Agricultural machinery driving fatigue grade division method based on multi-source physiological information

Abstract

The invention discloses an agricultural machinery driving fatigue grade division method based on multi-source physiological information, solves the problems of insufficient robustness and detection precision of the agricultural machinery driving fatigue grade division method, and belongs to the technical field of human-computer interaction, traffic safety and artificial intelligence cross application. The method comprises the steps that electrocardiosignals and face image data of a driver are collected, feature extraction is carried out, and a heart rate feature vector and a face feature vector of a current sliding window are obtained; fusing the standardized heart rate features and facial features to obtain fused feature representation; taking the fusion feature representation as an observation vector, inputting the observation vector into a time sequence model, and estimating potential fatigue strength; according to the standardized heart rate features and facial features, obtaining respective corresponding edge fatigue probabilities, and fusing the two edge fatigue probabilities into a joint fatigue confidence coefficient; and calibrating the potential fatigue strength by using the joint fatigue confidence, and carrying out ordered grading by using the calibrated fatigue strength to obtain a predicted fatigue grade.

Description

Technical Field

[0001] This invention relates to a method for classifying agricultural machinery driver fatigue levels based on multi-source physiological information, belonging to the technical field of human-computer interaction, traffic safety and artificial intelligence cross-application. Background Technology

[0002] With the rapid development of agricultural mechanization and intelligentization, fatigue driving is a common problem for agricultural machinery drivers during long hours and high-intensity operations. Traditional fatigue detection methods mainly include the following categories:

[0003] Single physiological signal monitoring methods, such as fatigue determination based on indicators like heart rate and EEG, are susceptible to noise and individual differences in practical applications, resulting in insufficient robustness.

[0004] Facial behavior recognition methods, such as blinking frequency, yawning, and head posture, are intuitive, but their recognition accuracy decreases in complex environments (such as changes in lighting or occlusion).

[0005] Methods based on vehicle operating parameters, such as steering wheel deflection frequency and operation response time, can only indirectly reflect the driver's fatigue level and lack direct physiological evidence.

[0006] Existing technologies generally suffer from the following shortcomings:

[0007] Limitations of single-modal features: relying solely on heart rate or facial features, lacking cross-modal fusion, resulting in insufficient fatigue detection.

[0008] Lack of hierarchical assessment: Most methods can only provide a binary result (fatigue / not fatigued), which cannot distinguish between mild, moderate and severe fatigue, making it difficult to meet the needs of precise intervention.

[0009] Insufficient model robustness: Detection performance significantly decreases in complex farmland environments such as noise interference and unstable lighting. Summary of the Invention

[0010] To address the issues of insufficient robustness and detection accuracy in methods for classifying fatigue levels of agricultural machinery drivers, this invention provides a method for classifying fatigue levels of agricultural machinery drivers based on multi-source physiological information.

[0011] The present invention provides a method for classifying agricultural machinery driver fatigue levels based on multi-source physiological information, comprising:

[0012] S1. Collect the driver's electrocardiogram signal and facial image data, and extract features to obtain the heart rate feature vector and facial feature vector of the current sliding window;

[0013] S2. Standardize each feature in the current sliding window's heart rate feature vector and facial feature vector;

[0014] S3. Fuse the standardized heart rate features and facial features to obtain the fused feature representation. ;

[0015] S4. Represent the fused features As an observation vector, it is input into the time series model to estimate the potential fatigue intensity. ;

[0016] S5. Based on the standardized heart rate features and facial features, obtain their respective edge fatigue probabilities, and fuse the edge fatigue probabilities corresponding to the heart rate features and facial features into a joint fatigue confidence score.

[0017] S6. Utilizing joint fatigue confidence to assess potential fatigue strength Perform calibration and utilize the calibrated fatigue strength. By performing an ordered classification, the predicted fatigue level can be obtained.

[0018] As a preferred option, the heart rate feature vector of the current sliding window is:

[0019]

[0020] in, This represents the mean interval between heartbeats. Standard deviation, The root mean square of the difference between adjacent heartbeats. The proportion of consecutive heartbeats with a difference of more than 50 ms. For low-frequency power, For high-frequency power, The ratio of sympathetic to parasympathetic tension. For sample entropy, It is the fractal dimension; The current sliding window fatigue tendency spectrum power ratio:

[0021]

[0022] This represents the power spectral density of the current sliding window heartbeat sequence. This is a low-activity frequency band associated with fatigue. Indicates the baseline frequency band. Represents the numerically stable term. Indicates frequency.

