponding to the peak characteristics of the α-wave band in an embodiment of the invention.

[0084] Figure 16B This invention presents a brain topography map showing the statistical t-values ​​corresponding to the peak characteristics of the α-wave band in an embodiment of the invention.

[0085] Figure 17 A statistical diagram showing the frequency of occurrence of different EEG features in embodiments of the present invention is shown;

[0086] Figure 18 This illustration shows brain topography maps of sensitive brain regions obtained using the chi-square test algorithm in an embodiment of the present invention.

[0087] Figure 19 This illustration shows brain topography maps of sensitive brain regions obtained using the Minimum Redundancy Maximum Correlation (MRMR) algorithm in an embodiment of the present invention.

[0088] Figure 20 This shows brain topography maps of sensitive brain regions obtained using the Kruskal-Wallis test algorithm in an embodiment of the present invention;

[0089] Figure 21 This shows brain topography maps of sensitive brain regions obtained using the ReliefF algorithm in an embodiment of the present invention;

[0090] Figure 22 This illustration shows brain topography maps of sensitive brain regions obtained using the Support Vector Machine-based Recursive Feature Ranking (SVM-REF) algorithm in an embodiment of the present invention.

[0091] Figure 23 This invention presents a brain topography map of a sensitive brain region obtained by integrating multiple feature selection algorithms in an embodiment of the invention. Detailed Implementation

[0092] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Although the description of the present invention is presented in conjunction with preferred embodiments, this does not mean that the features of the invention are limited to these embodiments. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of the present invention. To provide a deep understanding of the invention, many specific details will be included in the following description. The invention may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of the invention, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0093] It should be noted that in this specification, similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0094] The terms “first”, “second”, etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0095] In the description of this embodiment, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set up," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this embodiment based on the specific circumstances.

[0096] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0097] Electroencephalography (EEG) fatigue detection is a technique that uses electroencephalogram (EEG) signals to assess an individual's fatigue state. This technique is based on the fact that EEG waves reflect the brain's electrical activity, and these activity patterns change with the brain's fatigue state. EEG waves are generally divided into several main bands, including delta waves (0.5-4 Hz), theta waves (4-8 Hz), alpha waves (8-14 Hz), and beta waves (14-30 Hz). These bands are associated with different brain states; for example, alpha waves are associated with relaxation, while theta waves are associated with fatigue or deep thinking. In EEG fatigue detection, researchers analyze the time-domain characteristics (such as amplitude and power spectral density) and frequency-domain characteristics (such as band power) of EEG waves to identify changes in EEG patterns specific to normal sleep, wakefulness, and fatigue stages.

[0098] Current fatigue detection technologies using electroencephalograms (EEGs) typically require complex equipment to collect EEG signals from multiple locations on the head, sometimes as many as 64. These EEG signals correspond to various types of EEG features, and existing technologies, in order to achieve high-precision fatigue detection, also employ complex computational methods to process the EEG data and calculate these features. This current fatigue detection method involves large amounts of data collection, computational demands, high costs, and limited applicability, making it unsuitable for use by pilots during flight and failing to meet their needs for rapid and accurate fatigue detection. Therefore, it is necessary to further reduce the number of locations on the head where EEG signals are collected, and to reduce or simplify the EEG indicators that need to be calculated, while still ensuring high-precision fatigue detection, so as to improve flight safety during pilot operations.

[0099] Firstly, reference Figure 1 The present invention discloses a method for determining sensitive brain regions and electroencephalogram (EEG) features for fatigue detection. The sensitive brain regions are areas of the head where EEG features are collected. The determination method includes the following steps:

[0100] S1: Select multiple signal acquisition locations on the head and acquire the corresponding awake EEG and fatigue EEG signals for each signal acquisition location.

[0101] Among them, the awake EEG signal is the EEG signal in a conscious state, and the fatigue EEG signal is the EEG signal in a fatigued state.

[0102] S2: Based on the awake and fatigued EEG signals collected at each signal acquisition location, determine multiple awake EEG characteristics corresponding to the awake EEG signals and multiple fatigued EEG characteristics corresponding to the fatigued EEG signals.

[0103] S3: Determine the degree of difference between the same type of awake EEG features and fatigue EEG features at each signal acquisition location, and sort the signal acquisition locations and corresponding EEG features according to the degree of difference to obtain the sorting results. Based on the sorting results, select the final signal acquisition location and final EEG features from the signal acquisition locations and EEG features located in the preset high degree of difference range as the sensitive brain regions and EEG features used for fatigue judgment.

[0104] The preset high degree of difference range is the range with relatively high degree of difference in the ranking.

[0105] More specifically, "relatively high" refers to a higher degree of difference between awake and fatigue EEG characteristics of the same type at a specific signal acquisition location compared to the same type of awake and fatigue EEG characteristics at other signal acquisition locations within the ranking results. Alternatively, it could mean that among multiple types of awake and fatigue EEG characteristics at a specific signal acquisition location, the degree of difference between a certain type of awake and fatigue EEG characteristic is higher than other types of awake and fatigue EEG characteristics. Or, it could mean that the degree of difference between the same type of awake and fatigue EEG characteristics at a specific signal acquisition location is higher than the degree of difference between other types of awake and fatigue EEG characteristics at other signal acquisition locations. Signal acquisition locations and EEG characteristics within a preset high degree of difference correspond to higher degrees of difference in awake and fatigue EEG characteristics compared to other signal acquisition locations and other EEG characteristics.

[0106] Specifically, the preset high degree of difference range can be, for example, the top 50% of the ranking results. Further, the preset high degree of difference range can be, for example, the top 20%-30% of the ranking results. Even further, the preset high degree of difference range can be, for example, the top 5%-10% of the ranking results. The higher the preset high degree of difference range is in the overall ranking results, the more beneficial it is for subsequently selecting EEG features and sensitive brain regions that are more sensitive to fatigue.

[0107] Using the above technical solution, the method for determining sensitive brain regions and EEG features for fatigue detection in this invention can compare the degree of difference between awake and fatigue EEG features of the same type collected from different signal acquisition locations, and can also compare the sensitivity of different types of EEG features to fatigue. Furthermore, by sorting, signal acquisition locations and EEG features within a preset high degree of difference range are selected, and then sensitive brain regions and EEG features for fatigue judgment are selected. This method can accurately select suitable sensitive brain regions and EEG features for fatigue detection, and also results in a relatively small number of selected sensitive brain regions and relatively easier-to-calculate selected EEG features. Compared with the existing technology that typically detects EEG signals from multiple locations on the head (e.g., 64 locations) for fatigue judgment, the method of this invention can select a relatively small number of final signal acquisition locations and a relatively small number of final EEG features, eliminating most signal acquisition locations and EEG features that have low contribution to fatigue detection and are basically irrelevant to fatigue, thereby reducing the computational load.

[0108] Furthermore, the final signal acquisition location and the final EEG characteristics can each be multiple.

[0109] In the above embodiments, the types of awake and fatigued EEG characteristics can include delta wave energy characteristics, theta wave energy characteristics, alpha wave energy characteristics, alpha wave peak characteristics, and beta wave energy characteristics. All five characteristics can be calculated using the power spectrum obtained by performing a digital discrete Fourier transform on the EEG signal.

[0110] The inventors discovered that aperiodic features in EEG characteristics can also be used for fatigue assessment. These aperiodic features include, for example, aperiodic offset features and aperiodic scaling transform features. In the above embodiments, the types of awake and fatigued EEG characteristics include aperiodic offset features and aperiodic scaling transform features. The aperiodic offset features and aperiodic scaling transform features of the EEG signal are determined in the following way:

[0111] S01: Determine the logarithmic power spectrum of the EEG signal.

