Method, device and equipment for adjusting navigation task based on human intelligence function allocation

Abstract

This application discloses a method, apparatus, and equipment for adjusting navigation tasks based on human intelligence function allocation, belonging to the field of ship control technology. The method includes: performing cross-band dynamic functional connectivity analysis on physiological signals to extract neural state trajectories; performing task semantic parsing on navigation operation logs to extract behavioral log streams corresponding to the ship's sub-task types; constructing positive and negative sample pairs based on the synchronization state of the neural state trajectories and behavioral log streams; mapping the neural state trajectories, behavioral log streams, and environmental pressure vectors to a maritime context-aware latent space with the optimization objectives of maximizing the mutual information of positive sample pairs, minimizing the distance of negative sample pairs, and maximizing the correlation between negative sample pairs and the environmental pressure vector, thereby obtaining a comprehensive contextualized common representation of the crew; and calculating metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation tasks. This method can improve the contextual sensitivity of task adjustment.

Description

Technical Field

[0001] This application belongs to the field of ship control technology, and in particular relates to a method, device and equipment for adjusting navigation tasks based on human intelligence function allocation. Background Technology

[0002] With the increasing intelligence of ships, modern vessels are equipped with multi-source sensing systems such as radar, Automatic Identification System (AIS), and electronic charts. The collaboration between humans and intelligent systems has become a core element in ensuring navigational safety. However, current industry practices mainly focus on enhancing environmental perception and optimizing automated control logic, while the real-time cognitive state modeling and adaptive intervention mechanisms for the dynamic variable of "humans" remain weak.

[0003] Most navigation accidents are directly related to crew situational awareness failure, cognitive overload, or metacognitive bias. In high-pressure scenarios such as poor visibility, sudden changes in wave height, or dense ship traffic, existing systems rely on preset thresholds and fixed rule bases for task allocation. In complex navigation situations, they exhibit shortcomings such as delayed response, coarse granularity that ignores individual rhythms, and insufficient adaptability. They also lack situational sensitivity and cannot accurately perceive and respond to the instantaneous coupling changes of multi-source situational elements.

[0004] Therefore, there is an urgent need for a task adjustment method that can improve situational sensitivity. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a navigation mission adjustment method, apparatus, and device based on human intelligence function allocation, which can improve the situational sensitivity of mission adjustment.

[0006] Firstly, this application provides a method for adjusting navigation missions based on the allocation of human intelligence functions, the method comprising: The physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship are obtained. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data. Cross-band dynamic functional connectivity analysis was performed on the physiological signals to extract neural state trajectories; Perform task semantic parsing on the navigation operation logs to extract the behavior log streams corresponding to the sub-task types of the ship; Based on the synchronization status of the neural state trajectory and the behavior log stream, construct positive sample pairs and negative sample pairs; With the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, the neural state trajectory, the behavior log stream, and the environmental pressure vector are mapped to the maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew. Metacognitive alignment loss is calculated based on the comprehensive contextualized common representation to adjust the crew's navigation mission.

[0007] According to one embodiment of this application, the physiological signal includes EEG signal and HRV signal, and the step of performing cross-band dynamic functional connectivity analysis on the physiological signal to extract neural state trajectories includes: The EEG and HRV signals are synchronously time-separated to extract the neuropsychological load segments corresponding to the key stages of navigation operations, thus obtaining EEG segments and HRV segments. The EEG segment is subjected to multi-scale bandpass filtering to separate the theta band phase signal and the gamma band amplitude signal, and the HRV segment is subjected to time-domain sequence extraction to generate the RR interval sequence; Perform a Fourier transform on the RR interval sequence to calculate the HRV high-frequency power characteristics; The theta-band phase signal and the gamma-band amplitude signal are subjected to Hilbert transform to obtain the corresponding instantaneous phase sequence and instantaneous amplitude sequence, respectively. The instantaneous phase sequence is discretized into multiple phase intervals, and the average value of the instantaneous amplitude in each phase interval is calculated to generate a phase-amplitude coupled feature vector. The phase-amplitude coupling feature vector and the HRV high-frequency power feature are tensor-concatenated, and the concatenated high-dimensional feature vector is processed by a time-series sliding window to obtain the neural state trajectory.

[0008] According to one embodiment of this application, the behavior log stream includes steering command response delay features and radar target omission features. The step of performing task semantic parsing on the navigation operation log to extract the behavior log stream corresponding to the ship's sub-task type includes: Identify time anchors for critical operations in the navigation operation event log, including collision avoidance, turning, and berthing / departure. Centered on the aforementioned time anchor point, extend a time window forward and backward respectively, and extract the rudder command execution record, radar target interaction sequence and alarm response timestamp within the corresponding time period; The timing of the instruction issuance time and the instruction execution time in the rudder command execution record are aligned to determine the rudder command response delay characteristics; The radar target interaction sequence is subjected to target trajectory continuity detection to identify time periods of unresponsive targets and determine radar target omission features; The alarm trigger time and alarm response time in the alarm response timestamp are time-aligned to determine the alarm response delay characteristics; The steering command response delay feature, the radar target omission feature, and the alarm response delay feature are grouped according to the sub-task type corresponding to the time window to obtain the behavior log stream.

[0009] According to one embodiment of this application, after obtaining the behavior log stream and before constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream, the method further includes: The operation trajectory in the rudder command execution record is segmented into a time series to extract the trajectory segment corresponding to the navigation operation event; The trajectory segment is solved in reverse based on the ship dynamics model of the ship to generate an operation intention vector; The cosine similarity between the operational intent vector and the feature vector of the crew member's actual executed actions is used as the intent-execution deviation feature. The intent-execution deviation feature is fused with the steering command response delay feature, the radar target omission feature, and the alarm response delay feature to update the behavior log stream.

[0010] According to one embodiment of this application, constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream includes: The corresponding environmental pressure weight coefficient is determined based on the ship's navigation sub-task type and the semantic association strength of the environmental pressure vector. Based on the neural state trajectory and the behavior log stream, the time-varying mutual information value between the rate of change of neural state and the rate of change of behavioral abnormality index within each time window is determined, and the time-varying mutual information value is weighted according to the environmental pressure weighting coefficient to generate a weighted synchronization intensity sequence. Based on the navigation sub-task type and the weighted synchronization intensity sequence, online adaptive kernel density estimation is performed on the ship's historical mission data to obtain the probability density distribution curve of the weighted synchronization intensity; The upper quartile and lower quartile are extracted from the probability density distribution curve as dynamic dividing points. Time window segments with weighted synchronization intensity above the upper quartile are marked as positive sample pairs, and time window segments with weighted synchronization intensity below the lower quartile are marked as negative sample pairs. The regions above the upper quartile and the regions below the lower quartile exhibit bimodal separation characteristics in the probability density distribution.

[0011] According to one embodiment of this application, the optimization objective is to map the neural state trajectory, the behavior log stream, and the environmental pressure vector to a maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew member, with the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector. This includes: The outputs of the neural state encoder, behavior log encoder, and environmental stress encoder are mapped to a contrastive learning space of a unified dimension via a projection head network to construct a multimodal encoder group. In the contrastive learning space, the lower bound of mutual information between the neural state hidden representation and the behavior log hidden representation in the positive sample pair is estimated by noise contrast, and a mutual information maximization loss term is constructed. Based on the cosine distance between the negative sample pair and the hidden representation of the neural state and the hidden representation of the behavior log, a distance minimization loss term is constructed; Spearman rank correlation analysis was performed on the fusion latent representation and environmental pressure latent representation of the negative sample pairs to construct a correlation-maximizing loss term; The mutual information maximization loss term, the distance minimization loss term, and the correlation maximization loss term are weighted and summed to construct a joint optimization objective function; The parameters of the multimodal encoder group are iteratively optimized through gradient backpropagation, and the fused representation of the neural state trajectory, the behavior log stream, and the environmental pressure vector in the maritime context-aware latent space is used as the comprehensive contextualized common representation.

[0012] According to one embodiment of this application, the step of calculating metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation mission includes: Based on the predicted value of the behavioral error probability according to the comprehensive contextualized common representation, the prediction uncertainty index of the predicted value of the behavioral error probability is determined. The prediction uncertainty index and the environmental pressure vector are weighted and fused to obtain the metacognitive alignment loss value. If the metacognitive alignment loss value exceeds the loss threshold, a feature weight vector is determined based on the comprehensive contextualized common representation, the environmental pressure vector, and the metacognitive alignment loss value. Based on the feature weight vector, action adjustment is performed from a predefined action space, which includes operation permission downgrade, subtask handover, and cognitive recovery prompt. The feature weight vector is mapped to natural language explanatory text to generate an interpretable intervention instruction, which is used to perform parameterized adjustments to the navigation mission.

[0013] According to one embodiment of this application, before acquiring the physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship, the method further includes: Obtain the aforementioned marine environment context data; The visibility, wave height, and density of nearby vessels in the aforementioned marine environmental context data are quantified to obtain a quantitative sequence of environmental factors. Based on the Pearson correlation coefficient between each environmental factor and the amplitude of physiological signal fluctuations in the crew's historical missions, the individual sensitivity weight of each environmental factor is determined. The environmental factor quantification sequence is summed element-wise with the corresponding individual sensitivity weights to generate the environmental pressure vector.