[0023] As a preferred embodiment, the facial feature vector is:

[0024]

[0025] in, To determine the percentage of eyelid closure within a set time period, The blinking frequency per unit time The average duration of a blink. The frequency of yawning per unit time. This refers to the characteristics of the mouth's opening and closing amplitude. , , These are the nose pitch angle, yaw angle, and roll angle, respectively. The expression energy vector is obtained from the facial feature vector of the current sliding window; the fatigue expression consistency index is:

[0026]

[0027] The mean of facial expressions in a conscious state. The mean of facial expression embedding under fatigue state. It is a norm 2. This represents a numerically stable term.

[0028] Preferably, in S2, the method for standardizing each feature in the heart rate feature vector and facial feature vector includes:

[0029] Perform restricted quantile scaling on each feature to obtain the standardized heart rate features. and facial features :

[0030]

[0031] superscript for or , To be Cut off at the 1% and 99% percentiles. The median. This represents the median absolute deviation.

[0032] Preferably, in S3, the standardized heart rate features and facial features are fused to obtain a fused feature representation. The methods include:

[0033] The standardized heart rate features and facial features are encoded separately to obtain the encoded data. and ;

[0034] Using standardized heart rate features as the query The standardized facial features are the key. Sum Calculate the attention weights of facial features on heart rate features. :

[0035]

[0036] in, Scaling the dimensions for attention;

[0037] Using standardized facial features as the query The standardized heart rate characteristics are key. Sum Calculate the attention weights of heart rate features on facial features. :

[0038]

[0039] Will and The fusion is performed to obtain the fused feature representation. :

[0040]

[0041] in, To achieve mutual attention fusion weights, For vector concatenation, To fuse linear transformations.

[0042] Preferably, in S4, the fused feature representation is... As an observation vector, it is input into the time series model to estimate the potential fatigue intensity. The methods include:

[0043] The nonlinear state-space model is as follows:

[0044]

[0045]

[0046] in, It is a nonlinear transfer function. For process noise, For the observation vector, For nonlinear observation mapping function, To observe noise;

[0047] Estimating network using variational inference to Approximate posterior distribution The variational inference network is trained by maximizing the lower bound of evidence.

[0048]

[0049] in, For the loss of the lower boundary of evidence, For the set of parameters of the variational inference network, for divergence, As expected, This represents the prior distribution.

[0050] Preferably, in S5, the edge fatigue probabilities corresponding to the standardized heart rate features and facial features are obtained respectively. The method of fusing the edge fatigue probabilities corresponding to the heart rate features and facial features respectively into a joint fatigue confidence score includes:

[0051] The standardized heart rate features and facial features are respectively passed through a fully connected network to obtain their respective edge fatigue probabilities. and ;

[0052] edge fatigue probability and The input is fed into the Copula function to obtain the joint fatigue confidence score. :

[0053]

[0054]

[0055]

[0056] in Empirical CDF for the marginal distribution of heart rate characteristics. Empirical CDF for facial feature distribution , All are random variables that follow a uniform distribution on the interval [0,1]. This refers to the Copula function.

[0057] Preferably, in S5, the calibrated fatigue strength is used. Methods for obtaining predicted fatigue levels by classifying fatigue levels into ordered levels include:

[0058] calibrated fatigue strength Mapped to a four-level ordered hierarchy These represent being alert, mildly fatigued, moderately fatigued, and severely fatigued, respectively.

[0059] Using preset threshold parameters , and Calculate the cumulative probability:

[0060]

[0061]

[0062]

[0063]

[0064] in, For the Sigmoid function, The slope from fatigue strength to logarithmic probability;

[0065] Based on the obtained cumulative probability, calculate the probability of each fatigue level:

[0066]

[0067]

[0068]

[0069]

[0070] The fatigue level with the highest probability is the predicted level. :

[0071] .

