A system and method for evaluating fatigue state of a drone operator

Abstract

This invention discloses a fatigue state evaluation system and method for unmanned aerial vehicle (UAV) operators. It employs a strategy combining functional decomposition and task orientation to construct a hierarchical evaluation system encompassing four dimensions: eye-tracking perception, cognitive integration and attention allocation, task-related eye-tracking performance, and eye-tracking-fatigue iterative learning and adaptation. This system is quantitatively represented through 36 underlying indicators. A task-situation gating weighting model based on a normal cloud model and Shapley value game is constructed. Expert verbal ratings are converted into cloud drops to retain fuzziness and randomness. Shapley values ​​are used to eliminate redundant indicator coupling, and subjective and objective weights are dynamically integrated through task-situation gating factors. This invention can accurately quantify operator fatigue state in conjunction with task context, solving the problem that traditional static evaluation is difficult to adapt to dynamic environments and individual differences. It improves the robustness and accuracy of UAV inspection personnel status assessment and has strong applicability.

Description

Technical Field

[0001] This invention belongs to the field of human-computer interaction and aviation safety technology, and in particular relates to a fatigue state evaluation system and method for unmanned aerial vehicle (UAV) operators. Background Technology

[0002] With the rapid development of drone technology, drones have been widely used in bridge inspection. The scale of bridge construction is constantly expanding, especially with long-span bridges, which have large spans and long distances, requiring long inspection times and frequent trips. Operators need to maintain a high level of concentration for extended periods, searching for minute cracks or anomalies through monitor screens and controlling the drone to maintain stable flight in real time. This high-intensity, complex task combining visual search, cognitive decision-making, and manual control easily leads to operator fatigue. Once operators become fatigued, not only will critical defects be missed, but slow reactions or operational errors may also cause serious safety accidents such as drone collisions. While some solutions exist for detecting operator fatigue, their applicability to the precision inspection scenarios using drones remains insufficient.

[0003] In the prior art, patent CN202110995514.5 discloses a fatigue detection method based on physiological signals, which assesses fatigue by collecting electrocardiogram, pulse wave, and respiratory signals; patent CN202510459564.X proposes to utilize electroencephalogram (EEG) and electrooculogram (EOG) signals and solve the problem of individual differences through domain adaptive technology. Although such technologies can reflect the physiological arousal level of the human body, they face two major challenges in the practical application of drones for outdoor inspection: First, drone operators frequently make micro-movements during operations, and the outdoor electromagnetic environment is complex, making extremely weak EEG or ECG signals easily drowned out by artifacts generated by muscle movements or environmental noise, resulting in a low signal-to-noise ratio; second, simple physiological indicators can only determine whether the operator is drowsy, but cannot be directly used to reflect the core issue of whether their control ability has declined. Existing technologies struggle to establish a quantitative mapping between physiological signals and drone inspection task performance. Although patent CN202110295547.9 collected eye movement and hand movement data simultaneously, it processed the two separately and did not conduct an in-depth analysis of the temporal coordination relationship between eye movement information and hand movements.

[0004] Furthermore, current evaluation systems mostly employ static linear weighting methods for indicator weighting and model construction. These methods have significant limitations: ① They cannot objectively describe the inherent ambiguity and randomness in expert cognition, leading to distorted subjective evaluation results; ② There are often complex coupling and redundant relationships between various evaluation indicators, and traditional independent weighting methods are insufficient to eliminate biases caused by information overlap; ③ The environment of UAV inspection missions is dynamically changing, and static fixed weighting systems cannot adaptively adjust according to the mission situation, resulting in poor robustness of the model under different mission scenarios.

[0005] Therefore, there is an urgent need for a comprehensive evaluation system and method that can deeply integrate the drone inspection mission context, construct a hierarchical fatigue state evaluation system from multiple dimensions such as visual perception quality, cognitive decision-making efficiency, vision-hand coordination performance, and long-term adaptability, and overcome the ambiguity of expert cognition, eliminate indicator redundancy, and adapt to dynamic mission situations. Summary of the Invention

[0006] 1. The technical problem to be solved:

[0007] Existing technologies suffer from several shortcomings, including poor resistance to interference from physiological signals, a disconnect between evaluation indicators and tasks, static and rigid weighting methods in traditional evaluation models, difficulty in handling expert cognitive ambiguity, and redundant coupling between indicators.

[0008] 2. Technical Solution:

[0009] To address the above issues, this invention provides a fatigue state evaluation system for drone operators, comprising four primary indicators: eye-tracking detection and perception ability, cognitive integration and attention allocation ability, task execution-related eye-tracking performance ability, and indicators representing eye-tracking-fatigue iterative learning and adaptation ability.

[0010] The secondary feature indicators corresponding to eye-tracking detection and perception capabilities include: eye-tracking perception accuracy, eye-tracking perception time, eye-tracking perception breadth, and eye-tracking strategy abundance.

[0011] Secondary characteristic indicators of cognitive integration and attention allocation ability include: accuracy of alarm and target understanding, decision-making and attention switching time, and information processing and search complexity;

[0012] Secondary indicators of task-related eye-tracking performance include: operational coordination efficiency, operational accuracy, and task self-monitoring and error correction capabilities.

[0013] The secondary indicators characterizing eye-fatigue iterative learning and adaptation capabilities include: eye-movement efficiency improvement capability, perception accuracy improvement capability, and strategy diversity and self-adjustment capability.

[0014] Each secondary indicator has a tertiary indicator, which is the underlying indicator.

[0015] This invention provides a method for evaluating the fatigue status of drone operators, characterized by the following steps:

[0016] Step 1: Construct the aforementioned drone operator fatigue evaluation system;

[0017] Step 2: Design detailed evaluation indicators to represent the underlying level;

[0018] Step 3: Calculate the weighted combination of task situation gating based on the normal cloud model and Shapley value game;

[0019] Step 4: Obtain the final weights of each indicator to get the fatigue status score.

[0020] 3. Beneficial effects:

[0021] This invention constructs a hierarchical index system encompassing four dimensions: eye-tracking detection and perception, cognitive integration and attention allocation, task execution performance, and eye-tracking-fatigue iterative learning. It utilizes a normal cloud model to quantify the ambiguity and randomness of expert subjective cognition, introduces Shapley value game theory to eliminate redundant coupling of objective data, and finally achieves dynamic fusion of weights through a task situation gating mechanism. This enables accurate quantification, scientific assessment, and adaptive grading of the fatigue state of UAV operators in complex dynamic inspection scenarios. Attached Figure Description

[0022] Figure 1 This is a flowchart of the overall solution.

[0023] Figure 2 A process for constructing a fatigue assessment system for drone operators.

[0024] Figure 3 To construct a fatigue evaluation system for drone operators.

[0025] Figure 4 This is a schematic diagram showing how the changes in weighted amounts vary with the number of iterations.

[0026] Figure 5 A schematic diagram showing how the change in weights varies with the number of iterations when AHP is assigned weights as initial values.

[0027] Figure 6 This is a process illustration according to an embodiment of the present invention. Figure 3 .

[0028] Figure 7 This is a process illustration according to an embodiment of the present invention. Figure 4 . Detailed Implementation

[0029] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] A fatigue assessment system for drone operators, such as Figure 3 As shown, it includes four primary indicators: eye movement detection and perception ability, cognitive integration and attention allocation ability, task execution-related eye movement performance ability, and indicators representing eye movement-fatigue iterative learning and adaptation ability.