[0014] Secondly, this application provides a navigation mission adjustment device based on human intelligence function allocation, the device comprising: The acquisition module is used to acquire the physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data. The first processing module is used to perform cross-band dynamic functional connectivity analysis on the physiological signals and extract neural state trajectories. The second processing module is used to perform task semantic parsing on the navigation operation log and extract the behavior log stream corresponding to the sub-task type of the ship. The third processing module is used to construct positive sample pairs and negative sample pairs based on the synchronization status of the neural state trajectory and the behavior log stream. The fourth processing module is used to map the neural state trajectory, the behavior log stream, and the environmental pressure vector to the maritime context-aware latent space with the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, so as to obtain a comprehensive contextualized common representation of the crew. The fifth processing module is used to calculate the metacognitive alignment loss based on the comprehensive contextualized common representation, so as to adjust the crew's navigation mission.

[0015] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the navigation mission adjustment method based on human intelligence function allocation as described in the first aspect above.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application.

[0017] The navigation mission adjustment method, apparatus, and electronic device based on human intelligence function allocation provided in this application have the following advantages over the prior art: (1) By constructing a latent space representation mechanism that couples physiology, behavior and environment, it is possible to deeply perceive the instantaneous interaction between the individual cognitive rhythm of the crew and external pressure to make task adjustment decisions. The adjustment basis is shifted from the feedback of operation results to the prediction of cognitive-behavioral coupling trends, which enhances the dynamic perception accuracy and response agility of the system in complex navigation situations. It effectively solves the problems of adjustment lag, misjudgment or insufficient individual adaptability caused by the shallow perception of situation in traditional methods, and can improve the situation sensitivity of task adjustment.

[0018] (2) By integrating theta-gamma phase-amplitude coupling features with HRV high-frequency power features, a multidimensional neural state trajectory with neurophysiological interpretability was constructed; the dynamic sliding window standardization mechanism enables the feature extraction granularity to adapt to changes in crew cognitive elasticity, avoiding noise interference or loss of details caused by fixed windows; the trajectory accurately depicts the temporal evolution of attention fluctuations and working memory load, improving the perception depth and reliability of task adjustment decisions on crew cognitive state. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is one of the flowcharts illustrating the navigation mission adjustment method based on human intelligence function allocation provided in the embodiments of this application; Figure 2 This is a schematic diagram of the navigation mission adjustment device based on human intelligence function allocation provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] The following description, in conjunction with the accompanying drawings, details the navigation mission adjustment method, navigation mission adjustment device, electronic device, and readable storage medium based on human intelligence function allocation provided in this application, through specific embodiments and application scenarios.

[0022] like Figure 1 As shown, this navigation mission adjustment method based on human intelligence function allocation includes: Step 110: Obtain the physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data.

[0023] Among them, physiological signals are continuous time-series data of the crew's neurophysiological state, including brain electrical activity and autonomic nervous system response signals; and time-series physiological data collected in real time by non-invasive shipboard sensors, including electroencephalogram (EEG) signals and heart rate variability (HRV) signals.

[0024] The navigation operations log is a structured sequence of events stored by the Voyage Data Recorder (VDR) to record crew operations, including rudder command execution records, radar target interaction sequences, and alarm response timestamps.

[0025] The rudder command execution record includes the command content, issuance timestamp, and execution completion timestamp, etc. The radar target interaction sequence includes the target identity (ID), scan time, operator identity, interaction action type, and timestamp, etc. The alarm response timestamp includes the alarm type, trigger time, and first response time, etc.

[0026] Navigational environmental context data is a set of pressure source parameters of the external navigational environment, including visibility and wave height provided by weather instruments, as well as the density of nearby vessels resolved by AIS.

[0027] The environmental pressure vector is a numerical vector generated by encoding the maritime environmental context data through a preset maritime pressure mapping rule. It is used to characterize the comprehensive pressure intensity of the environment on a specific crew member. The preset maritime pressure mapping rule is set according to the environmental factor weight table in the International Maritime Organization (IMO) Human Factors Engineering Guidelines. During encoding, each parameter is first linearly normalized, and then multiplied element by element with the corresponding weight coefficient and summed to generate a three-dimensional environmental pressure vector.

[0028] In step 110, physiological signals such as crew members' EEG and heart rate variability are collected in real time through the shipborne physiological sensor array. Simultaneously, navigation operation logs such as rudder command execution sequence, radar target interaction sequence and alarm response log in the navigation data recorder are acquired. The navigation environment context data such as visibility, wave height and density of nearby ships collected by meteorological instruments, AIS and other equipment are encoded into environmental pressure vectors according to preset maritime pressure mapping rules.

[0029] Based on timestamps, physiological signals, navigation operation logs, and environmental pressure vectors are aligned along a unified time axis to form a ternary synchronized data stream, which is then cached in a memory buffer.

[0030] Step 120: Perform cross-band dynamic functional connectivity analysis on the physiological signals and extract neural state trajectories.

[0031] Among them, cross-band dynamic functional connectivity analysis quantifies the dynamic interaction between different neural oscillation frequency bands through time-frequency transformation and phase-amplitude coupling calculation.

[0032] The neural state trajectory is a temporal feature vector sequence that integrates central nervous activity and peripheral physiological response. It is used to characterize the crew's ability to maintain attention, working memory load level, and the dynamic evolution and fluctuation trend of alertness.

[0033] In step 120, the synchronized physiological signals are segmented based on the key event time anchors in the navigation operation log, and the neuropsychological load segments corresponding to each key operation stage are extracted. Within the physiological signals, the EEG signals are subjected to multi-band bandpass filtering to separate key frequency band signals such as theta and gamma. Hilbert transform is performed on each frequency band signal to extract instantaneous phase and amplitude sequences. The phase is discretized into intervals, the corresponding amplitude mean is calculated, and the cross-band phase-amplitude coupling strength is calculated to generate neural electrical activity features. These features are then fused with heart rate variability frequency domain features to form a high-dimensional temporal feature vector. A cognitive elasticity index is used to dynamically adjust the sliding window length, and the feature sequence is temporally standardized to output the neural state trajectory. For example, the cognitive elasticity index is the HRV to EEG power ratio.

[0034] Step 130: Perform task semantic parsing on the navigation operation log to extract the behavior log stream corresponding to the sub-task type of the ship.

[0035] Among them, the sub-task type is an atomic-level operation unit based on the semantic division of navigation operation events, including collision avoidance decision-making, course maintenance, berthing and unberthing operations, emergency response, etc.

[0036] A behavior log stream is a sequence of behavioral performance metrics organized chronologically and associated with each subtask type.

[0037] In step 130, the event semantic tags in the navigation operation log are parsed to identify the time anchors of key operations such as collision avoidance, turning, berthing and unberthing. An analysis window is set with the time anchor as the center, extending forward by 2 seconds and backward by 3 seconds. All operation records within the window, including rudder command execution records, radar target interaction sequences, and alarm response timestamps, are extracted. The time difference from command issuance to execution completion is calculated for the rudder command execution records to determine the rudder command response delay feature. For the radar target interaction sequences, a trajectory continuity detection algorithm based on the Euclidean distance between the predicted position and the actual observed position using Kalman filtering is employed to statistically analyze the percentage of missed targets as the radar target omission rate feature. The time difference from triggering to the first response is calculated for the alarm response timestamp as the alarm response timeliness feature. The three types of features are grouped according to the sub-task type corresponding to the window and arranged in chronological order to form a behavior log stream.

[0038] Step 140: Construct positive sample pairs and negative sample pairs based on the synchronization status of the neural state trajectory and the behavior log stream.

[0039] Among them, the synchronization state refers to the degree of coupling between the neural state trajectory and the behavior log stream in the time dimension, which is quantified by weighted time-varying mutual information; Positive sample pairs refer to time window segments in which both neural state and behavioral performance are stable and highly synchronous, representing the state of cognitive-behavioral coordination. Negative sample pairs refer to time window segments in which the behavior log stream shows significant abnormalities but the neural state trajectory does not respond synchronously, representing a cognitive-behavioral disconnect.

[0040] In step 140, the product of HRV high-frequency power and EEG alpha-theta power ratio is used as the current cognitive resilience index, and the sampling window length is dynamically set. Based on the sampling window, the neural state trajectory and behavior log stream are temporally segmented. For each segment, the Euclidean distance between the neural state vectors of adjacent windows is calculated as the neural state change rate. The change in the standard score (Z-score) of the abnormal indicators in the behavior log stream is calculated as the behavior abnormal indicator change rate, and the time-varying mutual information value between the neural state change rate and the behavior abnormal indicator change rate is calculated. Combining the preset semantic association weights of the current sub-task type and the environmental pressure vector, the time-varying mutual information value is weighted to generate a weighted synchronization strength. The crew's historical data of the same sub-task type is called, and the probability density distribution of the weighted synchronization strength is constructed using online kernel density estimation. The upper quartile and lower quartile are extracted as dynamic boundary points. Segments with strength higher than the upper quartile are marked as positive sample pairs, and segments with strength lower than the lower quartile are marked as negative sample pairs. The distribution is verified to show a bimodal separation feature.