[0112] refer to Figure 6 , Figure 6 Curve 2 in the figure represents the logarithmic power spectrum of the EEG signal. Specifically, the logarithmic power spectrum of the EEG signal can be calculated using the following formula 4:

[0113]

[0114] Where y is the power spectrum obtained after performing a digital discrete Fourier transform on the EEG signal, L is the length of the digital discrete Fourier transform, and F is the logarithmic power spectrum. The length of the digital discrete Fourier transform can be the same as the length of the EEG signal.

[0115] S02: Based on the logarithmic power spectrum, perform nonlinear fitting using the following formula 1 to obtain the corresponding fitting curve:

[0116] L(f)=ab×log(P(f))(Formula 1)

[0117] Where L(f) is the fitted curve, P(f) is the logarithmic power spectrum, a is the offset parameter of the fitted curve, b is the scaling transformation parameter of the fitted curve, a and b are determined by least squares fitting, and f is the frequency of the EEG signal, with a range of 2-5Hz and 14-30Hz.

[0118] refer to Figure 6 In this embodiment, Figure 6 Curve 1 in the figure is the fitted curve, where 'a' represents the left and right offset of curve 1, and 'b' represents the scaling transformation of curve 1. By selecting a frequency range of 2-5Hz and 14-30Hz for fitting, the α band is avoided. This reduces the impact on the α band during fitting, allowing for more accurate α band energy to be obtained subsequently.

[0119] S03: Determine that the offset parameter a in Formula 1 is an aperiodic offset feature, and determine that the scaling transformation parameter b in Formula 1 is an aperiodic scaling transformation feature.

[0120] In this invention, the aperiodic offset features and aperiodic scaling transformation features calculated by curve fitting using the above methods and formulas have a relatively high sensitivity to fatigue, which helps to improve the accuracy of fatigue detection.

[0121] Furthermore, research revealed that the alpha wave energy and peak characteristics commonly used in existing technologies are not sensitive enough to fatigue. In response, the inventors further improved the EEG characteristics corresponding to the alpha wave selected for fatigue detection. Their research showed that the alpha wave frequency band energy and peak characteristics are more sensitive to fatigue.

[0122] In the above embodiments, the types of awake EEG characteristics and fatigue EEG characteristics also include alpha wave band energy characteristics and alpha wave band peak characteristics. The alpha wave band energy characteristics and alpha wave band peak characteristics of the EEG signal are determined in the following way:

[0123] S04: Determine the periodic components of the α-wave spectrum using the following formula 2 based on the logarithmic power spectrum and the fitted curve:

[0124] P ′ (f)=P(f)-L(f)(Formula 2)

[0125] Among them, P ′ (f) represents the periodic component of the α-wave spectrum, L(f) is the fitted curve, P(f) is the logarithmic power spectrum, and f is the frequency of the EEG signal.

[0126] S05: The periodic energy component of the α-wave spectrum is determined based on the periodic component of the α-wave spectrum using the following formula 3:

[0127] P ″ (f) = 10 × 10 P′(f) (Formula 3)

[0128] Among them, P ′ (f) represents the periodic component of the α-wave spectrum, P ″ (f) represents the periodic energy component of the alpha wave spectrum, where f is the frequency of the EEG signal. (Reference) Figure 7 , Figure 7 Curve 3 in the figure represents the periodic energy component of the α-wave spectrum.

[0129] S06: Determine the energy characteristics and peak characteristics of the α-wave band based on the periodic energy components of the α-wave spectrum.

[0130] The periodic energy component P of the α-wave spectrum can be obtained through the above formula 3. ″ (f), α-wave spectral periodic energy component P″ (f) Dimensionless, and can then be determined according to P. ″ (f) Calculate the energy corresponding to the α band, i.e., the α-wave frequency band energy characteristics, and also obtain the α-wave frequency band peak characteristics. The α-wave frequency band energy characteristics can correspond to the periodic energy component P of the α-wave spectrum. ″ (f) The area enclosed by the coordinate axes in the α band (i.e., 8-13Hz) represents the peak characteristics of the α wave frequency band, which can correspond to the periodic energy component P of the α wave spectrum. ″ The peak value of (f).

[0131] The alpha wave band energy characteristics and alpha wave band peak characteristics calculated using the above methods and formulas are more sensitive to fatigue and help improve the accuracy of fatigue detection.

[0132] Further, refer to Figure 2 The aforementioned step S1 specifically includes the following steps:

[0133] S11: Collect the corresponding raw awake EEG signal and raw fatigue EEG signal at each of the multiple signal acquisition locations on the head.

[0134] Specifically, brainwaves of subjects can be collected separately in a conscious state and a fatigued state. Raw conscious brainwave signals and raw fatigued brainwave signals are obtained respectively.

[0135] S12: Preprocess the raw awake EEG signal and the raw fatigue EEG signal to obtain the awake EEG signal and the fatigue EEG signal. The preprocessing method includes at least one of the following: data segmentation, averaging, filtering, and windowing.

[0136] Specifically, data segmentation refers to dividing the collected EEG signals into segments to increase the amount of data and improve its comprehensiveness. This enhances the general applicability of the selected sensitive brain regions and EEG features used for fatigue detection. It also facilitates subsequent verification of the effectiveness of the selected sensitive brain regions and EEG features for fatigue detection, and helps in building a more stable and accurate fatigue detection model. For example, the EEG signal can be segmented in steps with a certain step size. For instance, for a 3-minute EEG signal, each segment could be 4-10 seconds long, with a step size of 1-5 seconds.

[0137] De-averaging can be used to eliminate baselines, perform baseline correction, and improve the accuracy of subsequent analysis of EEG signals.

[0138] The filtering can employ traditional FIR (Finite Impulse Response) filtering, which maintains the stability of phase changes and has a bandpass range of f1-f2 Hz. Specifically, f1 ranges from 0.5-2 Hz, and should preserve as many low-frequency bands in the EEG as possible, such as the delta band (0.5-4 Hz); f2 ranges from 30-45 Hz, and should preserve as many high-frequency bands in the EEG as possible, such as the beta band (14-30 Hz). For example, the bandpass range could be 1-45 Hz.

[0139] Windowing refers to applying a window to the EEG signal, which can reduce the influence of endpoint data on the spectral measurement of subsequent signals. Windowing options include Hamming windows and Hanning windows.

[0140] Using any one of the aforementioned preprocessing methods is beneficial for subsequent analysis of EEG signals. When all of these preprocessing methods are used, even better results can be achieved, further improving the accuracy of EEG signal analysis.

[0141] Furthermore, the degree of difference between awake and fatigue EEG features of the same type was determined using a feature selection algorithm. Analyzing the degree of difference between awake and fatigue EEG features using a feature selection algorithm helps to identify the types of EEG features that contribute more to fatigue detection.

[0142] refer to Figure 3 In step S3 above, the final signal acquisition location and final EEG characteristics are selected from the signal acquisition locations and EEG characteristics within a preset high degree of difference range based on the sorting results. This specifically includes the following steps:

[0143] S31: Determine the frequency of occurrence of different signal acquisition locations and different EEG features within a preset high degree of difference range.

[0144] S32: Select the signal acquisition locations and EEG features that appear relatively frequently as the location and feature candidate datasets corresponding to the feature selection algorithm, and select the final signal acquisition locations and final EEG features from the location and feature candidate datasets.

[0145] Among them, signal acquisition locations and EEG features that appear relatively frequently within the preset high degree of difference range can be considered as signal acquisition locations and EEG features that are relatively more sensitive to fatigue, which helps to improve the accuracy of fatigue detection.

[0146] More specifically, "relatively high frequency of occurrence" can refer to the fact that, within a preset range of high variability, specific signal acquisition locations and specific types of EEG features occur more frequently than other signal acquisition locations and other types of EEG features.