[0031] The secondary feature indicators corresponding to eye-tracking detection and perception capabilities include: eye-tracking perception accuracy, eye-tracking perception time, eye-tracking perception breadth, and eye-tracking strategy abundance.

[0032] The eye-tracking accuracy is used to characterize whether the operator's gaze is focused on the task-related area of ​​interest (AOI) and whether the fixation on key targets is adequate. Increased fatigue typically leads to gaze drift and an increased probability of missed targets. This metric is measured by gaze hit rate. gaze position error Key scenario coverage Key target fixation accuracy Four underlying metrics are used for characterization.

[0033] The gaze hit rate The percentage of time that the eye falls on the key area of ​​the task reflects whether visual attention is focused on key information. The specific calculation formula is as follows:

[0034] (1)

[0035] in, This indicates the number of gaze points that fall within the task-related area of ​​interest (AOI). This indicates the total number of fixations.

[0036] gaze position error This is used to represent the mean squared error between the fixation point and the reference position, characterizing fixation accuracy and stability. A larger value indicates a more severe deviation of the fixation from key elements of the task. The specific calculation formula is as follows:

[0037] (2)

[0038] in, Indicates the number of fixations sampled; Indicates the first The coordinates of the gaze point; For the first The corresponding reference coordinates of each gaze point.

[0039] The coverage of key scenarios This refers to the percentage of key scenes that operators successfully observe within a specified time, reflecting the overall completeness of monitoring. The specific calculation formula is as follows:

[0040] (3)

[0041] in, This indicates the number of key scenes that were scanned at least once within a specified time. This indicates the total number of critical scenarios that the task requires to be monitored.

[0042] The accuracy of fixation on the key target The calculation formula is as follows:

[0043] (4)

[0044] in, Indicates the number of key targets that have been correctly locked in by line of sight; This indicates the total number of key targets that should be locked.

[0045] The eye-tracking perception time reflects the speed at which an operator moves from seeing to focusing on key task information and the stability of sustained fixation; a higher value generally indicates a higher level of fatigue. This indicator is measured by the average first fixation time. and average fixation duration Two underlying metrics are used for characterization.

[0046] The average first fixation time The average time delay from stimulus appearance to first fixation on the key area; the average first fixation time is increased when operator fatigue worsens. It will increase accordingly, and the calculation formula is as follows:

[0047] (5)

[0048] in, Indicates the number of key AOIs; Indicates the first The time from presentation to first gaze of a key AOI.

[0049] The average fixation duration This represents the average duration of a single fixation. As operator fatigue increases, it can range from extremely short (attention cannot be maintained) to abnormally long (staring blankly). The calculation formula is as follows:

[0050] (6)

[0051] in, Indicates the first The fixation duration of a key AOI.

[0052] The eye-tracking span reflects the operator's spatial coverage of the task interface or external scene. When the operator is fatigued, it often manifests as "field of vision contraction" or "focusing on only a few areas," at which point the eye-tracking span is small. This indicator is measured by the maximum fixation range. and task AOI coverage Two underlying metrics are used for characterization.

[0053] The maximum gaze span This refers to the maximum spatial scanning range after weighting the overall task complexity, characterizing the field of view width of visual surveillance. The calculation formula is as follows:

[0054] (7)

[0055] in, Indicates the first The spatial span of the viewer's gaze within a screen or scene over a short period of time; Indicates the first The task complexity coefficient over a short period of time.

[0056] The task AOI coverage This refers to the proportion of key areas covered by the line of sight within a mission cycle, reflecting the breadth of surveillance. The calculation formula is as follows:

[0057] (8)

[0058] in, This indicates the number of AOIs that have been viewed; This represents the total number of AOIs defined in the task.

[0059] The abundance of eye-tracking strategies reflects the diversity of an operator's visual search patterns and whether there are stereotyped or singular strategies. Under high fatigue, behaviors such as "scanning only a fixed route" or "fixating on a single area" often occur, at which point the abundance of eye-tracking strategies is low. This indicator is measured by the number of identifiable eye-tracking strategy categories. The characterization is performed, and its calculation formula is as follows:

[0060] (9)

[0061] in, This represents the number of different eye-tracking strategy categories obtained by clustering saccadic-fixation sequences in the context of the current task.

[0062] The cognitive integration and attention allocation capabilities of a drone operator reflect their ability to understand and filter multi-source visual information and flexibly switch between different task sub-interfaces in complex bridge inspection tasks. When fatigue accumulates, it often manifests as characteristics such as alarm or target comprehension deviations, prolonged decision-making reaction time, and sluggish attention switching. These cognitive integration and attention allocation capabilities are characterized by three secondary feature indicators: accuracy of alarm and target comprehension, decision-making and attention switching time, and information processing and search complexity.

[0063] The alarm and target understanding accuracy is used to characterize the operator's ability to correctly perceive and interpret key information at the visual level, including the accuracy of their understanding of alarm information, task objectives, and status changes. This metric is measured by alarm recognition accuracy. Accuracy of task event recognition and false alarm / false negative rate Three underlying indicators characterize it.

[0064] The alarm recognition accuracy The formula used to reflect the accuracy of the operator's visual understanding and response to alarm information is as follows:

[0065] (10)

[0066] in, This indicates the number of alarms for which the line of sight falls within the alarm AOI before a response is given and the response is correct. This indicates the total number of alarms.

[0067] The accuracy of task event recognition The formula for calculating the degree of accuracy in understanding critical mission events (such as defect labeling and track deviation) is as follows:

[0068] (11)

[0069] in, This indicates that the line of sight covers the AOI (Area of ​​Interest) of the task and the action is handled correctly. This indicates the total number of task events.

[0070] The false alarm / false negative rate This refers to the overall proportion of false alarms and missed alarms generated by operators during alarm handling, reflecting their ability to reliably identify alarm information under fatigue conditions. The specific calculation formula is as follows:

[0071] (12)

[0072] in, This represents the number of false alarms. This represents the number of missed reports; This represents the total number of alarms. Increased fatigue usually leads to... It has increased significantly.

[0073] The decision-making and attention switching time is used to characterize the timeliness of an operator's process from "seeing key information" to "completing the cognitive-decision process." Fatigue leads to slower reaction times and delayed attention switching. This indicator is measured by decision reaction time. Note the switching time. and the cost of multitasking Three underlying indicators characterize it.

[0074] The decision response time This refers to the average latency between visual integration and behavioral decision-making. A higher value indicates slower cognitive processing. The calculation formula is as follows:

[0075] (13)

[0076] in, Indicates the number of decisions; Indicates the first In this decision-making process, the time from the first look at the relevant AOI to the issuance of a control command.

[0077] The attention switching time The efficiency of redistributing attention across different task sub-interfaces is calculated using the following formula:

[0078] (14)

[0079] in, Indicates the number of times a high-priority event occurs; Indicates the first The time from the occurrence of a high-priority event to the first time the line of sight falls into the corresponding AOI.

[0080] The cost of multitasking This is used to quantify the time cost for an operator to switch from one task interface to another in a multi-task scenario, relative to a single-task baseline, reflecting the impact of fatigue on multi-task attention allocation ability. The specific calculation formula is as follows:

[0081] (15)

[0082] in, To allow for multiple task switching times, For the first The actual time consumed by each multi-task switching This represents the average switching time under the same type of single-task conditions. The larger the value, the higher the overhead of multitasking under fatigue.