[0041] Step 150: With the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, the neural state trajectory, the behavior log stream, and the environmental pressure vector are mapped to the maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew.

[0042] Mutual information is the lower bound of the amount of shared information between the latent representation of neural state and the latent representation of behavioral log in a positive sample pair. Distance is the geometric interval between the neural and behavioral latent representations in the latent space of a negative sample pair; Correlation is the statistical strength of the association between the negative sample fusion latent representation and the environmental pressure latent representation; The maritime context-aware latent space is a vector space constructed by a multimodal encoder, whose geometric structure reveals the semantics of the coupled cognition, behavior, and environment. The integrated contextualized common representation is a vector located in the latent space that uniquely represents the current multimodal contextual coupling state of the crew.

[0043] In step 150, the multimodal mapping function is driven to fuse the three-source data into the maritime context perception latent space with the joint optimization objectives of maximizing the mutual information of positive sample pairs, minimizing the latent space distance of negative sample pairs, and maximizing the Spearman correlation coefficient between the latent representation of negative samples and the environmental pressure vector. This generates a comprehensive contextualized common representation that uniquely represents the real-time coupling relationship between the crew's cognitive state, behavioral performance, and environmental pressure. A multimodal encoder group is constructed based on a neural state encoder, a behavior log encoder, an environmental stress encoder, and an independent projector network. The neural state encoder consists of a three-layer fully connected network, the behavior log encoder consists of a two-layer long short-term memory (LSTM) network, the environmental stress encoder consists of a single-layer fully connected layer, and the independent projector network consists of two layers of linear transformation and nonlinear activation function.

[0044] The neural state trajectory is input into the neural state encoder, which outputs a neural state latent representation after passing through a linear transformation layer, a batch normalization layer, a nonlinear activation layer, and another linear transformation layer. The behavior log stream is input into the behavior log encoder, which captures temporal dependencies through a recurrent neural network layer, followed by a temporal pooling layer and a linear transformation layer, and outputs a behavior log latent representation. The environmental stress vector is input into the environmental stress encoder, which outputs an environmental stress latent representation after passing through a single-layer linear transformation. Each latent representation is mapped to a unified-dimensional contrastive learning space through an independent projector network. For the neural and behavioral latent representations of positive sample pairs, a noise contrastive estimation is used to calculate the lower bound of mutual information as the mutual information maximization term. The cosine distance between the latent representations of negative sample pairs is calculated, and a boundary threshold (hinge loss) with a boundary threshold is constructed as the distance minimization term. Spearman correlation analysis is performed on the negative sample fusion latent representation and the environmental stress latent representation, and the term with the maximum negative correlation coefficient is taken as the loss. The three losses are weighted and summed according to preset weights to form a joint objective function. The encoder parameters and projector parameters are iteratively optimized through stochastic gradient descent. The three-mode fusion vector output by the encoder after optimization of the current input is used as a comprehensive contextualized common representation.

[0045] Step 160: Calculate metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation mission.

[0046] Among them, metacognitive alignment loss is used to characterize the degree of deviation between the operational risk predicted by the comprehensive contextualized common representation and the historical safety benchmark; A navigation mission is a set of specific operational responsibilities assigned to the crew, and its adjustment involves parameterized changes to operating permissions, interface parameters, and alarm strategies.

[0047] In step 160, the integrated contextualized common representation is input into the lightweight classifier. The degree of deviation between the predicted operational risk and the historical safety benchmark is used as the predicted value of behavioral error probability and output. Historical safety benchmarks under similar sub-task types and environmental pressure conditions are retrieved from the crew history database. The variance of the predicted value is calculated through multiple random perturbation prediction processes and used as the prediction uncertainty index. The prediction uncertainty index is then multiplied element-wise with the environmental pressure vector and the mean is calculated and weighted to obtain the metacognitive alignment loss value.

[0048] The lightweight classifier is built on a single fully connected layer and the sigmoid function.

[0049] When the loss value exceeds the preset threshold, the system combines the low, medium, and high quantization levels of the environmental pressure vector with the loss value range, queries the SQLite database of the preset strategy rules, generates a feature weight vector based on the comprehensive contextualized common representation, the environmental pressure vector, and the loss value, selects the optimal adjustment action from the predefined action space, and matches it with the hierarchical adjustment instructions in the preset strategy library, such as interface information simplification and dynamic downgrading of operation permissions. The ship's central task allocation module parses the instructions and calls the human-computer interaction system API to perform parameter adjustments, such as the number of interface elements, permission coefficients, and alarm intensity thresholds, to achieve precise intervention in navigation missions.

[0050] Actions in the predefined action space include downgrading of operation permissions, handing over of subtasks, and cognitive recovery prompts.

[0051] According to the navigation mission adjustment method based on human intelligence function allocation provided in the embodiments of this application, by constructing a latent space representation mechanism that couples physiology, behavior, and environment, it can deeply perceive the instantaneous interaction state of individual crew members' cognitive rhythm and external pressure to make mission adjustment decisions. The adjustment basis is shifted from feedback of operation results to prediction of cognitive-behavioral coupling trends, which enhances the system's dynamic perception accuracy and response agility to complex navigation situations. It effectively solves the problems of adjustment lag, misjudgment, or insufficient individual adaptability caused by the shallow situation perception of traditional methods, and can improve the situation sensitivity of mission adjustment.

[0052] In some embodiments, the physiological signals include EEG signals and HRV signals, and the step of performing cross-band dynamic functional connectivity analysis on the physiological signals to extract neural state trajectories includes: The EEG and HRV signals are synchronously time-separated to extract the neuropsychological load segments corresponding to the key stages of navigation operations, thus obtaining EEG segments and HRV segments. The EEG segment is subjected to multi-scale bandpass filtering to separate the theta band phase signal and the gamma band amplitude signal, and the HRV segment is subjected to time-domain sequence extraction to generate the RR interval sequence; Perform a Fourier transform on the RR interval sequence to calculate the HRV high-frequency power characteristics; The theta-band phase signal and the gamma-band amplitude signal are subjected to Hilbert transform to obtain the corresponding instantaneous phase sequence and instantaneous amplitude sequence, respectively. The instantaneous phase sequence is discretized into multiple phase intervals, and the average value of the instantaneous amplitude in each phase interval is calculated to generate a phase-amplitude coupled feature vector. The phase-amplitude coupling feature vector and the HRV high-frequency power feature are tensor-concatenated, and the concatenated high-dimensional feature vector is processed by a time-series sliding window to obtain the neural state trajectory.

[0053] It is understandable that EEG signals refer to the electrophysiological timing signals of the crew's scalp collected through a dry electrode EEG cap, which characterize the electrophysiological activity of the cerebral cortex. HRV signal is an RR interval sequence extracted by the Pan-Tompkins algorithm after being acquired by a photoelectric heart rate sensor, used to reflect cardiac rhythm variability; The neuropsychological load segment is a physiological signal subsequence extracted 2-5 seconds before and after the time anchor point of the key event in the navigation operation; Theta-band phase signal is a low-frequency band related to EEG and attention regulation. Phase extraction yields an instantaneous phase sequence that characterizes the dynamics of neural oscillation phase. The gamma band amplitude signal is a high-frequency band related to EEG and information integration. After amplitude extraction, an instantaneous amplitude sequence representing the dynamic intensity of neural activity is obtained. RR interval sequences are continuous heartbeat interval time series; HRV high-frequency power characteristics are frequency domain quantitative indicators that characterize parasympathetic activity. The power spectral density integral value in the 0.15-0.4Hz frequency band is calculated by Fast Fourier Transform (FFT) using the RR interval sequence, which reflects the parasympathetic regulatory state on cognitive load. The phase-amplitude coupling eigenvector is a cross-frequency coupling characterization of the modulation intensity of low-frequency phase on high-frequency amplitude, representing the modulation intensity of low-frequency phase on high-frequency amplitude during working memory maintenance.

[0054] In actual execution, continuous physiological signal streams are acquired through EEG and HRV acquisition devices synchronized with hardware clocks to obtain EEG and HRV signals. Based on the timestamps of key events such as collision avoidance and berthing marked in the navigation operation log as time anchors, a dynamic time window is set with the timestamp as the center. The window length is preset according to the sub-task type. Continuous EEG and HRV segments of the corresponding time period are synchronously extracted from the raw EEG and HRV signal streams to ensure that the two signals are strictly aligned on the time axis.

[0055] Finite Impulse Response (FIR) digital bandpass filtering was applied to the EEG segments in the theta and gamma bands to separate the theta and gamma band signals, respectively. Time-domain analysis was performed on the HRV segments to extract the continuous RR interval sequence.

[0056] Fast Fourier Transform was performed on the RR interval sequence to calculate the power spectral density integral value of the high-frequency band dominated by the parasympathetic nervous system, which was used as the high-frequency power feature of HRV.

[0057] Perform Hilbert transforms on the theta band signal and the gamma band signal respectively to generate the corresponding instantaneous phase sequence and instantaneous amplitude sequence; The range of values ​​of the instantaneous phase sequence is uniformly discretized into 18 phase intervals. The instantaneous amplitude sequence is traversed, and the amplitude at each time point is assigned to the corresponding phase interval. The statistical mean of the amplitude within the interval is calculated to form an 18-dimensional phase-amplitude coupled feature vector.