[0147] For example, signal acquisition locations and EEG features can be sorted according to their frequency of occurrence to obtain a ranking result. Signal acquisition locations and EEG features with relatively high frequency of occurrence refer to those within a preset range of high frequency of occurrence in the ranking result. Signal acquisition locations and EEG features within the preset range of high frequency of occurrence occur more frequently than other signal acquisition locations and other EEG features. The preset range of high frequency of occurrence could be, for example, the top 80% in the ranking result. Further, the preset range of high frequency of occurrence could be, for example, the top 40%-60% in the ranking result. Even further, the preset range of high frequency of occurrence could be, for example, the top 10%-25% in the ranking result. The higher the preset range of high frequency of occurrence appears in the overall ranking result, the more beneficial it is for subsequently selecting EEG features and sensitive brain regions that are more sensitive to fatigue.

[0148] Furthermore, in the above embodiments, there are multiple feature selection algorithms. In step S3, selecting the final signal acquisition location and final EEG features from signal acquisition locations and EEG features within a preset high degree of difference range based on the sorting results further includes:

[0149] S33: Obtain the location and candidate feature dataset corresponding to each feature selection algorithm.

[0150] Specifically, each feature selection algorithm corresponds to a location and a dataset of candidate features, and multiple feature selection algorithms can obtain multiple datasets of candidate locations and features.

[0151] S34: In the union of multiple candidate datasets of locations and features, select the signal acquisition location and EEG feature that appear relatively frequently as the final signal acquisition location and final EEG feature.

[0152] When multiple feature selection algorithms are used, there are also multiple candidate datasets for locations and features, resulting in a union of these datasets. Accordingly, the final signal acquisition location and final EEG feature can be selected from this union. Specifically, the signal acquisition location and EEG feature that appear relatively frequently in the union of the multiple candidate datasets can be chosen as the final signal acquisition location and final EEG feature. The union refers to the sum of all elements in the multiple candidate datasets; that is, in the union, the frequency of each signal acquisition location is the sum of the frequencies of each signal acquisition location in each of the multiple candidate datasets, and the frequency of each EEG feature is the sum of the frequencies of each EEG feature in each of the multiple candidate datasets.

[0153] For example, the signal acquisition locations and EEG features can be sorted again according to their frequency of occurrence in the union of multiple locations and the candidate feature datasets, resulting in a ranking result. This ranking result can be denoted as the second ranking result. Here, the signal acquisition locations and EEG features with relatively high frequency of occurrence can refer to those within a second preset range of frequency of occurrence in the second ranking result. Signal acquisition locations and EEG features within the second preset range of frequency of occurrence appear more frequently in the union of multiple locations and the candidate feature datasets compared to other signal acquisition locations and other EEG features. For example, the second preset range of frequency of occurrence can be the top 80% in the second ranking result. Further, the second preset range of frequency of occurrence can be the top 40%-60% in the second ranking result. Even further, the second preset range of frequency of occurrence can be the top 10%-25% in the second ranking result. The higher the second preset range of frequency of occurrence appears in the entire second ranking result, the more beneficial it is for subsequently selecting EEG features and sensitive brain regions that are more sensitive to fatigue.

[0154] Furthermore, the signal acquisition location and EEG feature that appear most frequently can be selected from the candidate dataset of location and feature as the final signal acquisition location and final EEG feature, or the signal acquisition location and EEG feature that appear most frequently can be selected from the union of multiple candidate datasets of location and feature as the final signal acquisition location and final EEG feature.

[0155] When a specific signal acquisition location and a specific EEG feature appear relatively frequently in the union of multiple locations and candidate feature datasets, it indicates that this specific signal acquisition location and specific EEG feature rank relatively high and appear frequently when the difference between wakefulness and fatigue is calculated and ranked using different feature selection algorithms. This demonstrates high stability in the degree of difference and indicates greater sensitivity to fatigue. The final signal acquisition location and final EEG feature selected using the above method are more sensitive to fatigue while maintaining a relatively small number of occurrences.

[0156] Furthermore, multiple feature selection algorithms are included, such as machine learning-based and statistical-based algorithms. By employing feature selection methods from various fields, including machine learning and statistics, to evaluate and rank the differences between awake and fatigued EEG features, the evaluation method is more comprehensive, the evaluation results are more accurate and stable, and it helps to determine more reliable final signal acquisition locations and final EEG features, thus increasing versatility.

[0157] Furthermore, in the above embodiments, there may be 3 to 10 feature selection algorithms. Preferably, there may be 4 to 7 feature selection algorithms. This setting ensures the comprehensiveness of the evaluation method without having too many evaluation methods or too much computation, and can select EEG features and signal acquisition locations that are highly versatile, reliable, and more sensitive to fatigue.

[0158] For example, several feature selection algorithms include the Minimum Redundancy Maximum Relevance (MRMR) algorithm, the chi-square test algorithm, the ReliefF algorithm, the Kruskal-Wallis test algorithm, and the Support Vector Machine-based Recursive Feature Ranking (SVM-REF) algorithm.

[0159] The Minimum Redundancy Maximum Relevance (MRMR) algorithm, also known as the fscmrmr method, aims to find a subset of features from a feature set that has the highest relevance to the target variable and the lowest redundancy among the features.

[0160] The chi-square test algorithm, denoted as the fscchi2 method, predicts the variable most influential on the classification result. In this embodiment, the classification result could be, for example, a division into two categories: alert and fatigued.

[0161] The ReliefF algorithm, also known as the Relieff method, ranks the importance of predictor variables. The ReliefF algorithm is based on the idea that good features should reduce the distance between similar samples and increase the distance between dissimilar samples.

[0162] The Kruskal-Wallis test algorithm, also known as the Kruskal-Wallis method, is a nonparametric statistical method used to determine whether there is a statistically significant difference between the medians of three or more independent groups. The Kruskal-Wallis test can be performed on data using statistical software or programming languages ​​such as Python or R. The test results typically include an H-value (Kruskal-Wallis statistic) and a p-value. During the Kruskal-Wallis test, the rank of each sample is calculated. The rank is the ordinal position of a sample value within all the pooled datasets, and values ​​can be ordered according to their rank.

[0163] The recursive feature ranking algorithm based on support vector machines (SVM-REF) can be denoted as SVMRFECBR. SVM-RFE (Recursive Feature Elimination with Support Vector Machines) is a feature selection method that reduces the number of features by recursively removing features that have the least impact on the model.

[0164] Furthermore, in the above embodiments, step S3, which involves selecting the final signal acquisition location and final EEG features from signal acquisition locations and EEG features within a preset high degree of difference range based on the sorting results, may specifically include the following steps:

[0165] S35: Obtain the fatigue detection model, which detects fatigue by using signal acquisition location and EEG characteristics.

[0166] For example, the following machine learning models can be used to construct fatigue detection models: FineTree, MediumTree, CoarseTree, LinearDiscriminant, Quadratic Discriminant, Binary GLM Logistic Regression, Efficient Logistic Regression, Efficient Linear SVM, Gaussian Naive Bayes, Kernel Naive Bayes, Linear SVM, Quadratic SVM, Cubic SVM, Fine Gaussian SVM, Medium Gaussian SVM, Coarse Gaussian SVM, Fine KNN, and Medium KNN. The following are some examples of machine learning models used to construct fatigue detection models: KNN, Coarse KNN, Cosine KNN, Cubic KNN, Weighted KNN, Boosted Trees, Bagged Trees, Subspace Discriminant Analysis, Subspace KNN, RUS Boosted Trees (RUSBoosted Trees), Narrow Neural Networks, Medium Neural Networks, Wide Neural Networks, Bilayered Neural Networks, Trilayered Neural Networks, SVM Kernel, and Logistic Regression Kernel. Other types of machine learning models can also be used to construct fatigue detection models; this invention does not limit this, as long as the fatigue detection model can detect fatigue based on signal acquisition location and EEG characteristics.