[0083] The information processing and search complexity reflects the eye-tracking search workload and hesitation required by an operator to make a decision or handle a task event. This metric is measured by the number of AOI scans per unit time. complexity of line-of-sight trajectory and search strategy stability Three underlying indicators characterize it.

[0084] The number of AOI scans per unit time This is used to characterize how many task-related areas an operator can monitor per unit of time, reflecting the workload and efficiency of visual search and monitoring. The specific calculation formula is as follows:

[0085] (16)

[0086] in, The number of AOIs that are visible within a task cycle; This corresponds to the duration of the task.

[0087] The complexity of the gaze trajectory The formula used to characterize whether the scanning path of the line of sight on the task interface is concise and efficient, or contains a large number of invalid back-and-forths and detours, is as follows:

[0088] (17)

[0089] in, This is the sum of the lengths of all scan vectors during the task; The maximum gaze span is calculated using formula (7). The larger the value, the more "circuitous" the line-of-sight trajectory, and the lower the search efficiency.

[0090] The stability of the search strategy The formula used to characterize the stability and orderliness of the operator's eye-tracking search strategy is as follows:

[0091] (18)

[0092] in, Shannon entropy for different eye-tracking strategy categories; This represents the total number of identified strategy categories. Frequent strategy switching or disordered wandering under fatigue conditions can lead to... Increase, thus Decrease.

[0093] Task-related eye-tracking performance primarily reflects the coordination and precision between visual and operational behaviors of a UAV operator when executing control commands during bridge inspection missions, as well as their ability to monitor and correct task status. Increased fatigue often manifests as misalignment of eye-hand coordination timing, insufficient visual focus before operation, and delayed detection of abnormal states. This task-related eye-tracking performance is characterized by three secondary feature indicators: visual-operation coordination efficiency, visual-operation accuracy, and task monitoring and error correction capabilities.

[0094] The eye-operation coordination efficiency describes whether the operator's gaze can lock onto the relevant interface or control in advance when performing operations such as joystick movement, mode switching, and parameter adjustment, and whether this "eye-first, hand-later" mode is stable. This indicator is measured by the eye-hand lead time. Eye-hand synchronization rate and continuous operation segment view-hand consistency Three underlying indicators characterize it.

[0095] Eye-hand lead time This value reflects whether the line of sight arrives at the target area before the operation. A larger value indicates more complete visual-operation coordination. The calculation formula is as follows:

[0096] (19)

[0097] in, This indicates the number of operations performed. Indicates the first In this operation, the time difference from when the line of sight first falls into the target control area to when the actual control action occurs (positive value when the line of sight is in front).

[0098] Eye-hand synchronization rate This value reflects whether the operational behavior is based on sufficient visual confirmation. The higher the value, the better the coordination. The calculation formula is as follows:

[0099] (20)

[0100] in, This indicates the number of operations in which the line of sight has fallen on the relevant AOI within a preset time window (e.g., within 500 ms before the action occurs); This indicates the total time the target area was in a "needs monitoring" state.

[0101] Continuous operation segment view – hand consistency The formula used to measure whether the "eye-first, hand-later" pattern remains stable during prolonged continuous operation is as follows:

[0102] (twenty one)

[0103] in, For most operations in a continuous operation segment to satisfy The number of segments (i.e., the number of visual inputs that arrive at the target area before hand movements); This represents the total number of consecutive operation segments. A higher value indicates more stable long-term vision-hand coordination.

[0104] The visual-operational precision reflects whether the operator's line of sight is sufficiently focused on the target area before performing a single fine-tuning operation (such as defect marking, key point selection, or track fine-tuning), and whether the line of sight has significantly deviated before a misoperation occurs. This indicator is measured by the operator's gaze focus before operation. Deviance of line of sight before misoperation Confirm gaze ratio with micro-operations Three underlying indicators characterize it.

[0105] Pre-operation gaze focus This value describes whether the line of sight is focused on the key area before performing delicate operations. The larger the value, the more focused the eye is. The calculation formula is as follows:

[0106] (twenty two)

[0107] in, This indicates the number of operations performed. Indicates the first The number of times the gaze falls on the target AOI within the time window before the operation; Indicates the first Total number of fixations within the time window prior to the operation.

[0108] The deviation of the line of sight before the misoperation This measure is used to assess the spatial deviation of the line of sight relative to the target area before a misoperation occurs. A larger value indicates less stable visual positioning. The calculation formula is as follows:

[0109] (twenty three)

[0110] in, This indicates the number of fixations detected before the erroneous operation. For the first Screen coordinates of a gaze point, These are the reference coordinates for the corresponding target area.

[0111] Micro-operation confirms gaze ratio The formula used to characterize whether the operator has made sufficient visual checks before performing delicate operations is as follows:

[0112] (twenty four)

[0113] in, This refers to the number of operations in which at least one confirmatory gaze precedes a micro-operation (such as precise defect annotation or key point clicking). This represents the total number of micro-operations counted. The higher the value, the more thorough the visual inspection is before delicate operations.

[0114] The task execution monitoring and error correction capability reflects the operator's speed in detecting abnormal states or task deviations and their sustained focus on risk areas. This indicator is based on the lead time for error detection. Anomaly monitoring coverage and secondary inspection rate Three underlying indicators characterize it.

[0115] Error detection lead time This reflects the time cost from "noticing an anomaly" to "starting correction." A smaller value indicates more agile monitoring and error correction. The calculation formula is as follows:

[0116] (25)

[0117] in, This indicates the number of errors or significant deviations detected. Indicates the first The moment when the line of sight first fell upon the abnormal area in this incident. This corresponds to the start time of the corrective action.

[0118] The anomaly monitoring coverage This describes the level of continuous attention paid to key risk areas during the task execution phase. A higher value indicates more thorough monitoring. The calculation formula is as follows:

[0119] (26)

[0120] in, This indicates the cumulative time during task execution when the line of sight is focused on areas displaying potential risks or critical statuses. This indicates the total time these areas have been in a "needs monitoring" state.

[0121] Second inspection rate This describes whether an operator will revisit the key AOI for confirmation after handling a critical incident, reflecting their risk awareness and level of hesitation. The specific calculation formula is as follows:

[0122] (27)

[0123] in, To review the number of times the AOI (Area of ​​Interest) was used for a critical event after it has been resolved, Total number of critical events. Under fatigue conditions. It may be abnormally high (excessive hesitation) or too low (insufficient monitoring).

[0124] Eye-fatigue iterative learning and adaptation capability primarily reflects the degree to which a drone operator's eye-movement behavior and fatigue control strategies are continuously optimized with experience during multiple bridge inspection missions. Under long-term training, the ideal state should manifest as: more efficient search of key areas, more accurate understanding of alarms and mission events, smoother vision-operation coordination, and a gradual decrease in the proportion of high-fatigue zones. This invention characterizes eye-fatigue iterative learning and adaptation capability through three secondary feature indicators: eye-movement efficiency improvement capability, perception accuracy improvement capability, and strategy diversity and self-regulation capability.