[0058] The phase-amplitude coupling feature vector and the HRV high-frequency power feature are concatenated along the feature dimension to form a 19-dimensional high-dimensional feature vector. A sliding window is used to perform Z-score normalization on the concatenated high-dimensional feature vector to output the neural state trajectory. The window length is dynamically adjusted based on the real-time calculated cognitive resilience index.

[0059] For example, the high-frequency power characteristics of HRV are: in, For HRV high-frequency power; Frequency resolution; and , respectively, are the Discrete Fourier Transform (DFT) indices for the maximum and minimum frequencies; j is the DFT index. ; These are the DFT coefficients of the resampled RR sequence; The phase-amplitude coupling eigenvector is: Let m be the probably approximately correct PAC eigenvalue for the m-th phase interval, where m is the phase interval number. n is the sampling point number within the neural segment; Let be the instantaneous amplitude of the gamma band at time t; To fall into The number of sampling points, for example, ; The instantaneous phase of the theta band; This is the m-th phase interval; This represents the total number of phase intervals; for example, a value of 18. This is an indicator function; it is 1 if the condition is true, and 0 otherwise.

[0060] in, This is the concatenated neural state feature vector; They are respectively PAC eigenvector components; for A dimensional real vector space.

[0061] In this embodiment, a multidimensional neural state trajectory with neurophysiological interpretability is constructed by fusing theta-gamma phase-amplitude coupling features and HRV high-frequency power features. The dynamic sliding window standardization mechanism enables the feature extraction granularity to adapt to changes in crew cognitive elasticity, avoiding noise interference or loss of details caused by a fixed window. This trajectory accurately depicts the temporal evolution of attention fluctuations and working memory load, improving the perception depth and reliability of task adjustment decisions on crew cognitive state.

[0062] In some embodiments, the behavior log stream includes steering command response delay features and radar target omission features. The step of performing task semantic parsing on the navigation operation log to extract the behavior log stream corresponding to the ship's sub-task type includes: Identify time anchors for critical operations in the navigation operation event log, including collision avoidance, turning, and berthing / departure. Centered on the aforementioned time anchor point, extend a time window forward and backward respectively, and extract the rudder command execution record, radar target interaction sequence and alarm response timestamp within the corresponding time period; The timing of the instruction issuance time and the instruction execution time in the rudder command execution record are aligned to determine the rudder command response delay characteristics; The radar target interaction sequence is subjected to target trajectory continuity detection to identify time periods of unresponsive targets and determine radar target omission features; The alarm trigger time and alarm response time in the alarm response timestamp are time-aligned to determine the alarm response delay characteristics; The steering command response delay feature, the radar target omission feature, and the alarm response delay feature are grouped according to the sub-task type corresponding to the time window to obtain the behavior log stream.

[0063] It is understandable that critical operations are atomic-level operational units that have a decisive impact on safety during navigation, and they are clearly identified by semantic tags in the navigation operation event log, including collision avoidance decisions, turning operations, berthing and unberthing operations, and emergency response. A time anchor is a marker in the navigation operation event log that has a clearly defined semantic label indicating the start time of an operation, such as the start of a collision avoidance decision.

[0064] The time window is an analysis period that extends forward by 2 seconds and backward by 3 seconds, centered on the time anchor point. The rudder command execution record is the structured operational data stored by the navigation data recorder, which includes the rudder command content, the time of command issuance, the time of command completion, and the execution status indicator; Radar target interaction sequence is the time-series data of interaction events between the crew and the detected target recorded by the radar system. It includes the target's unique identifier, scan timestamp, operator identifier, interaction action type and corresponding timestamp. The alarm response timestamp is a collection of alarm event time records generated by the navigation assistance system, including alarm type identifier, alarm trigger time, and the time of the crew's first manual response; The moment the command is issued is the point in time when the steering command is generated by the operating terminal and sent to the ship's control system. The command execution time is the point in time when the ship's steering gear system confirms that the rudder angle has reached the required command value, based on the timestamp fed back by the actuator status sensor.

[0065] The characteristic of rudder command response delay is the average difference between the timestamp of rudder command issuance and the timestamp of execution completion; Target trajectory continuity detection is based on historical radar target trajectory data. Kalman filtering is used to generate a predicted trajectory. The actual observed position and the predicted position within a continuous scanning cycle are compared with the Euclidean distance. When the deviation continuously exceeds the dynamic tolerance threshold set based on the target motion characteristics and radar accuracy, the trajectory is determined to be interrupted. An unresponsive target is a target that meets the preset high-risk conditions and for which the crew has not performed any interactive actions such as tracking, marking, or evasion within a single radar scan cycle. The radar target omission feature is the proportion of high-risk targets within a preset range and time period that were not interactively operated during the radar scanning cycle; The alarm trigger time is the point in time when the navigation assistance system detects a preset risk condition and generates an audible and visual alarm signal. The alarm response time is the point at which a crew member first manually intervenes in an alarm via the operating terminal. The alarm response delay characteristic is the arithmetic mean of the differences between the alarm response time and the alarm trigger time of all alarm events within the time window.

[0066] In actual execution, the navigation operation event log is analyzed to identify the semantic tags and corresponding time anchors of key operations such as collision avoidance, turning, berthing and unberthing.

[0067] A time window is constructed centered on each time anchor point, and the rudder command execution record, radar target interaction sequence, and alarm response timestamp are extracted within the time window. The rudder command execution record includes the command content, issuance time, and execution completion time; the radar target interaction sequence includes the target ID, scanning time, and interaction action; and the alarm response timestamp includes the alarm type, trigger time, and response time.

[0068] For each rudder command execution record, calculate the time difference between the time the command is issued and the time the execution is completed, and calculate the arithmetic mean of all differences within the window as the rudder command response delay feature.

[0069] Kalman filtering is used to predict the target trajectory of the radar target interaction sequence. If the Euclidean distance between the actual observed position and the predicted position continuously exceeds the preset tolerance threshold, it is determined to be a target omission. The proportion of omission targets to the total number of high-risk targets is counted as the radar target omission feature. Calculate the time difference between the alarm response timestamp and the first response, and take the arithmetic mean of all differences within the window as the alarm response delay feature. The three types of features are logically grouped according to the sub-task type corresponding to the time window, and sorted by timestamp to form a behavior log stream.

[0070] In this embodiment, by strongly associating multi-dimensional behavioral characteristics such as operational timeliness, context awareness integrity, and emergency response capability with sub-task types, task confusion caused by global statistics is avoided; the objectivity and anti-interference capability of radar omission judgment are improved through trajectory continuity detection algorithm; and the behavior performance under specific sub-tasks is accurately anchored through behavior log stream, which can differentiate the evaluation of behavioral anomalies in different task scenarios and enhance the scenario adaptability of task adjustment strategy and the accuracy of behavior basis.

[0071] In some embodiments, after obtaining the behavior log stream and before constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream, the method further includes: The operation trajectory in the rudder command execution record is segmented into a time series to extract the trajectory segment corresponding to the navigation operation event; The trajectory segment is solved in reverse based on the ship dynamics model of the ship to generate an operation intention vector; The cosine similarity between the operational intent vector and the feature vector of the crew member's actual executed actions is used as the intent-execution deviation feature. The intent-execution deviation feature is fused with the steering command response delay feature, the radar target omission feature, and the alarm response delay feature to update the behavior log stream.

[0072] It is understandable that the operating trajectory is a continuous sequence of changes in control parameters such as rudder angle and speed over time; A trajectory segment is a subsequence of operational trajectories aligned with the time of a navigation operation event; The ship dynamics model is a parametric mathematical model built on the six-degree-of-freedom equations of motion of a ship, including the mass matrix, damping matrix, and propeller model; The operational intent vector is an eigenvector generated by inversely solving the expected rudder angle change pattern given the actual motion trajectory in the dynamic model. The actual action feature vector is a vector composed of features such as the rate of change of rudder angle, steering initiation delay, and correction redundancy extracted from the trajectory segment; The intent-execution deviation feature is the quantified deviation value of the cosine similarity between the operational intent vector and the actual executed action feature vector.

[0073] In actual execution, the raw data of the operation trajectory corresponding to the rudder command execution is extracted from the flight data recorder; based on the identified time anchor points, the operation trajectory is divided into trajectory segments corresponding to each sub-task. The system calls upon the ship dynamics model calibrated with the current ship type, load, and sea state. Taking the ship's actual position, heading, speed, and other motion states in the trajectory segment as input, it uses the gradient descent backpropagation algorithm to deduce the operator's expected control input sequence, such as the target rudder angle change curve, and encodes the sequence as an operation intention vector. Features such as actual rudder angle change rate, steering start delay, and correction redundancy are extracted from trajectory segments to form a feature vector of actual executed actions. Calculate the cosine similarity between the intention vector and the actual action feature vector, and convert the similarity into an intention-execution deviation feature; The deviation feature is concatenated along the dimension with the features of steering command response delay, radar target omission, and alarm response delay to update the behavior log stream.