[0167] S36: Select the undetermined signal acquisition location and undetermined EEG features from the preset high difference range, input the undetermined signal acquisition location and undetermined EEG features into the fatigue detection model and perform fatigue detection, and determine the accuracy of the fatigue detection model in judging fatigue.

[0168] For example, accuracy can be determined by calculating the confusion matrix of a fatigue detection model, which can be used to evaluate the model's performance. More specifically, the accuracy of the fatigue detection model on a test set can be calculated. The test set consists of EEG signals from awake and fatigued individuals. Random EEG signals from the test set are input into the fatigue detection model and judged. The judgment result is compared with the actual state corresponding to the input EEG signals to the model to determine the accuracy of the fatigue detection model in judging fatigue. More specifically, accuracy is the proportion of data correctly judged by the model out of the total number of data points.

[0169] S37: Change the number and / or type of the pending signal acquisition location and pending EEG features, and input the changed pending signal acquisition location and pending EEG features into the fatigue detection model to determine the accuracy of fatigue judgment, until the minimum number of pending signal acquisition locations and pending EEG features that meet the preset high accuracy conditions are obtained, and the last change of the corresponding pending signal acquisition location and pending EEG features is determined as the final signal acquisition location and final EEG feature.

[0170] This can be achieved by updating the number and / or type of pending signal acquisition locations and pending EEG features based on signal acquisition locations and EEG features within a preset range of high differences.

[0171] Furthermore, when using feature selection algorithms to rank the degree of difference, the number and / or type of the pending signal acquisition locations and pending EEG features can be updated based on the location corresponding to the feature selection algorithm and the signal acquisition locations and EEG features within the candidate feature dataset. Even further, when multiple feature selection algorithms are used, the number and / or type of the pending signal acquisition locations and pending EEG features can be updated based on the signal acquisition locations and EEG features within the intersection of multiple locations and candidate feature datasets, or based on the signal acquisition locations and EEG features within the intersection of multiple locations and candidate feature datasets.

[0172] For example, changing the number and / or type of pending signal acquisition locations and pending EEG features can specifically include gradually reducing the number of pending signal acquisition locations or gradually reducing the types of EEG features based on the ranking results obtained by the aforementioned method. When changing the number and / or type of pending signal acquisition locations and pending EEG features, priority is given to retaining pending signal acquisition locations and pending EEG features with higher accuracy. Specifically, when it is necessary to reduce the number of pending signal acquisition locations or gradually reduce the types of EEG features, priority can be given to removing signal acquisition locations and EEG features that rank relatively low in the ranking results or appear relatively infrequently within a preset high degree of difference range, that is, priority can be given to excluding signal acquisition locations and EEG features with low correlation to fatigue. In one possible embodiment, if the difference in accuracy after removing only pending signal acquisition location A and removing only pending signal acquisition location B is small, then priority is given to retaining the re-determined signal acquisition location with higher accuracy and removing the re-determined signal acquisition location with lower accuracy. Furthermore, the number of signal acquisition locations can be changed first, while the EEG features remain unchanged. After the final signal acquisition locations are determined, a similar approach can be used to gradually reduce the types of calculated EEG features. That is, when removing EEG features, the types of EEG features with low relevance are removed first, and when replacing EEG features, the types of EEG features with higher accuracy that have been re-determined are retained first.

[0173] Similarly, changing the number and / or type of pending signal acquisition locations and pending EEG features can specifically include gradually increasing the number of pending signal acquisition locations or gradually increasing the types of calculated EEG features based on the ranking results obtained by the aforementioned method. When changing the number and / or type of pending signal acquisition locations and pending EEG features, priority should be given to pending signal acquisition locations and pending EEG features with higher accuracy. Specifically, when it is necessary to increase the number of pending signal acquisition locations or gradually increase the types of EEG features, priority can be given to signal acquisition locations and EEG features that rank relatively high in the ranking results or appear relatively frequently within a preset high degree of difference range, that is, priority can be given to signal acquisition locations and EEG features that are highly correlated with fatigue. In one possible embodiment, if the difference in accuracy after only increasing pending signal acquisition location A and only increasing pending signal acquisition location B is small, priority can be given to increasing the re-determined signal acquisition location with higher accuracy. Furthermore, the number of signal acquisition locations can be changed first, while the EEG features remain unchanged. After the final signal acquisition locations are determined, the types of calculated EEG features can be gradually increased or replaced in a similar manner. That is, when increasing EEG features, the types of EEG features with high relevance should be increased first, and when replacing EEG features, the types of EEG features with higher accuracy that have been re-determined should be retained first.

[0174] The preset high accuracy condition can be, for example, a set accuracy threshold. When the determined accuracy is greater than or equal to the accuracy threshold, the preset high accuracy condition can be considered met. For example, the accuracy threshold can be 0.75-0.95. Further, the accuracy threshold can also be 0.80-0.90.

[0175] Furthermore, the number and / or type of signal acquisition locations can be adjusted first, while the number and type of EEG features remain unchanged. For example, the number of pending signal acquisition locations can be increased, decreased, or replaced. Simultaneously, all nine types of pending EEG features are selected, and the accuracy is determined. When the determined accuracy meets a preset high accuracy condition, the last adjusted pending signal acquisition location can be determined as the final signal acquisition location. After determining the final signal acquisition location, the type and / or number of EEG features can be adjusted. For example, the types of EEG features can be gradually reduced or replaced to minimize the number of calculated EEG features while maintaining high accuracy. Using the above method, EEG features and signal acquisition locations with the fewest possible numbers, the least computational load, and high fatigue detection accuracy can be determined.

[0176] A more specific embodiment is described below.

[0177] Multiple signal acquisition sites were selected on the head, and awake and fatigue EEG signals were acquired for each site. Awake EEG signals were those received during the waking state, and fatigue EEG signals were those received during the fatigued state. Subsequently, based on the awake and fatigue EEG signals acquired at each site, various awake EEG characteristics and fatigue EEG characteristics were determined. The degree of difference between the same type of awake and fatigue EEG characteristics at each site was determined, and the site and its corresponding EEG characteristics were ranked according to the magnitude of the difference, resulting in a ranking.

[0178] The degree of difference between awake and fatigued EEG features of the same type was determined using feature selection algorithms. Specifically, five feature selection algorithms were employed: Minimum Redundancy Maximum Correlation (MRMR) algorithm, chi-square test algorithm, ReliefF algorithm, Kruskal-Wallis test algorithm, and Support Vector Machine-based Recursive Feature Ranking (SVM-REF) algorithm. These five algorithms yielded five ranking results, denoted as rank1, rank2, rank3, rank4, and rank5, respectively.

[0179] In this experiment, 61 signal acquisition locations were set according to the 10-20 International Electroencephalogram (EEG) Leads definition, namely: 'Fp1', 'AF3', 'AF7', 'Fz', 'F1', 'F3', 'F5', 'F7', 'FC1', 'FC3', 'FC5', 'FT7', 'Cz', 'C1', 'C3', 'C5', 'T7', 'CP1', 'CP3', 'CP5', 'TP7', 'TP9', 'Pz', 'P1', 'P3', 'P5', 'P7', 'P O3', 'PO7', 'Oz', 'O1', 'Fpz', 'Fp2', 'AF4', 'AF8', 'F2', 'F4', 'F6', 'F8', 'FC2', 'FC4', 'FC6', 'FT8', 'C2', ' C4', 'C6', 'T8', 'CPz', 'CP2', 'CP4', 'CP6', 'TP8', 'TP10', 'P2', 'P4', 'P6', 'P8', 'POz', 'PO4', 'PO8', 'O2'. There are a total of nine EEG features, namely the delta wave energy feature (denoted as DeltaPower), theta wave energy feature (denoted as ThetaPower), the alpha wave energy feature (denoted as AlphaPower), the alpha wave peak feature (denoted as AlphaPowerPeak), the beta wave energy feature (denoted as BetaPower), the aperiodic shift feature (denoted as aAperiodic), the aperiodic scaling transformation feature (denoted as bAperiodic), the alpha wave band energy feature (denoted as AlphaPeriodicPower), and the alpha wave band peak feature (denoted as AlphaPeriodicPowerPeak). Therefore, a total of 61 * 9 = 549 features need to be sorted.