[0125] The eye-tracking efficiency improvement capability is used to reflect the degree of improvement in how quickly an operator can find key information and more continuously monitor risk areas after training, under the same task conditions. This indicator is measured by the improvement in first fixation time. Scan path length reduction And the improvement in AOI coverage efficiency within the task cycle Three underlying indicators characterize it.

[0126] The first fixation time increase The formula used to reflect whether the average response to the first fixation on the key area is faster after multiple training sessions is as follows:

[0127] (28)

[0128] in, The number of tasks included in the statistics; and The first The average first fixation time of the subtask in the baseline phase and the current phase is calculated using formula (5). The larger the value, the more significant the reduction in reaction time and the more significant the increase in reaction rate.

[0129] The amount of reduction in scan path length The formula used to reflect whether the gaze scanning path is more streamlined and efficient after multiple training sessions is as follows:

[0130] (29)

[0131] in, and The first Total length of the line-of-sight scanning path for the sub-task in the baseline and current phases. The larger the value, the more significant the path shortening and the more significant the improvement in search efficiency.

[0132] Improvement in AOI coverage efficiency during the task cycle This is used to reflect whether the operator's coverage of the key areas of the task is more sufficient and efficient under the same task background. The specific calculation formula is as follows:

[0133] (30)

[0134] in, and The first The AOI coverage of the sub-task in the baseline phase and the current phase is solved using formula (8). The larger the value, the more effective the monitoring of key areas has been after training.

[0135] The improved perception accuracy reflects the increase in the operator's accuracy in understanding alarms and task events when repeatedly performing similar inspection tasks under the same task background. This metric is measured by the improvement in target fixation accuracy. Improved alarm recognition accuracy and the decrease in false alarm / false alarm rate It is characterized by three underlying indicators.

[0136] The improvement in target fixation accuracy The formula used to characterize whether an operator's ability to fixate on a key target improves after multiple tasks is as follows:

[0137] (31)

[0138] in, Indicates the total number of tasks; , The first The fixation accuracy of the key target in the baseline phase and the current phase of the sub-task is calculated using formula (4).

[0139] The improvement in alarm recognition accuracy The formula used to reflect whether operators' understanding and response to alarm information are more accurate after iterative training is as follows:

[0140] (32)

[0141] in, Indicates the total number of tasks; , For the first The alarm identification accuracy of the sub-task in the baseline stage and the current stage is calculated using formula (10).

[0142] The decrease in false alarm / false negative rate The formula used to characterize whether the false alarm to false negative ratio in alarm processing has been effectively reduced after training and intervention is as follows:

[0143] (33)

[0144] in, Indicates the total number of tasks; , The first The false alarm / false negative rate of the sub-task in the baseline phase and the current phase is calculated using formula (12).

[0145] The strategy diversity and self-regulation capability are used to reflect the improvement in the operator's visual-operation coordination and overall fatigue control after iterative training. This indicator is expressed as the change in the number of strategy categories. Strategy entropy change and fatigue self-regulating eye movement characteristics Three underlying indicators characterize it.

[0146] The change in the number of strategy categories This describes whether the types of eye-tracking strategies an operator can employ become more diverse or gradually converge as training progresses. The specific calculation formula is as follows:

[0147] (34)

[0148] in, , represents the number of eye-tracking strategy categories identified in the baseline stage and the current stage, respectively, and is solved using formula (9).

[0149] The strategy entropy change The formula used to measure the tendency of eye-tracking strategies to converge from "disordered / dispersed" to "a few efficient modes" is as follows:

[0150] (35)

[0151] in, , These are the policy entropy values ​​for the baseline stage and the current stage, respectively. If the trained policy is more focused and efficient, then it often has... .

[0152] The fatigue self-adjusting eye movement characteristics The system's ability to maintain the operator in a "low fatigue / high performance" state during long-term tasks, after introducing closed-loop control and individualized thresholds, is characterized by the following calculation formula:

[0153] (36)

[0154] in, This refers to the cumulative time spent in the "low fatigue / high performance" range during the training period. This represents the cumulative time spent in the "high fatigue" zone. When closed-loop control is effective, Increased proportion Reduce, thus The gradual decrease over time indicates that the system has a certain degree of fatigue self-regulation capability.

[0155] This invention also provides a method for evaluating the fatigue status of drone operators, comprising the following steps:

[0156] Step 1: Construct the aforementioned drone operator fatigue status evaluation system.

[0157] The drone operator fatigue assessment system is constructed through the following steps, such as: Figure 2 As shown,

[0158] Step 1.1: Establish top-level evaluation objectives. The overall fatigue level of drone operators is established as the top-level evaluation indicator in this evaluation system framework.

[0159] Step 1.2: Constructing the primary evaluation dimensions. The top-level evaluation indicators are functionally decomposed. Based on the closed-loop logic of "perception-cognition-execution-adaptation" in UAV control tasks, the top-level indicators are decomposed into four primary capability dimensions: eye-tracking detection perception capability, cognitive integration and attention allocation capability, task execution-related eye-tracking performance capability, and eye-tracking-fatigue iterative learning and adaptation capability.

[0160] Step 1.3: Design secondary feature indicators. The primary evaluation dimensions constructed in Step 1.2 are decomposed. Considering that the primary dimensions are relatively macroscopic and abstract, making it difficult to directly achieve accurate assessment of fatigue state, this invention further decomposes each capability dimension from multiple attribute perspectives such as accuracy, timeliness, spatial breadth, and strategy abundance, thereby obtaining intermediate-layer feature indicators with clearer physical meaning.

[0161] Step 1.4: Define the underlying quantitative indicators. Further refine the secondary characteristic indicators designed in Step 1.3 to establish the underlying evaluation indicators that ultimately constitute the evaluation system architecture.

[0162] Step 1.5: Construct the final evaluation system framework. Based on steps 1.1-1.4, construct a four-layer evaluation system for the fatigue status of UAV operators.

[0163] Step 2: Design detailed evaluation indicators to represent the underlying level.

[0164] Step 3: Calculate the weighting of the task situation gating combination based on the normal cloud model and Shapley value game.

[0165] Step 4: Obtain the final weights of each indicator to get the fatigue status score.

[0166] In one embodiment, step 3 specifically involves the following steps:

[0167] Step 3.1: Data Preprocessing; Considering the sensitivity of objective weighting methods to outlier data, the original data is first preprocessed to remove or fill missing and outlier values. The removed values ​​are filled with random Gaussian values. Dimensional normalization is performed on the evaluation data corresponding to the underlying evaluation indicators, normalizing the score for each indicator to the positive [0,100] interval. Different dimensional normalization methods are selected based on different data distribution characteristics.

[0168] To address the issue of uniform distribution in the evaluation data, a linear dimensional normalization method is employed to eliminate the influence of dimensions. Specifically, the evaluation data corresponding to all evaluation indicators conforming to linear dimensional normalization are positively oriented. For evaluation indicators that are negatively correlated with the intelligence level of the unmanned system, their corresponding evaluation data are inversely oriented. Then, a normalization operation is performed. The linear dimensional normalization formula is as follows:

[0169] (37)

[0170] in, This represents the value obtained after normalizing the evaluation data; This is the original evaluation dataset; for The minimum value in; for The maximum value in.