[0072] The beneficial effects of this invention are that by introducing heart rate variability (time domain, frequency domain, nonlinear indicators) and facial behavioral features (PERCLOS, blinking, yawning, head posture, expression energy), and combining cross-modal hierarchical attention mechanisms and Copula probability calibration, it achieves joint modeling of physiological and behavioral dual channels, which greatly improves detection accuracy and robustness. Attached Figure Description

[0073] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0076] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0077] The method for classifying agricultural machinery driver fatigue levels based on multi-source physiological information in this embodiment includes:

[0078] Step 1: Collect the driver's electrocardiogram signal and facial image data, and extract features to obtain the heart rate feature vector and facial feature vector of the current sliding window;

[0079] Assume the data acquisition system has a length of w seconds and a step size of w. A sliding window of seconds is used to process multi-source data. The observations in the t-th window are:

[0080]

[0081] For all observation data in the t-th time window, This is a heart rate feature vector related to HRV. These are facial feature vectors related to facial expressions and postures.

[0082] The heart rate feature vector of the current sliding window in this application is:

[0083]

[0084] in, This represents the mean interval between heartbeats. Standard deviation, The root mean square of the difference between adjacent heartbeats. The proportion of consecutive heartbeats with a difference of more than 50 ms. For low-frequency power, For high-frequency power, The ratio of sympathetic to parasympathetic tension. For sample entropy, It is the fractal dimension; The current sliding window fatigue tendency spectrum power ratio:

[0085]

[0086] This represents the power spectral density of the current sliding window heartbeat sequence. This is a low-activity frequency band associated with fatigue. Indicates the baseline frequency band. Represents the numerically stable term. Indicates frequency.

[0087] The facial feature vector of the current sliding window in this application is:

[0088]

[0089] in, To determine the percentage of eyelid closure within a set time period, The blinking frequency per unit time The average duration of a blink. The frequency of yawning per unit time. This refers to the characteristics of the mouth's opening and closing amplitude. , , These are the nose pitch angle, yaw angle, and roll angle, respectively. The expression energy vector is obtained from the facial feature vector of the current sliding window. This index can be obtained through expression AU (action unit) or deep learning network; the fatigue expression consistency index is:

[0090]

[0091] The mean of facial expressions in a conscious state. The mean of facial expression embedding under fatigue state. It is a norm 2. This represents a numerically stable term.

[0092] This application simultaneously acquires electrocardiogram signals and facial videos, reflecting fatigue status from two independent channels: physiological (autonomic nervous system) and behavioral (visual performance).

[0093] ECG features are extracted in the time domain (SDNN, RMSSD), frequency domain (LF, HF, LF / HF), and nonlinear indices (sample entropy, complexity) to characterize autonomic nervous system regulation and stress response.

[0094] Facial features such as PERCLOS (percentage of eyelid closure time), blink frequency, number of yawns, head posture (pitch / yaw angle), and facial energy (intensity of facial motion units) are extracted to capture visual fatigue cues.

[0095] Furthermore, based on the characteristic changes in the regulatory patterns of the autonomic nervous system (ANS) under fatigue conditions, these changes are reflected in the power distribution of HRV signals at specific frequency bands. This application proposes a fatigue tendency spectrum power ratio for real-time monitoring of dynamic changes in fatigue state; in practical driver fatigue monitoring systems, especially in scenarios with high vibration and complex environments such as agricultural machinery, it provides a stable, continuous, and specific physiological fatigue indicator.

[0096] Since fatigue leads to a decline in the ability to control facial expression muscles, resulting in a tendency for the change pattern of facial expression features to become more consistent within a continuous time window, specifically, when fatigued, a driver's facial expressions become rigid and less varied, or periodic fatigue movements (such as frequent blinking, yawning, etc.) will occur. These will all reflect consistency in the expression feature sequence. This application proposes a fatigue expression consistency index, which provides a sensitive, stable, and interpretable fatigue detection feature by quantifying the consistency of facial expression change patterns.

[0097] This application's dual-channel features provide complementary information: physiological indicators are objective and stable, while behavioral indicators are intuitive and sensitive, jointly covering the internal and external manifestations of fatigue. This lays a data foundation for subsequent multimodal fusion and improves the comprehensiveness and robustness of the features.