[0180] For sensitive brain regions, the frequency of occurrence of specific signal acquisition locations is counted in the top 50 of each ranking result from rank 1 to rank 5 (a total of 549 features). Alternatively, the top 9% of all features can be selected, which in this embodiment is the top 9% of the 549 features. The more frequently a specific signal acquisition location appears, the more important that location is, and the more sensitive it is to fatigue. EEG signals acquired at this location help improve the accuracy of fatigue detection. For example, if "signal acquisition location 1" appears 10 times and "signal acquisition location 2" appears 6 times in the top 50 of rank 1, then "signal acquisition location 1" is more important than "signal acquisition location 2." Similarly, the advantageous signal acquisition locations for each of rank 1, rank 2, rank 3, rank 4, and rank 5 can be obtained, thus identifying sensitive brain regions. Preferably, the union of the advantageous signal acquisition locations for each of rank 1, rank 2, rank 3, rank 4, and rank 5 can be obtained to determine the optimal signal acquisition location under various ranking methods, thereby identifying sensitive brain regions that are more conducive to accurate fatigue detection.

[0181] For the advantageous EEG features used in fatigue detection, we can also count the frequency of occurrence of the above nine types of EEG features in the top 50 features of each ranking result (or we can select the top 9% of all features, which is the top 9% of 549 features in this embodiment) in rank1, rank2, rank3, rank4, and rank5. The more times a specific type of EEG feature appears, the more important that feature is and the more sensitive it is to fatigue. For example, if the delta wave energy feature appears 10 times and the theta wave energy feature appears 6 times in the top 50 of rank1, then the importance of the delta wave energy feature is greater than that of the theta wave energy feature. And so on, we can realize the advantageous features of each of rank1, rank2, rank3, rank4, and rank5 respectively. Combining the five ranking results, we can obtain the following: Figure 17 A statistical diagram. (Through...) Figure 17 It can be seen that the alpha wave frequency band energy characteristics and alpha wave frequency band peak characteristics calculated by this invention are more sensitive to fatigue than the alpha wave energy characteristics and alpha wave peak characteristics traditionally used in the prior art. Preferably, by taking the union of the advantageous EEG characteristics of rank1, rank2, rank3, rank4 and rank5, the optimal EEG characteristics under various ranking methods can be obtained, and thus the type of EEG characteristics that is more helpful for accurate fatigue detection can be determined.

[0182] In constructing the fatigue detection model, in this embodiment, the initial signal acquisition positions to be input into the fatigue detection model are selected based on the maximum allowed number of channels for fatigue detection and the sorting results obtained by the aforementioned feature selection algorithm (which can be the sorting results obtained by one feature selection algorithm or the sorting results obtained by a combination of multiple feature selection algorithms). Alternatively, a combination of multiple signal acquisition positions can be selected based on the specific usage environment.

[0183] Subsequently, full feature modeling is performed, meaning the initial input EEG features to the fatigue detection model are the nine types mentioned above. The datasets used for training and testing the model can be divided into training, validation, and test sets, with the data in the test set not included in the training or validation sets. The training set is used to train the model. The test set is used to evaluate the model's generalization ability. After the model is trained on the training set, its performance can be tested using the test set, for example, by calculating metrics such as accuracy and recall, or by obtaining the corresponding confusion matrix. The validation set is used for model selection and hyperparameter tuning, helping to avoid over-optimizing the model on the training set and ensuring that the model performs well on unseen data.

[0184] After the fatigue detection model is built, the accuracy of the test set can be compared to see if it meets the accuracy requirements for practical application. If not, the number of signal acquisition locations used in the fatigue detection model can be increased or decreased until the accuracy requirements are met. After determining the number and type of signal acquisition locations, the number of EEG features used in the fatigue detection model can be reduced based on the previously obtained ranking results, prioritizing the reduction of EEG features with low contribution. Simultaneously, the accuracy is checked to see if it meets the accuracy requirements; if so, the number of features can be further reduced. Using this method, a fatigue detection model with a relatively small number of signal acquisition locations, a relatively small number of selected EEG features, low computational cost, and satisfactory accuracy can be constructed.

[0185] The following specific experiments demonstrate the effectiveness of the methods described in the above embodiments of the present invention, as well as the accuracy of the sensitive brain regions identified by the methods described in the above embodiments and the EEG features used for fatigue detection.

[0186] In the validation experiments, a publicly available dataset was used. The dataset comes from the paper "a resting-state EEG dataset for sleep deprivation" published in ScientificData in 2024, and is described in detail below:

[0187] Regarding the participants, the experiment involved 71 participants with a mean age of 20 years (standard deviation 1.44 years) and an age range of 17–23 years. There were 34 females and 37 males. All participants reported no symptoms of mental illness, anxiety or depression, recent cold symptoms, or sleep problems such as insomnia, sudden awakenings, or difficulty breathing.

[0188] Regarding the experimental design, the experiment was conducted at the Sleep and Neuroimaging Center of Southwest University, spanning from March 2019 to October 2021. A within-subject design was used, meaning each participant experienced both normal sleep (NS) and sleep deprivation (SD). The two conditions were kept as consistent as possible throughout the day (morning or afternoon), with most participants experiencing a time difference of less than 1.5 hours between NS and SD. The interval between the two conditions was at least 7 days and at most one month to eliminate the sequence effect. The duration of sleep deprivation was 24 to 30 hours.

[0189] The experimental procedure involved participants undergoing tests under both NS and SD conditions in the laboratory. Prior to the EEG testing, participants monitored their sleep quality using a sleep diary or activity tracker to ensure adequate sleep before the experiment. Participants completed the Psychomotor Vigilance Task (PVT) test, as well as scales assessing mood and drowsiness, including the Positive and Negative Affect Scale (PANAS), Automatic Thoughts Questionnaire (ATQ), State Anxiety Inventory (SAI), Stanford Sleepiness Scale (SSS), Karolinska Sleepiness Scale (KSS), and a simplified version of the sleep diary. During EEG data collection, participants were asked to sit still with their eyes open for 5 minutes, followed by 5 minutes with their eyes closed (for some participants). Sixty-one Ag / AgCl active electrodes were used, arranged according to the extended 10-20 international electrode placement system, with FCz as the online reference electrode. The sampling rate was 500 Hz, and the electrode impedance was maintained below 5 kΩ.

[0190] Sleep deprivation (SD) conditions: Participants arrived at the laboratory at 9:00 PM the night before the experiment. Throughout the deprivation period, participants were continuously monitored by two experimenters to ensure they remained awake until the end of the experiment the following day. Participants were prohibited from consuming caffeinated, tea, or alcoholic beverages or foods, and were not allowed to lie down, sleep, or engage in strenuous exercise. There was no recovery sleep period after SD; participants immediately underwent the experiment.

[0191] The data was recorded as follows: The acquired dataset contained EEG data from 71 participants with their eyes open and EEG data from 38 participants with their eyes closed, with the recording duration standardized to 5 minutes. The uploaded data was not preprocessed.