[0171] To address the significant differences in the distribution of evaluation data across different precision ranges, a piecewise dimensional normalization method is employed to eliminate the influence of dimensions. The specific method is as follows:

[0172] All evaluation data corresponding to evaluation indicators that conform to piecewise dimension normalization are positively oriented. For evaluation indicators that are negatively correlated with the intelligence level of the unmanned system, their corresponding evaluation data are inversely oriented. A minimum value for the evaluation data is set. The corresponding score is 0; the passing score for the evaluation data is... The corresponding score is Maximum value of evaluation data The corresponding score is 100 points.

[0173] pass and Construct a range of values ​​in linear functions ,Will range value conversion Interval values; through and Construct a range of values ​​in linear functions ,Will range value conversion Interval values. Based on linear functions. and Normalize the data that meets the linear quantization conditions.

[0174] To address the issue of continuous and Gaussian-like distribution characteristics in the evaluation data, a dimensional normalization method is employed to eliminate the influence of dimensions. Specifically:

[0175] All evaluation data corresponding to evaluation indicators that conform to the normalization of quantities are positively oriented. For evaluation indicators that are negatively correlated with the intelligence level of the unmanned system, the corresponding evaluation data are inversely oriented. A minimum value for the evaluation data is set. The corresponding score is 0; the passing score for the evaluation data is... The corresponding score is Maximum value of evaluation data The corresponding score is 100 points.

[0176] pass and The range of construct values ​​is in Log-normalized function and through and Construct a range of values ​​in Log-normalized function ,Will range value conversion Interval values. Based on linear functions. and Normalize the data that meets the quantification criteria.

[0177] Based on the three evaluation index dimension normalization methods mentioned above, the evaluation data corresponding to the bottom-level evaluation indexes in the designed evaluation system framework are normalized. The table of selection of dimensionless methods for bottom-level evaluation indexes is shown in Table 1.

[0178] Table 1 Summary of methods for removing dimensions from underlying evaluation indicators

[0179] .

[0180] Step 3.2: Calculate the subjective weights of language cognition based on the normal cloud model; traditional precise numerical scoring is difficult to describe the ambiguity and randomness of experts' fatigue definitions. This invention utilizes the normal cloud model to transform experts' qualitative importance ratings into scores that include expected values. ,entropy and hyperentropy The quantitative cloud droplets are used to achieve a digital mapping of subjective perception. The specific method includes the following steps:

[0181] Step 3.21: Construct a language evaluation scale cloud generator.

[0182] Let the standard evaluation domain be To eliminate the differences in the dimensions of the indicators, the total number of importance levels for the evaluation indicators is set to an odd number. The serial number corresponding to each level The first [section] is generated based on the golden ratio. The cloud model feature vectors corresponding to each importance level The calculation logic is as follows:

[0183] (38)

[0184] in, Indicates the intermediate importance level number; Indicates the highest importance level number; This represents the expected value, indicating the central position of experts' evaluation of the importance of this indicator; Entropy represents the fuzziness of the evaluation concept, reflecting the range of values ​​that experts believe the importance level covers. This refers to hyperentropy, or entropy of entropy, which reflects the degree of hesitation and randomness of experts when giving scores. These are the discrete threshold coefficients for cloud droplets, used to constrain hyperentropy. With entropy The ratio is usually taken as 10.

[0185] Step 3.2.2: Generate expert cognitive cloud map.

[0186] invite Experts in the field on the first The importance of the first underlying evaluation indicator is determined, if the first... One expert believes that this indicator is of paramount importance. For each level, the corresponding standard cloud model feature is directly referenced. As an individual evaluation cloud of the expert The calculation formula is as follows:

[0187] (39)

[0188] in, Indicates if the first The expert commented on the first The expected score, entropy, and hyperentropy of each underlying evaluation metric.

[0189] Based on this, calculate the first... A comprehensive floating cloud model with several underlying evaluation indicators. Used to reflect the expert group's opinion on the first The comprehensive opinion of the underlying evaluation indicators is calculated using the following formula:

[0190] (40)

[0191] Step 3.2.3: Calculate the subjective weight based on cloud similarity.

[0192] To comprehensively measure the importance and credibility of the indicators, an ideal indicator with theoretically perfect importance and no cognitive ambiguity or disagreement is set as a benchmark, namely the absolute importance standard cloud. Based on the distribution characteristics of Gaussian clouds, the first... A comprehensive floating cloud model based on multiple indicators With absolutely important standard cloud similarity between This similarity score characterizes how closely the index approximates the ideal state of absolute importance, and is calculated using the following formula:

[0193] (41)

[0194] Finally, regarding similarity After normalization, we obtain the first... Subjective weights of each indicator The normalization formula is as follows:

[0195] (42)

[0196] in, This indicates the total number of evaluation indicators.

[0197] The final set of subjective weights for the indicators is obtained. .

[0198] Step 3.3: Calculate the objective weight of marginal contribution based on the Shapley value of cooperative game.

[0199] The specific method includes the following steps:

[0200] Step 3.3.1: Construct a reference sample set and define the feature function.

[0201] Construct a reference sample dataset containing different fatigue states. :

[0202] (43)

[0203] in, For the first time collected One underlying indicator; For the first The actual fatigue status labels corresponding to the underlying indicators were determined using the Karolinska Sleepiness Scale (KSS).

[0204] Based on the reference sample dataset ,set up This is the set of all underlying evaluation metrics. For any subset in Define the fuzzy measure feature function For a subset of indicators The trace of the inter-class scatter matrix in the sample space is used to quantify the ability of this index combination to distinguish different fatigue levels. The calculation formula is as follows:

[0205] (44)

[0206] in, Indicates based on a subset of indicators The constructed inter-class scatter matrix. The larger this value, the more separated the samples of different fatigue states are in the feature space constituted by these indicators, and the better the recognition effect. Indicates the number of preset fatigue state categories; Indicates the first Prior probability of fatigue-like samples; No. fatigue-like samples in the index subset Mean vector in dimension; This indicates that all samples are in the subset of indicators. The global mean vector in dimension; This represents the matrix transpose operation; The trace of a matrix is ​​the sum of its diagonal elements, used to condense matrix information into a scalar value for comparison.

[0207] Step 3.3.2: Calculate the objective weights of the indicators based on Shapley values.

[0208] No. The objective weight of each underlying metric depends on its inclusion in different subsets. The average information gain resulting from this, according to the axiomatic definition of the Shapley value, is the... Shapley values ​​of underlying metrics The calculation formula is as follows:

[0209] (45)

[0210] in, Indicates from the entire set of indicators Remove indicators The remaining set afterwards; Indicates the number of indicators contained in the subset; Indicates the first The variation value of each evaluation indicator; The higher the value, the more information the indicator provides.

[0211] To standardize the units of measurement, the calculated Shapley values ​​are... Perform absolute value normalization to obtain the first... The final objective weight of each indicator The calculation formula is as follows:

[0212] (46)

[0213] Finally, the objective weight set of the indicators is obtained. .

[0214] Step 3.4: Calculate the combined weights based on task situation gating. The credibility of subjective weights is affected by the dispersion of expert cognition. This invention proposes a geometric logarithmic convergence method for task situation gating, which nonlinearly and dynamically fuses the weights obtained in steps 3.2 and 3.3. The specific method is as follows:

[0215] Step 3.4.1: Subjective weight credibility modeling.