[0074] For the same subtask instance within the same time window, it is confirmed that the extracted rudder command response delay features, radar target omission features, alarm response delay features, and newly calculated intent-execution deviation features strictly correspond in the time dimension. The above four scalar features are combined into a single behavioral feature vector in a preset fixed order: rudder command response delay, radar target omission, alarm response delay, and intent-execution deviation. This expands the behavioral representation dimension of the time window from 3-dimensional to 4-dimensional. While maintaining the original time sequence order and subtask type grouping logic of the behavioral log stream, the behavioral feature vector corresponding to the original time window is replaced with the concatenated enhanced vector. An updated behavioral log stream is generated, in which each time window contains four-dimensional behavioral features of the timeliness of the fused operation results and the consistency of intent-execution, and the grouping structure and time index are completely consistent with those before the update.

[0075] In this embodiment, cognitive-behavioral disconnect is identified through dynamic inversion, and intention-execution deviation features are used as a deep criterion for behavioral abnormalities, thereby improving the detection sensitivity of hidden operational errors. The updated behavior log stream integrates the dual perspectives of ought intention and actual execution, enabling neuro-behavioral synchronous analysis to have causal explanation capabilities, and task adjustments can accurately target the root causes of cognitive deviations.

[0076] In some embodiments, constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream includes: The corresponding environmental pressure weight coefficient is determined based on the ship's navigation sub-task type and the semantic association strength of the environmental pressure vector. Based on the neural state trajectory and the behavior log stream, the time-varying mutual information value between the rate of change of neural state and the rate of change of behavioral abnormality index within each time window is determined, and the time-varying mutual information value is weighted according to the environmental pressure weighting coefficient to generate a weighted synchronization intensity sequence. Based on the navigation sub-task type and the weighted synchronization intensity sequence, online adaptive kernel density estimation is performed on the ship's historical mission data to obtain the probability density distribution curve of the weighted synchronization intensity; The upper quartile and lower quartile are extracted from the probability density distribution curve as dynamic dividing points. Time window segments with weighted synchronization intensity above the upper quartile are marked as positive sample pairs, and time window segments with weighted synchronization intensity below the lower quartile are marked as negative sample pairs. The regions above the upper quartile and the regions below the lower quartile exhibit bimodal separation characteristics in the probability density distribution.

[0077] It is understandable that the semantic association strength of the environmental pressure vector is The environmental pressure weighting coefficient is a weighting coefficient that is dynamically set based on the semantic correlation strength between the navigation sub-task type and environmental factors; The rate of change of neural state is a quantified value of the difference in feature vectors between adjacent windows of the neural state trajectory; The rate of change of abnormal behavior indicators is the time-series change of abnormal indicators in the behavior log stream; The rate of change of anomaly metrics is the inter-window variation of the standardized Z-score values ​​of anomaly metrics in the behavior log stream; The time-varying mutual information value is a dynamic estimate of the degree of sharing between the neural state and behavioral abnormality change rate sequences, estimated using the sliding window k-nearest neighbor method. The weighted synchronization intensity sequence is a sequence obtained by multiplying the time-varying mutual information value by the environmental pressure weighting coefficient; Historical mission data is a collection of the crew member's neural state trajectory fragments, behavior log stream fragments, environmental stress vectors, and corresponding sub-task type labels, all strictly aligned with timestamps, from historical voyages. Online adaptive kernel density estimation is a probability density estimation method based on historical data and dynamically calculated using the Gaussian kernel function; The weighted synchronization strength is a quantitative value of neuro-behavioral synchronicity calculated for a single time window and corrected by the environmental pressure weighting coefficient. Its value is equal to the product of the time-varying mutual information value of that window and the corresponding environmental pressure weighting coefficient. The probability density distribution curve is a continuous probability density function curve generated by the online adaptive kernel density estimation method for the weighted synchronization intensity sequence of the same sub-task type in historical task data. The horizontal axis represents the range of weighted synchronization intensity values, and the vertical axis represents the probability density value. The upper quartile is the weighted synchronization intensity value corresponding to the 75th percentile of the cumulative distribution function in the probability density distribution curve, which represents the lower boundary threshold of the high synchronization region; The lower quartile is the weighted synchronization intensity value corresponding to the cumulative distribution function reaching the 25th percentile in the probability density distribution curve, representing the upper boundary threshold of the low synchronization region; The dynamic dividing line is the upper and lower quartiles calculated in real time based on the current crew's historical mission data. It serves as an adaptive threshold for dividing positive and negative samples, and its value is dynamically updated according to individual cognitive characteristics and mission type. The time window segment is a combination of neural state trajectory subsequences and behavioral log stream subsequences within a fixed time interval centered on the key operation time anchor point, which serves as the basic analysis unit for sample construction.

[0078] The bimodal separation characteristic is a distribution pattern in the probability density distribution curve where the ratio of the valley depth to the peak height between the high synchronization peak and the low synchronization peak is greater than 0.3.

[0079] In actual execution, based on the current subtask type, the preset semantic association weight table is queried to determine the weight coefficients corresponding to each environmental factor and generate environmental pressure weight coefficients. The rule base is constructed based on maritime operation specifications, such as the wave height association strength being higher in collision avoidance tasks than in berthing tasks. The Euclidean distance between the feature vectors of adjacent windows is calculated using the sliding window difference method to generate a sequence of neural state change rates. The change between windows is calculated after normalizing the abnormal indicators in the behavior log stream using the sliding window Z-score method to generate a sequence of behavioral abnormal indicator change rates. The mutual information value between the sequence of neural state change rates and the sequence of behavioral abnormal indicator change rates in each window is calculated using the sliding window k-nearest neighbor method to generate a time-varying mutual information value sequence. The mutual information values ​​in the time-varying mutual information value sequence are multiplied element by element with the environmental pressure weighting coefficient to generate a weighted synchronization intensity sequence. The weighted synchronization intensity sequence from the crew member’s historical data of the same sub-task type is called, and an online adaptive kernel density estimation method is used. The kernel function type and bandwidth parameters are dynamically adjusted according to the data distribution to construct the weighted synchronization intensity probability density distribution curve. The upper and lower quartiles are extracted from the distribution curve as dynamic dividing points. Segments with weighted synchronization intensity higher than the upper quartile are marked as positive sample pairs, and segments with weighted synchronization intensity lower than the lower quartile are marked as negative sample pairs. The valley depth and peak height ratio between the two peaks are calculated. If the ratio is greater than a preset threshold, it is determined to be a bimodal separation, in order to verify whether the distribution curve satisfies the bimodal separation characteristic.

[0080] In this embodiment, the synchronous intensity assessment is dynamically coupled with the sub-task requirements by using the environmental pressure weighting coefficient, thereby improving the contextual adaptability of sample labeling; dynamic boundary points are generated based on adaptive kernel density estimation of individual historical data, avoiding the neglect of individual differences by using fixed thresholds; the bimodal separation feature verification ensures that positive and negative samples are naturally separable in distribution, providing high-quality training samples for latent space mapping; and the ability of samples to represent the cognitive-behavioral coupling state is enhanced.

[0081] In some embodiments, the optimization objective of mapping the neural state trajectory, the behavior log stream, and the environmental pressure vector to a maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew member includes: The outputs of the neural state encoder, behavior log encoder, and environmental stress encoder are mapped to a contrastive learning space of a unified dimension via a projection head network to construct a multimodal encoder group. In the contrastive learning space, the lower bound of mutual information between the neural state hidden representation and the behavior log hidden representation in the positive sample pair is estimated by noise contrast, and a mutual information maximization loss term is constructed. Based on the cosine distance between the negative sample pair and the hidden representation of the neural state and the hidden representation of the behavior log, a distance minimization loss term is constructed; Spearman rank correlation analysis was performed on the fusion latent representation and environmental pressure latent representation of the negative sample pairs to construct a correlation-maximizing loss term; The mutual information maximization loss term, the distance minimization loss term, and the correlation maximization loss term are weighted and summed to construct a joint optimization objective function; The parameters of the multimodal encoder group are iteratively optimized through gradient backpropagation. The fused representation of the neural state trajectory, the behavior log stream, and the environmental pressure vector in the maritime context-aware latent space is used as the comprehensive contextualized common representation.

[0082] It is understandable that the neural state encoder is a three-layer fully connected network, including a batch normalization (BatchNorm, BN) layer and a ReLU function; a parameterized mapping module is used to map neural state trajectories to latent representations, with the input being the temporal vector of the neural state trajectory and the output being the latent representation of the neural state; The behavior log encoder is a parameterized mapping module consisting of a two-layer LSTM followed by a fully connected layer, used to map the behavior log stream to an implicit representation. The input is the timing vector of the behavior log stream, and the output is the behavior log implicit representation. The environmental pressure encoder is a single-layer fully connected network used as a parameterized mapping module to map environmental pressure vectors to implicit representations. The input is the environmental pressure vector, and the output is the environmental pressure implicit representation. The projection head network consists of two fully connected layers attached to each encoder. It is used to map the output of each encoder to a lightweight parameterized mapping module with a unified dimension. The input is the hidden representation of each encoder and the output is a vector with a unified dimension. The contrastive learning space is a vector space for aligning multimodal features; The multimodal encoder group is a set of parameterized mapping modules consisting of a neural state encoder, a behavior log encoder, and an environmental stress encoder. The neural state latent representation is a fixed-dimensional vector output by the neural state encoder after parameterizing the neural state trajectory, representing the abstract features of the crew's neural electrophysiological activities in the latent space.