[0192] The data preprocessing and feature extraction methods are as follows: Signal acquisition locations are the aforementioned 61 locations. The acquired EEG signals are preprocessed according to the data segmentation, averaging, filtering, and windowing procedures described in the embodiments of this invention. The aforementioned nine EEG features are calculated, and the aforementioned five feature selection algorithms are used for feature extraction and sorting. A fatigue detection model is constructed and trained using machine learning. Finally, 2100 awake EEG data points and 2163 fatigue EEG data points are obtained. Significance correction is performed using paired t-tests and Bonferroni correction, and statistical p-values ​​and t-values ​​for the aforementioned nine EEG features are obtained. A smaller p-value indicates better differentiation between awake and fatigue for that type of EEG feature; that is, in the brain topography map corresponding to the statistical p-value, a larger blue area coverage is better, and a wider range of t-values ​​is better, indicating a broader effective brain region for that type of EEG feature.

[0193] The results obtained are as follows: Figure 8A Brain topography maps showing statistical p-values ​​corresponding to delta wave energy characteristics. Figure 8B Brain topography maps showing statistical t-values ​​corresponding to delta wave energy characteristics. Figure 9A Brain topography maps showing statistical p-values ​​corresponding to the theta wave energy characteristics are shown. Figure 9B Brain topography maps showing the statistical t-values ​​corresponding to the theta wave energy characteristics. Figure 10A Brain topography maps showing statistical p-values ​​corresponding to alpha wave energy characteristics are shown. Figure 10B Brain topography maps showing the statistical t-values ​​corresponding to the alpha wave energy characteristics. Figure 11A Brain topography maps showing the statistical p-values ​​corresponding to the peak energy characteristics of alpha waves. Figure 11B Brain topography maps showing the statistical t-values ​​corresponding to the peak energy characteristics of α waves. Figure 12A Brain topography maps showing statistical p-values ​​corresponding to β-wave energy characteristics. Figure 12B Brain topography maps showing the statistical t-values ​​corresponding to the β-wave energy characteristics. Figure 13A Brain topography maps showing statistical p-values ​​corresponding to aperiodic scaling transformation features. Figure 13B Brain topography maps showing statistical t-values ​​corresponding to non-periodic scaling transformation features. Figure 14A Brain topography maps showing statistical p-values ​​corresponding to non-periodic shift features. Figure 14B Brain topography maps showing statistical t-values ​​corresponding to non-periodic offset features. Figure 15A Brain topography maps showing statistical p-values ​​corresponding to the energy characteristics of the alpha wave band. Figure 15BBrain topography maps showing the statistical t-values ​​corresponding to the energy characteristics of the α-wave band. Figure 16A Brain topography maps showing the statistical p-values ​​corresponding to the peak characteristics of the alpha wave band. Figure 16B The brain topography map showing the statistical t-values ​​corresponding to the peak features of the alpha wave band is illustrated. For example, it can be seen that the dominant brain regions corresponding to the alpha wave band energy features and peak features have broad coverage, and the absolute value of the t-value can reach greater than 20, with a wider range of t-values.

[0194] Table 1: Classification accuracy statistics of 61 channels using 5 traditional features and 9 features of this invention

[0195]

[0196] To more objectively compare the effectiveness of the features, 34 machine learning models were used to process the feature data, as shown in Table 1. Table 1 shows that after using the features proposed in this invention, the classification accuracy improved from 90.07% to 90.25%, proving the effectiveness of the invention. In Table 1, the traditional five features refer to delta-wave energy features, theta-wave energy features, alpha-wave energy features, alpha-wave peak features, and beta-wave energy features. The nine features of this invention refer to delta-wave energy features, theta-wave energy features, alpha-wave energy features, alpha-wave peak features, beta-wave energy features, aperiodic offset features, aperiodic scaling transform features, alpha-wave band energy features, and alpha-wave band peak features.

[0197] Furthermore, the classification accuracy results demonstrate that the superior brain region proposed in this invention achieves higher classification accuracy compared to the traditional method of randomly selecting several brain regions. Specific experimental results are as follows: Figures 18-23 The diagram shows the dominant brain regions. Figures 18-23 More specifically, it shows the degree of difference between the same type of awake EEG characteristics and fatigue EEG characteristics in each region. The redder the color, the greater the difference, which means that the signal collected in that region is more sensitive to fatigue and can be regarded as a sensitive brain region.

[0198] in, Figure 18 The brain topography map of sensitive brain regions obtained using the chi-square test algorithm is shown. Specifically, it is drawn based on the number of times different signal acquisition locations appear in the ranking results obtained using the chi-square test algorithm. Figure 19 The brain topography map of the sensitive brain region is shown using the Minimum Redundancy Maximum Correlation (MRMR) algorithm, specifically drawn based on the number of times different signal acquisition locations appear in the ranking results obtained using the Minimum Redundancy Maximum Correlation (MRMR) algorithm. Figure 20The brain topography map of sensitive brain regions obtained using the Kruskal-Wallis test algorithm is shown. Specifically, it is drawn based on the frequency of occurrence of different signal acquisition locations in the ranking results obtained using the Kruskal-Wallis test algorithm. Figure 21 The brain topography map of the sensitive brain region is shown using the ReliefF algorithm, specifically drawn based on the frequency of occurrence of different signal acquisition locations in the sorting results obtained using the ReliefF algorithm. Figure 22 The brain topography map of sensitive brain regions is shown using the support vector machine-based recursive feature ranking algorithm (SVM-REF), specifically drawn based on the frequency of occurrence of different signal acquisition locations in the ranking results obtained using the support vector machine-based recursive feature ranking algorithm (SVM-REF). Figure 23 This illustration shows a brain topography map of sensitive brain regions obtained by combining five feature selection algorithms in an embodiment of the present invention. Specifically, it is drawn based on the total number of times different signal acquisition locations appear in the ranking results obtained by the aforementioned five feature selection algorithms; that is, it is drawn based on the union of the first five feature selection algorithms. Sensitive brain regions are indicated by a reddish hue. (Reference) Figures 18-23 The statistical map of dominant brain regions shown shows that F3, F4, O1, O2, and Oz are dominant brain regions (i.e., sensitive brain regions). In other words, the five signal acquisition locations of F3, F4, O1, O2, and Oz are finally identified as sensitive brain regions.

[0199] Referring to Table 2, the average accuracy of the five electrodes (F3, F4, O1, O2, Oz) in the dominant brain region is 4.2% higher than that of the five random electrodes (excluding F3, F4, O1, O2, Oz). This demonstrates the effectiveness of the sensitive brain regions selected by the method described above in fatigue detection.

[0200] Table 2: Classification accuracy statistics of 5 electrodes (F3, F4, O1, O2, Oz) and 5 random electrodes in dominant brain regions.

[0201]

[0202] Secondly, refer to Figure 4 The present invention discloses a device 1 for determining sensitive brain regions and EEG features for fatigue detection, wherein the sensitive brain regions and EEG features for fatigue detection are determined by the method for determining sensitive brain regions and EEG features for fatigue detection in any of the foregoing embodiments. The device 1 for determining sensitive brain regions and EEG features for fatigue detection includes a signal acquisition module 11, a feature calculation module 12, and a sensitive brain region and EEG feature determination module 13.

[0203] The signal acquisition module 11 is connected to the feature calculation module 12, and the scaffold image acquisition module 12 is connected to the sensitive brain region and EEG feature determination module 13.

[0204] Continue to refer to Figure 4 The signal acquisition module 11 is used to select multiple signal acquisition locations on the head and acquire the corresponding awake and fatigued EEG signals for each location. The awake EEG signal is the EEG signal in a conscious state, and the fatigued EEG signal is the EEG signal in a fatigued state.