[0216] Since the credibility of subjective weights is affected by the dispersion of expert cognition, an overall cognitive ambiguity index is defined. The calculation formula is as follows:

[0217] (47)

[0218] in, Indicates the total number of evaluation indicators; Indicates the first The superentropy of each underlying evaluation index is calculated by formula (39).

[0219] When expert opinions become increasingly divergent (i.e.) The larger the value, the lower the credibility of expert perception; thus, a subjective credibility factor is constructed. The calculation formula is as follows:

[0220] (48)

[0221] in, and This is the adjustment coefficient; The information entropy of the subjective weight vector is used to prevent the weight allocation from being too extreme. This is the reference entropy. A larger value indicates that the subjective weight is more reliable.

[0222] Step 3.4.2: Modeling the strength of objective weighted evidence.

[0223] The effectiveness of objective weights depends on the quality of real-time data. The definition of the first... Data efficiency of each underlying evaluation indicator With stability coefficient The calculation formula is as follows:

[0224] (49)

[0225] in, This represents the theoretical total number of samples within the sliding time window, which is the product of the window duration and the sampling frequency. This indicates that the current length is Within the sliding window, the first The number of data points that are missing or not collected for each indicator; This indicates that the current length is Within the sliding window, the first The number of data points identified as outliers in each indicator; This is a stability adjustment coefficient used to control the sensitivity of the stability score to data fluctuations. Indicates the number of elements in the current sliding window. Normalized score time series vectors of each indicator; This is the median absolute deviation function, used to robustly assess the dispersion of a sequence; This represents the reference deviation benchmark value, which is the median of all index deviations, used to eliminate the influence of dimensions. It is a very small positive number used to prevent calculation errors caused by a denominator of zero.

[0226] Based on data efficiency With stability coefficient Calculate the strength of objective evidence The calculation formula is as follows:

[0227] (50)

[0228] Step 3.4.3 Construct the task status gating factor.

[0229] To enable the weights to be adaptively adjusted under different task conditions, this invention introduces a gating coefficient. This is used to control the contribution ratio of subjective and objective factors in the combined weighting. It defines the task situation strength. The calculation formula is as follows:

[0230] (51)

[0231] in, This represents the normalized value of task complexity. This represents the normalized value of alarm event density per unit time; Normalized value for interface / mode switching density per unit time; This is a normalized measure of risk exposure intensity. is the proportionality coefficient, and the sum is 1 and both are greater than 0.

[0232] Further calculation of the gating coefficient The calculation formula is as follows:

[0233] (52)

[0234] in, This is the gating sensitivity coefficient. When the mission situation is more complex and the objective evidence is stronger... An increase indicates a greater emphasis on objective weights; conversely, a decrease indicates a greater emphasis on subjective weights.

[0235] 3.4.4: Information geometric logarithmic convergence generates the final combined weights.

[0236] The weight vector is treated as a distribution on a probability simplex, and a logarithmic opinion pooling technique from information geometry is used for fusion to minimize the weighted KL divergence and preserve the geometric structure of the probability space. The formula for calculating the unnormalized combined weights of the indicators is as follows:

[0237] (53)

[0238] The normalization formula is as follows:

[0239] (54)

[0240] Finally, the optimal combination weight set of each underlying indicator in the drone operator fatigue state evaluation system is obtained. .

[0241] Example

[0242] Using the Yangtze River Bridge No. 3 UAV inspection mission as the application scenario, the experiment was conducted within a virtual bridge inspection corridor of approximately 10 km in length, and a simulation platform was constructed using real UAV mission planning software. The scenario included standard inspection targets such as conventional bridge deck structures, bridge towers, and cable anchors, with randomly superimposed defects such as cracks, spalling, and corrosion, as well as various mission events such as wind disturbances, communication alarms, and flight path deviations, to simulate long-duration, high-load actual inspection conditions.

[0243] The experiment involved K participants with drone operation experience, each of whom completed three rounds of full inspection tasks.

[0244] 1) Baseline Wheel (Low Fatigue): The task lasts about 20 minutes, with moderate task intensity, and contains only a small number of defects and alarm events. It is used to collect the individual's "low fatigue / high performance" baseline eye movement pattern.

[0245] 2) Cumulative Round (Medium Fatigue): Executed immediately after the baseline round ends, it increases defect density and task complexity, inserts multi-task switching and temporary instructions, and lasts for about 30 minutes. It is used to induce medium fatigue.

[0246] 3) High-load wheel (high fatigue): Continue execution after the cumulative wheel, appropriately extend the task time and increase the alarm frequency, requiring the subject to maintain continuous monitoring and frequent operation, in order to collect eye movement and operation behavior under high fatigue state.

[0247] The experiment used head-mounted or telemetry-based eye-tracking devices to record gaze data at a sampling frequency of no less than 120 Hz; simultaneously, UAV control commands, task event logs, and system interface status were recorded. Based on the indicator system constructed in step one, indicators were extracted from the raw eye-tracking data. Furthermore, using subjective questionnaires (such as KSS and NASA-TLX) and the automatic fatigue grading results from the "UAV Operator Fatigue Status Evaluation System" constructed based on this invention, each task data segment was labeled as "mild fatigue," "moderate fatigue," and "severe fatigue" for subsequent performance evaluation and comparative analysis. To ensure statistical sufficiency, the entire experiment was repeated at least 100 times under the same scenario configuration, constructing a comprehensive dataset containing multiple subjects, multiple rounds, and multiple fatigue levels.

[0248] Based on the collected eye-tracking and task execution data, using Outlier samples were identified using a combination of criteria and box plot methods. Isolated outliers were corrected using neighborhood median or linear interpolation. For a small number of missing data points caused by device frame drops, the mean of similar task segments or time-series-based interpolation methods were used to complete the data. The statistics of missing and outlier values ​​for each indicator are shown in Table 2.

[0249] Table 2. Statistics on missing values ​​and outliers for each indicator. .

[0250] Subsequently, linear, piecewise, or logarithmic functions were selected for dimensional normalization of different types of underlying indicators, and indicators negatively correlated with fatigue levels were first positively normalized. After normalization, all underlying indicators were scaled to the [0,100] interval to ensure that weighted calculations could be performed under a unified dimension. The dimensional normalization methods used for each indicator are shown in Table 1.

[0251] To verify the effectiveness of the "Unmanned Aerial Vehicle Operator Fatigue Status Evaluation System" constructed in this invention and the consistency of results under different weighting methods, this section designs three fatigue index calculation schemes for comparison:

[0252] 1) Option 1: Equal-weighted linear summation of fatigue index After normalization, all underlying indicators are assigned the same weight to obtain a baseline fatigue index, which is used to characterize the "simple average evaluation effect without modeling".

[0253] 2) Option Two: AHP Subjective Weight Fatigue Index Five domain experts were invited to score the importance of the four primary capability dimensions and key secondary indicators.

[0254] 3) Option 3: Task Situation Gating Combination Weight Fatigue Index The task situation is gated and weighted based on subjective cloud model mapping and objective Shapley game. The weight values ​​corresponding to each underlying indicator are listed in Table 3.

[0255] Table 3. Weight values ​​for each underlying indicator

[0256] .