[0098] Step 2: Standardize each feature in the current sliding window's heart rate feature vector and facial feature vector;

[0099] This step involves independently standardizing each feature to eliminate differences in the units and ranges of different features, thus preventing large-scale features from dominating model learning. Specifically, each feature is subjected to restricted quantile scaling to obtain the standardized heart rate feature. and facial features :

[0100]

[0101] superscript for or , To be Cut off at the 1% and 99% percentiles. The median. This represents the median absolute deviation.

[0102] This invention brings different modal features to the same order of magnitude, which facilitates cross-modal comparison and fusion, improves model convergence speed and numerical stability, and enhances generalization ability.

[0103] Step 3: Fuse the standardized heart rate features and facial features to obtain the fused feature representation. ;

[0104] This step employs intramodal self-attention: self-attention calculations are performed separately for heart rate and facial features to extract key information within each modality. Cross-modal cross-attention: heart rate features are used as queries, and facial features as keys (or vice versa), enabling bidirectional information exchange. The attention-weighted bimodal features are then concatenated to form a fused representation.

[0105] Specifically, the standardized heart rate features and facial features are encoded separately to obtain the encoded data. and ;

[0106] Using standardized heart rate features as the query The standardized facial features are the key. Sum Calculate the attention weights of facial features on heart rate features. :

[0107]

[0108] in, Scaling the dimensions for attention;

[0109] Using standardized facial features as the query The standardized heart rate characteristics are key. Sum Calculate the attention weights of heart rate features on facial features. :

[0110]

[0111] Will and The fusion is performed to obtain the fused feature representation. :

[0112]

[0113] in, To achieve mutual attention fusion weights, For vector concatenation, To fuse linear transformations.

[0114] This step dynamically weights and highlights fatigue-related features, suppresses noise and irrelevant information, achieves deep interaction between physiological and behavioral features, enhances the discriminative power of feature representation, and provides high-quality input for time series modeling.

[0115] Step 4: Represent the fused features As an observation vector, it is input into the time series model to estimate the potential fatigue intensity. ;

[0116] Representing fusion features As the observation vector, it is input to a nonlinear state-space model (such as a variational autoencoder based on RNN or Transformer). Potential fatigue strength. For continuous latent variables, their posterior distribution is approximated through variational inference. A stability regularization term (such as a latent state smoothing constraint) is introduced to suppress short-term fluctuations and capture the gradual evolution trend of fatigue. Specifically, the nonlinear state-space model is as follows:

[0117]

[0118]

[0119] in, It is a nonlinear transfer function. For process noise, For the observation vector, For nonlinear observation mapping function, To observe noise;

[0120] Estimating network using variational inference to Approximate posterior distribution The variational inference network is trained by maximizing the lower bound of evidence.

[0121]

[0122] in, For the loss of the lower boundary of evidence, For the set of parameters of the variational inference network, for divergence, As expected, This represents the prior distribution.

[0123] This step models the temporal dependence of fatigue states, avoiding the jitter problem of independent frame determination. The estimated potential intensity provides a continuous and smooth fatigue metric, laying the foundation for subsequent ordered classification.

[0124] Step 5: Based on the standardized heart rate features and facial features, obtain their respective edge fatigue probabilities, and fuse the edge fatigue probabilities corresponding to the heart rate features and facial features into a joint fatigue confidence score.

[0125] This step inputs the standardized heart rate features and facial features into a classifier (such as logistic regression) to obtain the edge fatigue probability. A Copula function (such as Gaussian Copula) is then used to model the dependency structure of the two modal probabilities, which are then fused into a joint fatigue confidence score. Specifically, the standardized heart rate features and facial features are passed through a fully connected network to obtain their respective edge fatigue probabilities. and ;

[0126] edge fatigue probability and The input is fed into the Copula function to obtain the joint fatigue confidence score. :

[0127]

[0128]

[0129]

[0130] in Empirical CDF for the marginal distribution of heart rate characteristics. Empirical CDF for facial feature distribution , All are random variables that follow a uniform distribution on the interval [0,1]. This refers to the Copula function.