[0083] The behavior log latent representation is a fixed-dimensional vector output by the behavior log encoder after parameterizing the behavior log stream, representing the abstract features of crew operation behavior patterns in the latent space.

[0084] The mutual information lower bound is the lower bound of the amount of shared information between positive sample pairs estimated by contrastive learning (InfoNCE) loss. Noise contrastive estimation treats positive sample pairs as positive examples and other sample pairs in the same batch as negative examples. It calculates the cosine similarity between positive and negative examples in the contrastive learning space, and after softmax normalization, takes the negative logarithm of the positive example probability as the loss. The negative value of this loss is the lower bound estimate of mutual information.

[0085] The mutual information maximization loss term is a loss function term calculated based on noise contrast estimation. Its value is equal to the negative of the lower bound estimate of the mutual information of positive samples. During optimization, minimizing this loss term strengthens the semantic consistency of positive samples with the neural-behavioral latent representation.

[0086] The distance minimization loss term is a constraint term on the distance between negative samples and the latent representations; The fusion latent representation is a single vector generated by concatenating the neural state latent representation and the behavior log latent representation of the negative sample pair as vectors and then performing a linear transformation. It represents the overall latent state of the negative sample pair.

[0087] The environmental pressure latent representation is a fixed-dimensional vector output by the environmental pressure encoder after parameterizing and mapping the environmental pressure vector. It represents the abstract characteristics of external environmental pressure in the encoder's hidden layer and reflects the modulation effect of environmental pressure on the cognitive-behavioral coupling state.

[0088] Spearman rank correlation analysis involves sorting the fusion implicit representation and the environmental pressure implicit representation in ascending order according to the values ​​of each dimension to generate a rank sequence, and then calculating the Pearson correlation coefficient between the two rank sequences. The resulting coefficient is the Spearman rank correlation coefficient, which characterizes the monotonic correlation strength between the two vectors.

[0089] The correlation maximization loss term is a loss function term constructed with the negative value of the Spearman rank correlation coefficient. During optimization, minimizing this loss term makes the negative sample fusion latent representation and the environmental pressure latent representation present a strong statistical correlation in the latent space. The joint optimization objective function is a weighted combination of multi-objective losses; Gradient backpropagation is a deterministic optimization process that, in the computation graph of the parameterized mapping module, starts from the joint optimization objective function according to the chain rule, calculates the partial derivatives of each parameter with respect to the loss layer by layer, and propagates the gradient backward along the computation path to drive the iterative update of parameters.

[0090] The maritime context perception latent space is a vector space formed after training with a joint optimization objective function, which can uniquely encode the coupled semantic relationship between crew members' cognitive state, behavioral performance, and environmental pressure.

[0091] The fusion representation is a single vector generated in the maritime context perception latent space by vector concatenation and linear transformation after the current input neural state trajectory, behavior log stream, and environmental pressure vector are respectively mapped by the optimized multimodal encoder group and the projection head network, and then used as a comprehensive contextualized common representation.

[0092] In actual implementation, a multimodal encoder group is constructed based on the neural state encoder, behavior log encoder, environmental stress encoder and projection head network. For the neural state encoder, the neural state trajectory is input, the first linear transformation layer outputs the intermediate dimension, which is then passed through the batch normalization layer, the nonlinear activation function layer, and the second linear transformation layer to output the preset hidden dimension, thus outputting the neural state hidden representation. For the behavior log encoder, the input behavior log stream is captured by the recurrent neural network layer to capture the temporal dependencies, and the output is the hidden state sequence. The temporal pooling layer takes the final hidden state, and the linear transformation layer outputs the preset hidden dimension, outputting the behavior log hidden representation. For an environmental pressure encoder, the input environmental pressure vector is transformed by a single-layer linear transformation to output a preset hidden dimension, and the output is an implicit representation of the environmental pressure. The neural state hidden representation, behavior log hidden representation, and environmental stress hidden representation are respectively input into the corresponding projection head network, and the three are mapped to a contrastive learning space of the same dimension. Each projection head consists of two linear transformation layers with nonlinear activation in between. In the contrastive learning space, the neural state hidden representation and behavior log hidden representation of positive sample pairs are estimated using a noisy contrastive estimation method. By constructing positive and negative sample pairs, the softmax normalized similarity is calculated to obtain the lower bound of mutual information. The InfoNCE loss is calculated for positive sample pairs as the mutual information maximization loss term. Calculate the cosine distance between the neural state hidden representation and the behavior log hidden representation in the negative sample pair, and construct a set boundary threshold (hinge loss) as the distance minimization loss term; The neural and behavioral latent representations of negative sample pairs are concatenated into a fused latent representation. The Spearman rank correlation coefficient between the fused latent representation and the environmental stress latent representation is calculated, and the negative value is taken as the correlation maximization loss term. The three losses are weighted and summed according to preset weight coefficients to form a joint optimization objective function; all parameters of the multimodal encoder group are iteratively optimized through gradient backpropagation of the Adam optimizer; the current input neural state trajectory, behavior log stream and environmental pressure vector are forward propagated through the optimized encoder group and the projector head to output their fusion vector in the maritime context perception latent space as a comprehensive contextualized common representation.

[0093] In this embodiment, the semantic consistency of positive sample pairs to neural-behavioral representations is enhanced by the mutual information maximization loss term, the distance minimization loss term explicitly increases the distance between negative sample pairs to representations, and the correlation maximization loss term associates environmental pressure with the latent representations of negative samples. The triple optimization objective enables the latent space of maritime context perception to simultaneously possess intra-class compactness, inter-class separability, and environmental sensitivity. The generated comprehensive contextualized common representation not only encodes the current multimodal state of the crew members, but also implicitly contains the rules of how environmental pressure modulates cognitive-behavioral coupling.

[0094] In some embodiments, calculating the metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation mission includes: Based on the predicted value of the behavioral error probability according to the comprehensive contextualized common representation, the prediction uncertainty index of the predicted value of the behavioral error probability is determined. The prediction uncertainty index and the environmental pressure vector are weighted and fused to obtain the metacognitive alignment loss value. When the metacognitive alignment loss value exceeds the loss threshold, a feature weight vector is calculated based on the comprehensive contextualized common representation, the environmental pressure vector, and the metacognitive alignment loss value. Based on the feature weight vector, the action is adjusted from a predefined action space, which includes operation permission downgrade, subtask handover, and cognitive recovery prompt. The feature weight vector is mapped to natural language explanatory text to generate an interpretable intervention instruction, which is used to perform parameterized adjustments to the navigation mission.

[0095] Understandably, the predicted probability of behavioral errors is based on the predicted operational risk probability output by a single-layer fully connected layer and a sigmoid function, using a comprehensive contextualized common representation. The prediction uncertainty index is a quantitative representation of the degree of fluctuation in the prediction results. It is determined by the variance of the predicted values ​​obtained by performing 10 Monte Carlo Dropout samplings on the comprehensive contextualized common representation.

[0096] Metacognitive alignment loss is a quantified risk value resulting from the coupling of prediction uncertainty and environmental pressure. The loss threshold is a preset boundary for judging metacognitive alignment loss values, which is dynamically set based on the statistical distribution of metacognitive alignment loss values ​​in the crew's historical safety operation data. The feature weight vector is calculated through a differentiable attention mechanism, the query vector, the joint feature dot product, and the softmax operation, to represent the dimensional importance weights and characterize the weight distribution of the contribution of each contextual element to the decision. The predefined action space is a set of adjustment actions that the system can execute, including downgrading operation permissions, handing over subtasks, and cognitive recovery prompts; Degrading operational authority involves changing the crew member's operational control authority over the current navigation sub-task from master control mode to auxiliary or monitoring mode. Subtask handover is the process of completely transferring a specific navigation subtask currently assigned to the crew to a preset standby operating terminal or a designated standby crew member; Cognitive recovery prompts are non-invasive intervention information pushed by the system to crew members to promote cognitive recovery. The type of prompt content, presentation duration, and interaction intensity are dynamically matched based on the current metacognitive alignment loss value and the level of environmental stress. Natural language interpretation text is a text description that conforms to the maritime operation terminology standard and is generated by matching semantic tags extracted from the salient dimension of the feature weight vector with a preset interpretation template library. Explainable intervention instructions are structured instructions consisting of three parts: optimal adjustment action identifier, parameter adjustment amount, and natural language explanation text. They are used to make numerical modifications to interface parameters, permission coefficients, alarm intensity, etc.