[0205] The feature calculation module 12 is used to determine multiple awake EEG features corresponding to awake EEG signals and multiple fatigue EEG features corresponding to fatigue EEG signals based on the awake EEG signals and fatigue EEG signals collected at each signal acquisition location.

[0206] The sensitive brain region and EEG feature determination module 13 is used to determine the degree of difference between the same type of awake and fatigue EEG features at each signal acquisition location, and to sort the signal acquisition locations and corresponding EEG features according to the degree of difference, obtaining a sorting result. Based on the sorting result, the final signal acquisition location and final EEG feature are selected from the signal acquisition locations and EEG features located within a preset high degree of difference range as the sensitive brain region and the EEG feature used for fatigue judgment. The preset high degree of difference range is the range with relatively high degree of difference in the sorted data.

[0207] Using the above technical solution, the device 1 for determining sensitive brain regions and EEG features for fatigue detection of the present invention can compare the degree of difference between the same type of awake EEG features and fatigue EEG features collected from different signal acquisition locations, and can also compare the sensitivity of different types of EEG features to fatigue. Then, by sorting, it selects signal acquisition locations and EEG features located within a preset high degree of difference range, and then selects sensitive brain regions and EEG features for fatigue judgment. It can accurately select sensitive brain regions and EEG features suitable for fatigue detection, and can also make the number of selected sensitive brain regions relatively small and the selected EEG features relatively easy to calculate.

[0208] Furthermore, the signal acquisition module 11 can also be used to execute the aforementioned steps S11, S12, and related steps. The feature calculation module 12 can also be used to execute steps for determining the aperiodic shift features, aperiodic scaling transform features, alpha wave band energy features, and alpha wave band peak features of the EEG signal, such as steps S01, S02, S03, S04, S05, S06, and related steps. The sensitive brain region and EEG feature determination module 13 can also be used to execute the aforementioned steps S31, S32, S33, S34, S35, S36, S37, and related steps.

[0209] Thirdly, embodiments of the present invention disclose a fatigue detection method, which uses the sensitive brain regions and EEG characteristics determined by the method for determining sensitive brain regions and EEG characteristics for fatigue detection in any of the foregoing embodiments to perform fatigue detection.

[0210] By employing the above technical solution, the fatigue detection method of the present invention can achieve accurate fatigue detection using as few sensitive brain regions and EEG features as possible, while reducing computational load. More specifically, the above fatigue detection method can reduce the number of channels for acquiring EEG signals, simplify data acquisition methods, simplify the calculation of EEG features, reduce computational complexity, shorten computation time, and improve the convenience of fatigue detection.

[0211] Furthermore, the identified sensitive brain regions include the F3, F4, O1, O2, and Oz electrode locations of the 10-20 international standard lead system, and the identified EEG characteristics include delta wave energy characteristics, theta wave energy characteristics, alpha wave energy characteristics, alpha wave peak characteristics, beta wave energy characteristics, aperiodic shift characteristics, aperiodic stretching transformation characteristics, alpha wave band energy characteristics, and alpha wave band peak characteristics.

[0212] The process involves employing different numbers or types of feature selection algorithms to obtain ranking results, and then selecting sensitive brain regions and EEG features for fatigue detection. This may result in the selection of sensitive brain regions and EEG features different from the five electrode locations and nine EEG features mentioned above. Furthermore, for different individuals, datasets, and application scenarios, the method for determining sensitive brain regions and EEG features for fatigue detection in any of the aforementioned embodiments may also select other sensitive brain regions and EEG features for fatigue detection. In this invention, the sensitive brain regions and EEG features (i.e., the five electrode locations and nine EEG features mentioned above) determined by the method for determining sensitive brain regions and EEG features for fatigue detection in any of the aforementioned embodiments meet the needs of fatigue detection in most cases and have universality. In actual fatigue detection, the aforementioned determined sensitive brain regions and EEG features can be preferentially retained.

[0213] Fourthly, refer to Figure 5 The present invention also discloses an electronic device 2, including a memory 21 and a processor 22. The memory 21 stores at least one instruction, which, when executed by the processor 22, implements the method for determining sensitive brain regions and EEG characteristics for fatigue detection in any of the embodiments, and / or implements the fatigue detection method in any of the aforementioned embodiments. The memory 21 may, for example, include system memory, fixed non-volatile storage media, etc. The system memory may, for example, store an operating system, application programs, a boot loader, and other programs.

[0214] In this embodiment, the electronic device 2 can compare the differences between the same type of awake and fatigue EEG features acquired from different signal acquisition locations, and can also compare the sensitivity of different types of EEG features to fatigue. It then selects signal acquisition locations and EEG features within a preset high degree of difference range through sorting, and further selects sensitive brain regions and EEG features for fatigue judgment. This allows for accurate selection of suitable sensitive brain regions and EEG features for fatigue detection, while also minimizing the number of selected sensitive brain regions and making the selected EEG features easier to calculate. Furthermore, during fatigue detection, accurate fatigue detection can be achieved with as few sensitive brain regions and EEG features as possible, reducing computational load.

[0215] Fifthly, embodiments of the present invention disclose a computer-readable storage medium storing at least one instruction, which, when executed, implements the method for determining sensitive brain regions and EEG features for fatigue detection in any of the foregoing embodiments, and / or implements the fatigue detection method in any of the foregoing embodiments.

[0216] Using the above technical solution, the computer-readable storage medium can compare the differences between awake and fatigue EEG features acquired from different signal acquisition locations, and also compare the sensitivity of different types of EEG features to fatigue. Furthermore, by sorting and selecting signal acquisition locations and EEG features within a preset high-difference range, it can further select sensitive brain regions and EEG features for fatigue assessment. This allows for accurate selection of suitable sensitive brain regions and EEG features for fatigue detection, while minimizing the number of selected sensitive brain regions and making the selected EEG features easier to compute. Moreover, during fatigue detection, accurate fatigue detection can be achieved with as few sensitive brain regions and EEG features as possible, reducing computational load.

[0217] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0218] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0220] While the present invention has been illustrated and described with reference to certain preferred embodiments, those skilled in the art should understand that the above description is a further detailed explanation of the invention in conjunction with specific embodiments, and should not be construed as limiting the specific implementation of the invention to these descriptions. Various changes in form and detail can be made by those skilled in the art, including several simple deductions or substitutions, without departing from the spirit and scope of the invention.

Claims

1. A method for determining sensitive brain regions and electroencephalographic features for fatigue detection, characterized in that, The sensitive brain area is a region of the head where the brain electrical characteristics are collected, and the determination method comprises: selecting multiple signal collection positions on the head and collecting corresponding wake brain electrical signals and fatigue brain electrical signals at each signal collection position; wherein the wake brain electrical signals are brain electrical signals in a wake state, and the fatigue brain electrical signals are brain electrical signals in a fatigue state; determining multiple wake brain electrical characteristics corresponding to the wake brain electrical signals and multiple fatigue brain electrical characteristics corresponding to the fatigue brain electrical signals according to the wake brain electrical signals and the fatigue brain electrical signals collected at each signal collection position; determining the difference degree of the same type of wake brain electrical characteristics and fatigue brain electrical characteristics of each signal collection position, and sorting the signal collection positions and corresponding brain electrical characteristics according to the size of the difference degree to obtain a sorting result, and selecting a final signal collection position and a final brain electrical characteristic from the signal collection positions and brain electrical characteristics within a preset high difference degree range as a sensitive brain area and a brain electrical characteristic for fatigue judgment.