[0131] This step utilizes Copula to characterize the correlation between modes, improving the reliability of probability estimation. The resulting joint confidence score integrates dual-channel information, providing an uncertainty measure for subsequent calibration and enhancing interpretability.

[0132] Step 6: Utilize joint fatigue confidence to assess potential fatigue strength. Perform calibration and utilize the calibrated fatigue strength. By performing an ordered classification, the predicted fatigue level can be obtained.

[0133] Potential fatigue strength This is a continuous value estimated from a time-series model, representing the driver's intrinsic fatigue level. However, this estimate is subject to uncertainty, especially in cases of poor observation data quality or conflicting multimodal information. The joint fatigue confidence score reflects the reliability of the current estimate and can therefore be used for calibration. Specifically, direct weighted calibration can be used, adjusted according to the joint fatigue confidence level. The weights are adjusted so that the weights shrink towards the baseline when the confidence level is low. Alternatively, a threshold adjustment calibration can be used; a threshold is set, and the result is determined. Is it reliable?

[0134] Using calibrated fatigue strength The specific steps of orderly hierarchical classification include:

[0135] calibrated fatigue strength Mapped to a four-level ordered hierarchy These represent being alert, mildly fatigued, moderately fatigued, and severely fatigued, respectively.

[0136] Using preset threshold parameters , and Calculate the cumulative probability:

[0137]

[0138]

[0139]

[0140]

[0141] in, For the Sigmoid function, The slope from fatigue strength to logarithmic probability;

[0142] Based on the obtained cumulative probability, calculate the probability of each fatigue level:

[0143]

[0144]

[0145]

[0146]

[0147] The fatigue level with the highest probability is the predicted level. :

[0148] .

[0149] The ordered regression output in this step produces a four-level probability distribution, enabling fine-grained fatigue grading and supporting differentiated intervention strategies. The entire process is probabilistically interpretable and can trigger multi-level feedback through risk scores (fusion level, confidence level, fatigue momentum).

[0150] This application achieves multimodal information fusion: by introducing heart rate variability (time domain, frequency domain, nonlinear indicators) and facial behavioral features (PERCLOS, blinking, yawning, head posture, expression energy), and combining cross-modal hierarchical attention mechanism and Copula probability calibration, it realizes joint modeling of physiological and behavioral dual channels, which greatly improves detection accuracy and robustness.

[0151] This application employs a tiered assessment method: based on potential fatigue intensity and an ordered logistic regression model, driver fatigue is categorized into four levels: alert, mild, moderate, and severe. This breaks through the binary classification limitations of traditional methods, providing a basis for differentiated intervention. This application utilizes temporal dynamic modeling: employing a nonlinear state-space model and variational inference methods, it can capture the temporal evolution trend of fatigue states and suppress short-term fluctuations through stability regularization, achieving smooth and reliable assessment. This application features interpretability and a feedback mechanism: by fusing risk scores with level, confidence, and fatigue momentum, multi-level intervention thresholds are set, and the results can be projected onto key features such as PERCLOS and LF / HF, providing drivers with intuitive explanations and real-time feedback. This application demonstrates applicability and safety: the method is robust to complex environments and can be deployed at low cost in actual agricultural machinery operation scenarios, effectively reducing the risk of accidents caused by fatigued driving and improving the safety and efficiency of agricultural production.

[0152] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for classifying the fatigue level of agricultural machinery driving based on multi-source physiological information, characterized in that, Comprise: S1, collect the electrocardiosignal and facial image data of the driver, and perform feature extraction to obtain a heart rate feature vector and a facial feature vector of a current sliding window; S2, standardize each feature in the heart rate feature vector and the facial feature vector of the current sliding window; S3, fuse the normalized heart rate features and facial features to obtain a fused feature representation ; S4, fusing the feature representations As an observation vector, input to the time series model, the latent fatigue intensity is estimated ; S5, obtain respective corresponding edge fatigue probabilities according to the standardized heart rate features and facial features, and fuse the respective corresponding edge fatigue probabilities of the heart rate features and the facial features into a joint fatigue confidence; S6. Utilizing the joint fatigue confidence to predict potential fatigue strength Perform calibration, utilize the calibrated fatigue strength Perform ordered ranking, obtain the predicted fatigue ranking.

2. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, The heart rate feature vector of the current sliding window is: wherein, is the mean of the interbeat intervals, is the standard deviation, is the root mean square of the difference between adjacent heartbeats, is the proportion of adjacent heartbeats that differ by more than 50 ms, is the low frequency power, is the high frequency power, is the sympathetic to parasympathetic tone ratio, is the sample entropy, is the fractal dimension; is the current sliding window fatigue tendency spectral power ratio: the power spectral density of the current sliding window heartbeat sequence, a low activity band related to fatigue, denotes the baseline band, denotes a numerical stabilization term, denotes the frequency.

3. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, The facial feature vector is: wherein, is the proportion of eyelid closure in a set time, is the blink frequency per unit time, is the average blink duration, is the yawn frequency per unit time, is the mouth opening and closing amplitude feature, , , are the head pitch angle, yaw angle, roll angle, respectively; is the expression energy vector obtained from the face feature vector of the current sliding window; the fatigue expression consistency index is: is the mean expression embedding for the awake state, is the mean expression embedding for the fatigue state, is the two-norm, denotes the numerical stability term.

4. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, In S2, the method for standardizing each feature in the heart rate feature vector and the facial feature vector comprises: Restricted quantile scaling is done on each feature to get the normalized heart rate features and facial features : superscript for or , To be Cut off at the 1% and 99% percentiles. The median. This represents the median absolute deviation.

5. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, In S3, the normalized heart rate features and the facial features are fused to obtain a fused feature representation The method comprises: The standardized heart rate features and facial features are respectively coded to obtain coded and ; with the normalized heart rate features as queries , the normalized face features as keys , and values , compute attention weights of face features to heart rate features : wherein, is an attention scaling dimension; with the standardized facial features as queries , the standardized heart rate features as keys and values , and compute attention weights of the heart rate features to the facial features : Fusing and to obtain a fused feature representation : wherein, is an attention fusion weight, is a vector concatenation, is a fusion linear transformation.

6. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 5, characterized in that, In S4, the fusion feature representation As an observation vector, input to the time series model, the latent fatigue intensity is estimated The method comprises: The nonlinear state space model is: wherein, is a non-linear transfer function, is process noise, is an observation vector, is a non-linear observation mapping function, is observation noise; Estimating with variational inference networks To Approximate posterior distributions A variational inference network is trained by maximizing the lower bound on the evidence: wherein, is the evidence lower bound loss, is the set of variational inference network parameters, is the divergence, is the expectation, denotes the prior distribution.

7. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, In S5, the method for obtaining respective corresponding edge fatigue probabilities according to the standardized heart rate features and facial features, and fusing the respective corresponding edge fatigue probabilities of the heart rate features and the facial features into a joint fatigue confidence comprises: The standardized heart rate features and facial features are respectively input into a full connection network to obtain respective corresponding edge fatigue probabilities and ; The edge fatigue probability and is input into the Copula function to obtain the joint fatigue confidence : wherein is the empirical CDF of the heart rate feature edge distribution, is the empirical CDF of the face feature distribution, , are both random variables that follow a uniform distribution on the interval [0,1], is a Copula function.

8. The method for classifying the driving fatigue level of the agricultural machine based on the multi-source physiological information according to claim 1, characterized in that, In S5, the calibrated fatigue strength The method for performing an ordered ranking to obtain a predicted fatigue class comprises: calibrated fatigue strength mapped to a four-level ordinal scale representing, respectively, sober, mild fatigue, moderate fatigue, and severe fatigue; Using a preset threshold parameter , and , the cumulative probability is calculated: wherein, is a Sigmoid function, is the slope of the fatigue strength to logit. According to the obtained cumulative probability, the probability of each fatigue level is calculated: The most probable level of fatigue rating is the predicted level : 。 9.A device for classifying driving fatigue level of agricultural machinery based on multi-source physiological information, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method for classifying the fatigue levels of agricultural machinery driving based on multiple source physiological information according to any one of claims 1 to 8.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method for classifying the fatigue levels of agricultural machinery driving based on multiple source physiological information according to any one of claims 1 to 7.

Images