[0097] In actual implementation, the comprehensive contextualized common representation is input into the lightweight classifier, and the predicted value of the behavior error probability is output through a single-layer linear transformation and sigmoid activation. Through multiple random perturbation prediction processes, for example, applying a random mask to the intermediate layer of the prediction module, the variance of the predicted value is calculated as the prediction uncertainty index. The metacognitive alignment loss value is obtained by multiplying the prediction uncertainty index element by element of each dimension of the environmental pressure vector and then taking the mean value. When the metacognitive alignment loss value exceeds the preset threshold, the comprehensive contextualized common representation, environmental pressure vector and loss value are concatenated into a joint feature vector, which is input to the feature weight calculation module composed of a linear transformation layer and a softmax layer to generate the weight distribution of each dimension as the feature weight vector; based on the feature weight vector, the preset weights of each action in the predefined action space are weighted and summed, and the action with the highest weight score is selected as the optimal adjustment action; Dimensions with significantly higher weights than the mean in the feature weight vector are extracted and matched with a preset interpretation template library. The preset templates contain dimension semantic labels and natural language fragments to generate natural language interpretation text. The optimal adjustment action and interpretation text are combined to form an interpretable intervention instruction. The instruction is parsed by the ship's central task allocation module and the human-computer interaction system application programming interface (API) is called to perform parameterized adjustments, such as the number of interface elements, operation permission coefficients, alarm intensity thresholds, etc.

[0098] Metacognitive alignment loss is: in, Let be the scalar loss value of the metacognitive alignment loss at time t; t is the current system sampling time. For time t, Monte Carlo Dropout outputs the variance on NS(t) to predict uncertainty; This is an indicator function for the increase in uncertainty within the time interval {0,1} from t-1 to t, used to capture the trend of risk deterioration. 1 = the current uncertainty is higher than the previous moment; 0.2 is the amplification factor for the increase. Let t be the environmental pressure intensity at time t; The sub-task types for time t include collision avoidance, berthing, channel holding, and narrow waterway. It is a set of high-risk tasks, including collision avoidance and narrow waterways; The high-risk task indicator is {0,1}, where 1 indicates that the current task is high-risk and triggers enhanced monitoring; 0.3 is the high-risk enhancement coefficient. τ(t) represents the historical over-threshold ratio at time t, which is the proportion of times in the past Δt when the loss exceeds τ(t); 0.1 is the historical feedback suppression coefficient, which is greater than 0.15 and causes response hysteresis, and less than 0.05 and introduces noise. Historical over-threshold ratio at time t for: Where k is the summation variable, located within the window [t-Δt, t-1], and Δt is the length of the history window; Let be the scalar loss value of the metacognitive alignment loss at time k; The threshold is set dynamically based on the individual crew member's historical data. As an indicator function for risk anomalies, in In this case, the risk at time t is significantly higher than the individual's historical normal level, and is included in the cumulative risk statistics. In this case, the risk at time t is within the normal fluctuation range of an individual and can be ignored; The dynamic thresholds are as follows: in, The historical average loss at time t, within the window [t-Δt, t-1]. The arithmetic mean; Let be the historical loss standard deviation at time t.

[0099] In this embodiment, by weighted fusion of prediction uncertainty and environmental pressure, high uncertainty combined with high pressure increases the loss value and triggers priority intervention. Through action selection driven by feature weight vector and generation of natural language interpretation, the decision-making process becomes transparent and traceable, improving crew members' trust and acceptance of the system's suggestions and avoiding over-intervention or under-intervention.

[0100] In some embodiments, before acquiring the physiological signals of the crew members, navigation operation logs, and the environmental pressure vector of the ship, the method further includes: Obtain the aforementioned marine environment context data; The visibility, wave height, and density of nearby vessels in the aforementioned marine environmental context data are quantified to obtain a quantitative sequence of environmental factors. Based on the Pearson correlation coefficient between each environmental factor and the amplitude of physiological signal fluctuations in the crew's historical missions, the individual sensitivity weight of each environmental factor is determined. The environmental factor quantification sequence is summed element-wise with the corresponding individual sensitivity weights to generate the environmental pressure vector.

[0101] Understandably, visibility is the maximum horizontal distance at which a crew member can clearly identify a target ahead in a navigational environment; Wave height refers to the height of the waves on the sea surface. The density of nearby vessels is the number of other vessels per unit area within a circular monitoring area centered on the vessel and with a radius of 3 nautical miles. The environmental factor quantification sequence is a numerical sequence of independent parameter variables in marine environmental context data that characterize external environmental pressure sources after linear normalization. Environmental factors are independent parametric variables used in marine environmental context data to characterize external environmental stress sources; The amplitude of physiological signal fluctuations is the standard deviation sequence of the HRV high-frequency power characteristic sequence calculated by sliding window, which characterizes the fluctuation intensity of crew members' autonomic nervous activity in a short time scale; The Pearson correlation coefficient between environmental factors and physiological signal fluctuation amplitude is a linear correlation measure calculated for the same historical task segment after strictly aligning the quantified sequence of environmental factors with the sequence of physiological signal fluctuation amplitude in time.

[0102] Individual sensitivity weights are weighting coefficients calculated based on crew historical data, which characterize the intensity of the influence of environmental factors on the physiological response of a specific crew member.

[0103] In actual implementation, real-time maritime environmental context data, including visibility, wave height and density of nearby vessels, is acquired, and each parameter is linearly normalized to generate a quantitative sequence of environmental factors. The crew's historical mission database was retrieved, and historical segments with normalized environmental factor values ​​that were similar to the current environmental conditions with differences less than the preset tolerance were selected. The Pearson correlation coefficient between the environmental factor sequence and the physiological signal fluctuation amplitude sequence in each historical segment was calculated. The absolute value of the correlation coefficient of the same environmental factor in all historical segments is taken and the mean is calculated as the individual sensitivity weight of the environmental factor. The current environmental factor quantification sequence is multiplied element by element by the corresponding individual sensitivity weight and then summed to generate an environmental pressure vector that retains each weighted component.

[0104] For example, adaptive phase binning and HRV frequency domain feature tensor fusion: Environmental pressure vector calculation: in, The output value represents the environmental pressure intensity at time t. The larger the output value, the stronger the pressure on the crew's cognitive load under the current maritime situation. This is the timestamp; i is the environmental factor index, where i=1 represents visibility, i=2 represents wave height, and i=3 represents the density of nearby vessels. Let be the quantified observation value of the i-th environmental factor at time t; Let i be the individual sensitivity weight of the crew member to the i-th environmental factor index; The high-risk threshold for the i-th environmental factor index; The extreme threshold for the index of the i-th environmental factor; This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. This is a counter for over-threshold factors, which activates enhancement when two factors simultaneously exceed the threshold. , is the ReLU truncation operator; This is used to determine the existence of extreme events; it is triggered when any factor exceeds the extreme threshold.

[0105] In this embodiment, the actual pressure differences of the same environment on different crew members are accurately quantified by individual sensitivity weights, so that the environmental pressure vector has the ability to be individually calibrated. This enables proactive adjustment strategies to be triggered in advance for sensitive crew members under high pressure, while avoiding unnecessary intervention for tolerant crew members. The weight calculation based on historical data ensures that the method is objective and feasible, and improves the individual adaptation accuracy and humanization level of task adjustment.

[0106] The navigation mission adjustment method based on human intelligence function allocation provided in this application can be executed by a navigation mission adjustment device based on human intelligence function allocation. This application uses the example of a navigation mission adjustment device based on human intelligence function allocation executing the navigation mission adjustment method to illustrate the navigation mission adjustment device based on human intelligence function allocation provided in this application.

[0107] This application also provides a navigation mission adjustment device based on human intelligence function allocation.

[0108] like Figure 2 As shown, the navigation mission adjustment device based on human intelligence function allocation includes: The acquisition module 210 is used to acquire the physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data. The first processing module 220 is used to perform cross-band dynamic functional connectivity analysis on the physiological signal and extract neural state trajectories. The second processing module 230 is used to perform task semantic parsing on the navigation operation log and extract the behavior log stream corresponding to the sub-task type of the ship. The third processing module 240 is used to construct positive sample pairs and negative sample pairs based on the synchronization status of the neural state trajectory and the behavior log stream; The fourth processing module 250 is used to map the neural state trajectory, the behavior log stream, and the environmental pressure vector to the maritime context-aware latent space with the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, so as to obtain a comprehensive contextualized common representation of the crew member. The fifth processing module 260 is used to calculate the metacognitive alignment loss based on the comprehensive contextualized common representation in order to adjust the crew's navigation mission.

[0109] According to the navigation mission adjustment device based on human intelligence function allocation provided in the embodiments of this application, by constructing a latent space representation mechanism that couples physiology, behavior, and environment, it can deeply perceive the instantaneous interaction state of individual crew members' cognitive rhythm and external pressure to make mission adjustment decisions. The adjustment basis is shifted from feedback of operation results to prediction of cognitive-behavioral coupling trends, which enhances the system's dynamic perception accuracy and response agility to complex navigation situations. It effectively solves the problems of adjustment lag, misjudgment, or insufficient individual adaptability caused by the shallow perception of situation in traditional methods, and can improve the situational sensitivity of mission adjustment.

[0110] The navigation mission adjustment device based on human intelligence function allocation in this application embodiment can be an electronic device or a component of an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a laptop computer, a mobile internet device (MID), an ultra-mobile personal computer (UMPC), a server, network attached storage (NAS), a personal computer (PC), etc., and this application embodiment does not specifically limit it.