2. The method of claim 1, wherein the sensitive brain regions and electroencephalographic features for fatigue detection are determined by: The types of the wake brain electrical characteristics and the fatigue brain electrical characteristics include non-periodic offset characteristics and non-periodic stretching transformation characteristics, and the non-periodic offset characteristics and the non-periodic stretching transformation characteristics of the brain electrical signals are determined by the following methods: determining the logarithmic power spectrum of the brain electrical signals; nonlinearly fitting the logarithmic power spectrum according to the following formula 1 to obtain a corresponding fitting curve: L(f) = a-b* log(P(f)) (Formula 1) wherein L(f) is the fitting curve, P(f) is the logarithmic power spectrum, a is an offset parameter of the fitting curve, b is a stretching transformation parameter of the fitting curve, a and b are determined by least square fitting, and f is the frequency of the brain electrical signals, the range of f is 2-5 Hz and 14-30 Hz; determining the offset parameter a in the formula 1 as the non-periodic offset characteristic, and determining the stretching transformation parameter b in the formula 1 as the non-periodic stretching transformation characteristic.

3. The method of claim 2, wherein the sensitive brain regions and electroencephalographic features for fatigue detection are determined by: The types of the wake brain electrical characteristics and the fatigue brain electrical characteristics also include alpha wave frequency band energy characteristics and alpha wave frequency band peak value characteristics, and the alpha wave frequency band energy characteristics and the alpha wave frequency band peak value characteristics of the brain electrical signals are determined by the following methods: determining an alpha wave frequency spectrum periodic component according to the logarithmic power spectrum and the fitting curve by the following formula 2: P ′ (f) = P(f) - L(f) (Equation 2) where P ′ (f) is the periodic component of the alpha wave spectrum, L(f) is the fitted curve, P(f) is the log power spectrum, and f is the frequency of the electroencephalogram signal. determining an alpha wave frequency spectrum periodic energy component according to the alpha wave frequency spectrum periodic component by the following formula 3: P ″ (f) = 10 x 10 P′(f) (Equation 3) where P ′ (f) is the alpha wave spectral periodic component, P ″ (f) is the alpha wave spectral periodic energy component, f is the frequency of the electroencephalogram signal; determining the alpha wave frequency band energy characteristics and the alpha wave frequency band peak value characteristics according to the alpha wave frequency spectrum periodic energy component.

4. The method of claim 1, wherein, The collection of the corresponding wake brain electrical signals and fatigue brain electrical signals at each signal collection position in the multiple signal collection positions on the head specifically comprises: collecting corresponding original wake brain electrical signals and original fatigue brain electrical signals at each signal collection position in the multiple signal collection positions on the head; The original wake EEG signal and the original fatigue EEG signal are preprocessed to obtain the wake EEG signal and the fatigue EEG signal; wherein the preprocessing mode includes at least one of the following: data segmentation, deaverage, filtering, windowing.

5. The method of claim 1, wherein, The difference degree of the same type of wake EEG feature and fatigue EEG feature is determined by a feature selection algorithm, and the final signal acquisition position and the final EEG feature are selected from the signal acquisition positions and EEG features within the preset high difference degree range according to the sorting result, specifically including: Determine the number of occurrences of different signal acquisition positions and different EEG features within the preset high difference degree range; Select the signal acquisition position and the EEG feature with relatively more occurrences as the position and feature candidate data set corresponding to the feature selection algorithm, and select the final signal acquisition position and the final EEG feature from the position and feature candidate data set.

6. The method of claim 5, wherein the sensitive brain regions and electroencephalographic features for fatigue detection are determined by: The feature selection algorithm is multiple, and the final signal acquisition position and the final EEG feature are selected from the signal acquisition positions and EEG features within the preset high difference degree range according to the sorting result, further including: Obtain the position and feature candidate data set corresponding to each feature selection algorithm; In the union set of multiple position and feature candidate data sets, select the signal acquisition position and the EEG feature with relatively more occurrences as the final signal acquisition position and the final EEG feature.

7. The method of claim 6, wherein the sensitive brain regions and electroencephalographic features for fatigue detection are determined by: The multiple feature selection algorithms include machine learning-based feature selection algorithms and statistical-based feature selection algorithms.

8. The method of claim 7, wherein the sensitive brain regions and electroencephalographic features for fatigue detection are determined by: The multiple feature selection algorithms include minimum redundancy maximum relevance MRMR algorithm, chi-square test algorithm, ReliefF algorithm, Kruskal-Wallis test algorithm, and support vector machine-based recursive feature ranking algorithm SVM-REF.

9. The method of claim 1, wherein, The final signal acquisition position and the final EEG feature are selected from the signal acquisition positions and EEG features within the preset high difference degree range according to the sorting result, specifically including: Obtain a fatigue detection model, wherein the fatigue detection model detects fatigue through signal acquisition positions and EEG features; Select a pending signal acquisition position and a pending EEG feature from the preset high difference degree range, input the pending signal acquisition position and the pending EEG feature into the fatigue detection model and perform fatigue detection to determine the accuracy rate of the fatigue detection model for fatigue judgment; Change the number and / or type of the pending signal acquisition position and the pending EEG feature, and input the changed pending signal acquisition position and the pending EEG feature into the fatigue detection model to determine the accuracy rate of fatigue judgment, until the least number of the pending signal acquisition position and the pending EEG feature are obtained and the determined accuracy rate meets the preset high accuracy rate condition, and the pending signal acquisition position and the pending EEG feature corresponding to the last change are determined as the final signal acquisition position and the final EEG feature.

10. An apparatus for determining sensitive brain regions and electroencephalographic features for fatigue detection, characterized in that, The sensitive brain area is the area of the head where the EEG feature is collected, and the determination device includes: The signal collection module is configured to select multiple signal collection positions on the head and collect a clear brain electrical signal and a fatigue brain electrical signal corresponding to each signal collection position; the clear brain electrical signal is a brain electrical signal in a clear state, and the fatigue brain electrical signal is a brain electrical signal in a fatigue state; The feature calculation module is configured to determine multiple clear brain electrical features corresponding to the clear brain electrical signal and multiple fatigue brain electrical features corresponding to the fatigue brain electrical signal according to the clear brain electrical signal and the fatigue brain electrical signal collected at each signal collection position; The sensitive brain area and brain electrical feature determination module is configured to determine a difference degree of the clear brain electrical feature and the fatigue brain electrical feature of the same type at each signal collection position, sort the signal collection positions and the corresponding brain electrical features according to the size of the difference degree, obtain a sorting result, and select a final signal collection position and a final brain electrical feature as a sensitive brain area and a brain electrical feature used for fatigue judgment from the signal collection positions and the brain electrical features in a preset high difference degree range according to the sorting result.

11. A fatigue detection method characterized by, The fatigue detection method is used for fatigue detection by using the sensitive brain area and the brain electrical feature determined by the method for determining a sensitive brain area and a brain electrical feature for fatigue detection according to any one of claims 1-9.

12. The fatigue detection method of claim 11, wherein, The determined sensitive brain area includes F3, F4, O1, O2 and Oz electrode positions of the 10-20 international standard lead system, and the determined brain electrical feature includes a delta wave energy feature, a theta wave energy feature, an alpha wave energy feature, an alpha wave peak value feature, a beta wave energy feature, a non-periodic offset feature, a non-periodic stretching and transforming feature, an alpha wave frequency band energy feature and an alpha wave frequency band peak value feature.

13. An electronic device, comprising: The electronic device includes a processor and a memory, and the memory stores at least one instruction, and the at least one instruction is executed by the processor to implement the method for determining a sensitive brain area and a brain electrical feature for fatigue detection according to any one of claims 1-9, and / or implement the fatigue detection method according to any one of claims 11-12.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, and the at least one instruction is executed to implement the method for determining a sensitive brain area and a brain electrical feature for fatigue detection according to any one of claims 1-9, and / or implement the fatigue detection method according to any one of claims 11-12.

Images