[0257] During the weight optimization process, the "sum of changes in the contribution differences of the overall underlying evaluation indicators" is used as the convergence and effectiveness criterion. Figure 4 and Figure 5 The changes in change with the number of iterations are presented for initial values ​​using equal weights and AHP-weighted weights, respectively. It can be seen that with equal weights as the initial value, the initial change is approximately 4.8, and it converges to below the set threshold after about 60 iterations. With AHP as the initial value, the initial change is approximately 3.7, and it stabilizes after about 45 iterations. In both cases, the change shows a monotonically decreasing trend, indicating that the model's attribution weight optimization process is stable and effective.

[0258] To further compare the evaluation effects of different fatigue indices, this section conducts performance analysis from the following two aspects:

[0259] 1) Consistency of Fatigue Grading: Three fatigue levels labeled by experts were used as "reference labels." The consistency rate between the fatigue index and the labels, as well as the weighted Kappa coefficient, were statistically analyzed under different schemes. The results show that the equal weighting scheme... The grading consistency rate is approximately 78%, AHP scheme The success rate is increased to approximately 84%, while the task situation gating combination scheme adopted in this invention... It can be further increased to about 89%, and the Kappa coefficient increases from 0.62 to 0.78.

[0260] 2) Index stability and separability: For samples under the same fatigue level, calculate the mean and standard deviation of the fatigue index, and use the standard deviation to characterize the stability of the evaluation results; at the same time, calculate the mean difference and overlap between different fatigue levels (such as Fisher's discrimination rate). Figure 6 and Figure 7 A comparison of the score distributions of the three schemes across three fatigue levels is presented. It can be seen that... The intra-class variance is relatively large, especially the distribution of moderate fatigue samples is relatively discrete; This reduced the intra-class dispersion to some extent; While maintaining the mean gradient (mild < moderate < severe), the overlap of the three score intervals is significantly reduced, and the overall standard deviation is reduced by about 20% compared to the AHP scheme.

Claims

1. A fatigue assessment system for unmanned aerial vehicle (UAV) operators, characterized in that: It includes four primary indicators: eye-tracking detection and perception ability, cognitive integration and attention allocation ability, task-related eye-tracking performance ability, and indicators representing eye-tracking-fatigue iterative learning and adaptation ability. The secondary feature indicators corresponding to eye-tracking detection and perception capabilities include: eye-tracking perception accuracy, eye-tracking perception time, eye-tracking perception breadth, and eye-tracking strategy abundance. Secondary characteristic indicators of cognitive integration and attention allocation ability include: accuracy of alarm and target understanding, decision-making and attention switching time, and information processing and search complexity; Secondary indicators of task-related eye-tracking performance include: visual-operational coordination efficiency, visual-operational accuracy, and task self-monitoring and error correction capabilities; The secondary indicators characterizing eye-fatigue iterative learning and adaptation capabilities include: eye-movement efficiency improvement capability, perception accuracy improvement capability, and strategy diversity and self-adjustment capability. Each secondary indicator has a tertiary indicator.

2. The method for evaluating the fatigue status of UAV operators as described in claim 1, characterized in that: The eye-tracking accuracy is used to characterize whether the operator's gaze is focused on the task-related area and whether the fixation on key targets is adequate. Increased fatigue leads to gaze drift and an increased probability of missed target scanning. This metric is measured by fixation hit rate. gaze position error Key scenario coverage Key target fixation accuracy Four tertiary indicators are used for characterization; The eye-tracking time is measured by the average first fixation time. and average fixation duration The eye-tracking perception span is characterized by two tertiary indicators; the eye-tracking perception span is measured by the maximum fixation range. and task AOI coverage The abundance of eye-tracking strategies is characterized by two tertiary indicators, namely the number of identifiable eye-tracking strategy categories. Characterize it.

3. The method for evaluating the fatigue status of UAV operators as described in claim 2, characterized in that: The alarm and target understanding accuracy is used to characterize the operator's ability to correctly perceive and interpret key information at the visual level, and is measured by alarm recognition accuracy. Accuracy of task event recognition and false alarm / false negative rate The decision-making and attention switching time is characterized by three tertiary indicators, which are expressed as decision reaction time. Note the switching time. and the cost of multitasking Three tertiary indicators characterize the information processing and search complexity: reflecting the eye-tracking search workload and hesitation required by an operator to make a decision or handle a task event. This indicator is measured by the number of AOI scans per unit time. complexity of line-of-sight trajectory and search strategy stability Three underlying indicators characterize it.

4. The method for evaluating the fatigue status of UAV operators as described in claim 3, characterized in that: The aforementioned indicators of eye-tracking performance related to task execution are characterized by three secondary features: visual-operation coordination efficiency, visual-operation accuracy, and task execution monitoring and error correction capabilities. The eye-operation coordination efficiency is used to characterize whether the operator's gaze can lock onto the relevant interface or control in advance when performing operations such as joystick movement, mode switching, and parameter adjustment, by measuring eye-hand lead time. Eye-hand synchronization rate and continuous operation segment view-hand consistency Three tertiary indicators characterize this; The visual-operational accuracy reflects whether the operator's gaze is sufficiently focused on the target area before performing a single fine operation, and whether the gaze has significantly deviated before a misoperation occurs. This is measured by the operator's gaze focus before operation. Deviance of line of sight before misoperation Confirm gaze ratio with micro-operations Three tertiary indicators characterize this; The task execution monitoring and error correction capabilities reflect the operator's speed of detecting abnormal states or task deviations and their continuous attention to risk areas, through error detection lead time. Anomaly monitoring coverage and secondary inspection rate Three underlying indicators characterize it.

5. The method for evaluating the fatigue status of UAV operators as described in claim 1, characterized in that: The indicators characterizing eye-tracking fatigue iterative learning and adaptation capabilities are based on three secondary features: the ability to improve eye-tracking efficiency, the ability to improve perceptual accuracy, and the ability to diversify strategies and self-regulate. The eye movement efficiency improvement capability is measured by the increase in first fixation time. Scan path length reduction And the improvement in AOI coverage efficiency within the task cycle Three tertiary indicators characterize this; The improved perception accuracy is achieved by increasing the target fixation accuracy. Improved alarm recognition accuracy and the decrease in false alarm / false alarm rate Three tertiary indicators are used for characterization; The strategy diversity and self-adjustment capabilities are expressed through the change in the number of strategy categories. Strategy entropy change and fatigue self-regulating eye movement characteristics Three tertiary indicators characterize the system.

6. A method for evaluating the fatigue status of drone operators, characterized in that: Includes the following steps: Step 1: Construct a drone operator fatigue status evaluation system as described in any one of claims 1-5; Step 2: Design detailed evaluation indicators to represent the underlying level; Step 3: Calculate the weighted combination of task situation gating based on the normal cloud model and Shapley value game; Step 4: Obtain the final weights of each indicator to get the fatigue status score.

7. The method for evaluating the fatigue status of UAV operators as described in claim 6, characterized in that: The specific steps of step 3 are as follows: Step 3.1: Data Preprocessing; First, the raw data is preprocessed to remove or fill missing and outlier values. Removed values ​​are filled with random Gaussian values. Dimensional normalization is performed on the evaluation data corresponding to the underlying evaluation indicators, normalizing the score for each indicator to the positive interval [0,100]. Different dimensional normalization methods are selected based on different data distribution characteristics. For cases where the evaluation data distribution is uniform, linear dimensional normalization is used to eliminate its dimensional influence. For cases where the evaluation data distribution has significant differences within different precision intervals, piecewise dimensional normalization is used to eliminate its dimensional influence. For cases where the evaluation data distribution is continuous and exhibits a Gaussian-like distribution, logarithmic dimensional normalization is used to eliminate its dimensional influence. Step 3.2: Calculate the subjective weights of language cognition based on the normal cloud model; use the normal cloud model to transform the expert's qualitative importance rating into a rating that includes the expected score. ,entropy and hyperentropy Quantitative cloud droplets enable the digital mapping of subjective perception; Step 3.3: Calculate the objective weight of marginal contribution based on the Shapley value of cooperative game theory; Step 3.4: Calculate the combined weights based on task situation gating, and perform nonlinear dynamic fusion of the weights obtained in Step 3.2 and Step 3.