[0111] The navigation mission adjustment device based on human intelligence function allocation in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0112] The navigation mission adjustment device based on human intelligence function allocation provided in this application embodiment can realize the various processes implemented in the navigation mission adjustment method embodiment based on human intelligence function allocation as described above. To avoid repetition, it will not be described again here.

[0113] In some embodiments, such as Figure 3As shown, this application embodiment also provides an electronic device 300, including a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301. When the program is executed by the processor 301, it implements the various processes of the above-described embodiment of the navigation mission adjustment method based on human intelligence function allocation and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0114] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0115] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0116] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for adjusting navigation missions based on the allocation of human intelligence functions, characterized in that, include: The physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship are obtained. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data. Cross-band dynamic functional connectivity analysis was performed on the physiological signals to extract neural state trajectories; Perform task semantic parsing on the navigation operation logs to extract the behavior log streams corresponding to the sub-task types of the ship; Based on the synchronization status of the neural state trajectory and the behavior log stream, construct positive sample pairs and negative sample pairs; With the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, the neural state trajectory, the behavior log stream, and the environmental pressure vector are mapped to the maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew. Metacognitive alignment loss is calculated based on the comprehensive contextualized common representation to adjust the crew's navigation mission.

2. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, The physiological signals include EEG signals and HRV signals. The cross-band dynamic functional connectivity analysis of the physiological signals to extract neural state trajectories includes: The EEG and HRV signals are synchronously time-separated to extract the neuropsychological load segments corresponding to the key stages of navigation operations, thus obtaining EEG segments and HRV segments. The EEG segment is subjected to multi-scale bandpass filtering to separate the theta band phase signal and the gamma band amplitude signal, and the HRV segment is subjected to time-domain sequence extraction to generate the RR interval sequence; Perform a Fourier transform on the RR interval sequence to calculate the HRV high-frequency power characteristics; The theta-band phase signal and the gamma-band amplitude signal are subjected to Hilbert transform to obtain the corresponding instantaneous phase sequence and instantaneous amplitude sequence, respectively. The instantaneous phase sequence is discretized into multiple phase intervals, and the average value of the instantaneous amplitude in each phase interval is calculated to generate a phase-amplitude coupled feature vector. The phase-amplitude coupling feature vector and the HRV high-frequency power feature are tensor-concatenated, and the concatenated high-dimensional feature vector is processed by a time-series sliding window to obtain the neural state trajectory.

3. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, The behavior log stream includes steering command response delay features and radar target omission features. The step of performing task semantic parsing on the navigation operation log to extract the behavior log stream corresponding to the ship's sub-task type includes: Identify time anchors for critical operations in the navigation operation event log, including collision avoidance, turning, and berthing / departure. Centered on the aforementioned time anchor point, extend a time window forward and backward respectively, and extract the rudder command execution record, radar target interaction sequence and alarm response timestamp within the corresponding time period; The timing of the instruction issuance time and the instruction execution time in the rudder command execution record are aligned to determine the rudder command response delay characteristics; The radar target interaction sequence is subjected to target trajectory continuity detection to identify time periods of unresponsive targets and determine radar target omission features; The alarm trigger time and alarm response time in the alarm response timestamp are time-aligned to determine the alarm response delay characteristics; The steering command response delay feature, the radar target omission feature, and the alarm response delay feature are grouped according to the sub-task type corresponding to the time window to obtain the behavior log stream.

4. The navigation mission adjustment method based on human intelligence function allocation according to claim 3, characterized in that, After obtaining the behavior log stream, and before constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream, the method further includes: The operation trajectory in the rudder command execution record is segmented into a time series to extract the trajectory segment corresponding to the navigation operation event; The trajectory segment is solved in reverse based on the ship dynamics model of the ship to generate an operation intention vector; The cosine similarity between the operational intent vector and the feature vector of the crew member's actual executed actions is used as the intent-execution deviation feature. The intent-execution deviation feature is fused with the steering command response delay feature, the radar target omission feature, and the alarm response delay feature to update the behavior log stream.

5. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, The step of constructing positive and negative sample pairs based on the synchronization state of the neural state trajectory and the behavior log stream includes: The corresponding environmental pressure weight coefficient is determined based on the ship's navigation sub-task type and the semantic association strength of the environmental pressure vector. Based on the neural state trajectory and the behavior log stream, the time-varying mutual information value between the rate of change of neural state and the rate of change of behavioral abnormality index within each time window is determined, and the time-varying mutual information value is weighted according to the environmental pressure weighting coefficient to generate a weighted synchronization intensity sequence. Based on the navigation sub-task type and the weighted synchronization intensity sequence, online adaptive kernel density estimation is performed on the ship's historical mission data to obtain the probability density distribution curve of the weighted synchronization intensity; The upper quartile and lower quartile are extracted from the probability density distribution curve as dynamic dividing points. Time window segments with weighted synchronization intensity above the upper quartile are marked as positive sample pairs, and time window segments with weighted synchronization intensity below the lower quartile are marked as negative sample pairs. The regions above the upper quartile and the regions below the lower quartile exhibit bimodal separation characteristics in the probability density distribution.

6. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, The optimization objective is to maximize the mutual information of the positive sample pairs, minimize the distance of the negative sample pairs, and maximize the correlation between the negative sample pairs and the environmental pressure vector. The neural state trajectory, the behavior log stream, and the environmental pressure vector are mapped to the maritime context-aware latent space to obtain a comprehensive contextualized common representation of the crew member, including: The outputs of the neural state encoder, behavior log encoder, and environmental stress encoder are mapped to a contrastive learning space of a unified dimension via a projection head network to construct a multimodal encoder group. In the contrastive learning space, the lower bound of mutual information between the neural state hidden representation and the behavior log hidden representation in the positive sample pair is estimated by noise contrast, and a mutual information maximization loss term is constructed. Based on the cosine distance between the negative sample pair and the hidden representation of the neural state and the hidden representation of the behavior log, a distance minimization loss term is constructed; Spearman rank correlation analysis was performed on the fusion latent representation and environmental pressure latent representation of the negative sample pairs to construct a correlation-maximizing loss term; The mutual information maximization loss term, the distance minimization loss term, and the correlation maximization loss term are weighted and summed to construct a joint optimization objective function; The parameters of the multimodal encoder group are iteratively optimized through gradient backpropagation, and the fused representation of the neural state trajectory, the behavior log stream, and the environmental pressure vector in the maritime context-aware latent space is used as the comprehensive contextualized common representation.

7. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, The step of calculating metacognitive alignment loss based on the comprehensive contextualized common representation to adjust the crew's navigation mission includes: Based on the predicted value of the behavioral error probability according to the comprehensive contextualized common representation, the prediction uncertainty index of the predicted value of the behavioral error probability is determined. The prediction uncertainty index and the environmental pressure vector are weighted and fused to obtain the metacognitive alignment loss value. If the metacognitive alignment loss value exceeds the loss threshold, a feature weight vector is determined based on the comprehensive contextualized common representation, the environmental pressure vector, and the metacognitive alignment loss value. Based on the feature weight vector, action adjustment is performed from a predefined action space, which includes operation permission downgrade, subtask handover, and cognitive recovery prompt. The feature weight vector is mapped to natural language explanatory text to generate an interpretable intervention instruction, which is used to perform parameterized adjustments to the navigation mission.

8. The navigation mission adjustment method based on human intelligence function allocation according to claim 1, characterized in that, Before acquiring the physiological signals of the crew members, navigation operation logs, and the environmental pressure vector of the ship, the method further includes: Obtain the aforementioned marine environment context data; The visibility, wave height, and density of nearby vessels in the aforementioned marine environmental context data are quantified to obtain a quantitative sequence of environmental factors. Based on the Pearson correlation coefficient between each environmental factor and the amplitude of physiological signal fluctuations in the crew's historical missions, the individual sensitivity weight of each environmental factor is determined. The environmental factor quantification sequence is summed element-wise with the corresponding individual sensitivity weights to generate the environmental pressure vector.

9. A navigation mission adjustment device based on human intelligence function allocation, characterized in that, include: The acquisition module is used to acquire the physiological signals of the crew members on the ship, the navigation operation log, and the environmental pressure vector of the ship. The environmental pressure vector is obtained by encoding the ship's maritime environmental context data. The first processing module is used to perform cross-band dynamic functional connectivity analysis on the physiological signals and extract neural state trajectories. The second processing module is used to perform task semantic parsing on the navigation operation log and extract the behavior log stream corresponding to the sub-task type of the ship. The third processing module is used to construct positive sample pairs and negative sample pairs based on the synchronization status of the neural state trajectory and the behavior log stream. The fourth processing module is used to map the neural state trajectory, the behavior log stream, and the environmental pressure vector to the maritime context-aware latent space with the optimization objectives of maximizing the mutual information of the positive sample pairs, minimizing the distance of the negative sample pairs, and maximizing the correlation between the negative sample pairs and the environmental pressure vector, so as to obtain a comprehensive contextualized common representation of the crew. The fifth processing module is used to calculate the metacognitive alignment loss based on the comprehensive contextualized common representation, so as to adjust the crew's navigation mission.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the navigation mission adjustment method based on human intelligence function allocation as described in any one of claims 1-8.

Images