3.

8. The method for evaluating the fatigue status of UAV operators as described in claim 7, characterized in that: In step 3.2, calculating the subjective weights of language cognition based on the normal cloud model includes the following steps: Step 3.2.1: Construct a language evaluation scale cloud generator: Let the standard evaluation domain be... To eliminate the differences in the dimensions of the indicators, the total number of importance levels of the evaluation indicators is set to an odd number. The serial number corresponding to each level The first one is generated based on the golden ratio. The cloud model feature vectors corresponding to each importance level The calculation logic is as follows: in, Indicates the intermediate importance level number; Indicates the highest importance level number; This represents the expected value, indicating the central position of experts' evaluation of the importance of this indicator; Entropy represents the fuzziness of the evaluation concept, reflecting the range of values ​​that experts believe the importance level covers. Hyperentropy represents the degree of hesitation and randomness among experts when assigning scores; These are the discrete threshold coefficients for cloud droplets, used to constrain hyperentropy. With entropy proportional relationship; Step 3.2.2: Generate Expert Cognition Cloud Map: Invite Experts in the field on the first The importance of the first underlying evaluation indicator is determined, if the first... One expert believes that this indicator is of paramount importance. For each level, the corresponding standard cloud model feature is directly referenced. As an individual evaluation cloud of the expert The calculation formula is as follows: in, Indicates if the first The expert commented on the first Expected score, entropy, and hyperentropy of each underlying evaluation metric; Calculate the first A comprehensive floating cloud model with several underlying evaluation indicators. Used to reflect the expert group's opinion on the first The comprehensive opinion of the underlying evaluation indicators is calculated using the following formula: ； Step 3.2.3: Calculate the subjective weight based on cloud similarity: Set an ideal indicator with theoretically full importance and no cognitive ambiguity or disagreement as a benchmark, standard cloud. Based on the distribution characteristics of Gaussian clouds, the first... A comprehensive floating cloud model based on multiple indicators With absolutely important standard cloud similarity between This similarity score characterizes the degree to which the index approximates the ideal state of absolute importance, and the calculation formula is as follows: Finally, regarding similarity After normalization, we obtain the first... Subjective weights of each indicator The normalization formula is as follows: in, Indicates the total number of evaluation indicators. The final set of subjective weights for the indicators is obtained. .

9. The method for evaluating the fatigue status of UAV operators as described in claim 8, characterized in that: In step 3.3, calculating the objective weight of the marginal contribution based on the Shapley value in cooperative game theory includes the following steps: Step 3.3.1: Construct a reference sample set and define feature functions: Construct a reference sample dataset containing different fatigue states. : ， in, For the first time collected One underlying indicator; For the first The actual fatigue state labels corresponding to each underlying indicator were determined using the Karolinska Sleepiness Scale (KSS). Based on the reference sample dataset ,set up For the set of all underlying evaluation metrics, any subset in Define the fuzzy measure feature function For a subset of indicators The trace of the between-class scatter matrix in the sample space is calculated using the following formula: ， in, Indicates based on a subset of indicators The constructed inter-class scatter matrix, Indicates the number of preset fatigue state categories; Indicates the first Prior probability of fatigue-like samples; No. fatigue-like samples in the index subset Mean vector in dimension; This indicates that all samples are in the subset of indicators. The global mean vector in dimension; This represents the matrix transpose operation; Represents the trace of a matrix; Step 3.3.2: Calculate the objective weights of the indicators based on the Shapley value: According to the axiomatic definition of the Shapley value, the first... Shapley values ​​of underlying metrics The calculation formula is as follows: in, Indicates from the entire set of indicators Remove indicators The remaining set afterwards; Indicates the number of indicators contained in the subset; Indicates the first The variation value of each evaluation indicator; To standardize the units of measurement, the calculated Shapley values ​​are... Perform absolute value normalization to obtain the first... The final objective weight of each indicator The calculation formula is as follows: Finally, the objective weight set of the indicators is obtained. .

10. The method for evaluating the fatigue status of UAV operators as described in claim 9, characterized in that: In step 3.4, calculating the combined weights based on task situation gating includes the following steps: Step 3.4.1: Subjective weight credibility modeling, defining the overall cognitive ambiguity index. The calculation formula is as follows: in, Indicates the total number of evaluation indicators; Indicates the first Hyperentropy of a fundamental evaluation metric; Constructing subjective credibility factors The calculation formula is as follows: (48) in, and This is the adjustment coefficient; The information entropy of the subjective weight vector is used to prevent the weight allocation from being too extreme. As a reference entropy, A larger value indicates that the subjective weight is more reliable; Step 3.4.2: Objective weighted evidence strength modeling, defining the first... Data efficiency of each underlying evaluation indicator With stability coefficient The calculation formula is as follows: in, This represents the total number of theoretical samples within the sliding time window; This indicates that the current length is Within the sliding window, the first The number of data points that are missing or not collected for each indicator; This indicates that the current length is Within the sliding window, the first The number of data points identified as outliers in each indicator; This is a stability adjustment coefficient used to control the sensitivity of the stability score to data fluctuations. Indicates the number of elements in the current sliding window. Normalized score time series vectors of each indicator; This is the median absolute deviation function, used to robustly assess the dispersion of a sequence; This represents the reference deviation benchmark value, which is the median of all index deviations, used to eliminate the influence of dimensions. It is a very small positive number, used to prevent calculation errors caused by a denominator of zero. Based on data efficiency With stability coefficient Calculate the strength of objective evidence The calculation formula is as follows: ； Step 3.4.3: Construct the task status gating factor and introduce the gating coefficient. It is used to control the contribution ratio of subjective and objective factors in combined weighting and to define the task situation strength. The calculation formula is as follows: in, This represents the normalized value of task complexity. This represents the normalized value of alarm event density per unit time; Normalized value for interface / mode switching density per unit time; This is a normalized measure of risk exposure intensity. The constants are proportionality coefficients, and their sums are all greater than 0, with a total of 1. Further calculation of the gating coefficient The calculation formula is as follows: in, This is the gating sensitivity coefficient, which is used when the mission situation is more complex and the objective evidence is stronger. An increase indicates a greater emphasis on objective factors; conversely, a decrease indicates a greater emphasis on subjective factors. Step 3.4.4: Log-based convergence in information geometry generates the final combined weights. The weight vector is treated as a distribution on the probability simplex, and a log-based opinion pooling method from information geometry is used for fusion to minimize the weighted KL divergence and preserve the geometric structure of the probability space. The formula for calculating the unnormalized combined weights of the indicators is as follows: The normalization formula is as follows: Finally, the optimal combination weight set of each underlying indicator in the drone operator fatigue state evaluation system is obtained. .

Images