Flight training assessment system integrating EEG characteristics and physiological indicators

Abstract

This invention relates to the field of flight training assessment technology and discloses a flight training assessment system that integrates electroencephalogram (EEG) characteristics and physiological indicators. The system includes a physiological signal acquisition module that simultaneously captures multi-channel raw EEG signals and surface physiological indicator data, including ECG R-R interval sequences, respiratory wave frequency and amplitude, and skin conductance response amplitude; a multimodal fusion module that analyzes the ECG R-R interval sequences to generate a heart rate variability feature vector and establishes a dynamic correlation mapping between EEG entropy values ​​and physiological feature vectors; a cognitive state modeling module that generates a cognitive load index based on the dynamic correlation mapping and constructs a cognitive stability quantification matrix; an adaptive feedback module that receives relevant data and dynamically adjusts simulated flight scenario parameters; and an assessment output module that integrates data to generate a comprehensive training assessment report including neurophysiological coordination scores and operational accuracy ratings. This system enables comprehensive assessment and dynamic training adjustment of the pilot's cognitive state.

Description

Technical Field

[0001] This invention relates to the field of flight training assessment technology, specifically a flight training assessment system that integrates electroencephalogram (EEG) characteristics and physiological indicators. Background Technology

[0002] In the field of flight training, the evaluation of pilot training effectiveness has long relied on operational trajectory data and subjective scoring, a method with significant limitations. Traditional evaluation systems focus primarily on the pilot's operational precision, such as flight path deviation and operational response time, but fail to deeply reflect their internal cognitive state. For example, during flight missions simulating complex weather conditions, pilots may experience cognitive fatigue due to sustained high concentration, but operational data alone often cannot capture this change in a timely manner, leading to biased evaluation results.

[0003] In existing technologies, some assessment systems attempt to incorporate single physiological indicators, such as heart rate or skin conductance. However, when used in isolation, these indicators are insufficient to comprehensively map a pilot's cognitive processes. Heart rate changes may be caused by a combination of physiological stress and increased cognitive load; without the aid of other indicators, it is easy to misjudge their cognitive state. Meanwhile, although electroencephalogram (EEG) signals can reflect brain activity, current processing methods are mostly limited to single-band analysis, ignoring the dynamic correlation between signals from different frequency bands and physiological indicators, thus failing to construct a complete model of cognitive state.

[0004] Traditional assessment systems often employ static feedback mechanisms, providing fixed feedback based on preset thresholds, which fails to adapt to the needs of different flight phases. Pilots' cognitive load requirements differ significantly between takeoff and cruise phases, making it difficult for static feedback to allow for targeted training adjustments and hindering training effectiveness. These shortcomings mean that existing assessment systems have room for improvement in terms of accuracy, comprehensiveness, and dynamic adaptability. Summary of the Invention

[0005] The purpose of this invention is to provide a flight training assessment system that integrates electroencephalogram (EEG) characteristics and physiological indicators to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a flight training assessment system that integrates electroencephalogram (EEG) characteristics and physiological indicators, the system comprising:

[0007] The physiological signal acquisition module is used to simultaneously capture multi-channel raw EEG signals and surface physiological index data of the pilot during simulated flight missions. The surface physiological index data includes at least the ECG RR interval sequence, respiratory wave frequency amplitude, and skin conductance response amplitude.

[0008] The multimodal fusion module is used to perform time-frequency domain decomposition processing on the raw EEG signal to extract energy entropy values ​​of different frequency bands, and at the same time, analyze the ECG RR interval sequence to generate heart rate variability feature vectors, and establish a dynamic correlation mapping between EEG entropy values ​​and physiological feature vectors through a cross-modal attention mechanism.

[0009] The cognitive state modeling module is used to generate a cognitive load index reflecting the pilot's attention concentration based on the dynamic correlation mapping, and to construct a cognitive stability quantification matrix by combining the respiratory wave frequency amplitude fluctuation rate.

[0010] An adaptive feedback module is used to receive the cognitive load index and cognitive stability quantification matrix, and dynamically adjust the visual stimulus intensity and joystick force feedback coefficient of the simulated flight scenario according to preset key operation nodes of the flight phase.

[0011] The evaluation output module integrates cognitive load index, cognitive stability quantification matrix and simulated flight operation trajectory data to generate a comprehensive training evaluation report that includes neurophysiological coordination score and operation accuracy rating.

[0012] Preferably, the physiological signal acquisition module includes an EEG signal preprocessing unit, which performs power frequency noise filtering and eye movement artifact correction on the multi-channel raw EEG signals, and uses independent component analysis to separate the effective EEG components;

[0013] The physiological signal acquisition module also includes a physiological signal synchronization unit, which binds the corrected EEG components, ECG RR interval sequences, and skin conductance response amplitudes to the same time coordinate system through a timestamp alignment mechanism, forming a time-domain aligned neurophysiological synchronization dataset.

[0014] Preferably, the multimodal fusion module includes a frequency domain decomposition engine, which performs wavelet packet transform on the EEG components to divide them into Delta, Theta, Alpha, and Beta frequency bands, and calculates the energy entropy value sequence of each frequency band.

[0015] The multimodal fusion module also includes a physiological feature analysis unit, which analyzes the sample entropy and fractal dimension of the ECG RR interval sequence through a nonlinear dynamics algorithm to generate a heart rate variability feature vector.

[0016] Preferably, the cognitive state modeling module includes a cognitive load index generation unit, which receives a frequency band energy entropy value sequence and a heart rate variability feature vector, uses a convolutional neural network model to learn the nonlinear relationship between the energy entropy value and the heart rate variability feature vector, and outputs a cognitive load index curve that changes with flight mission time.

[0017] The cognitive state modeling module also includes a stability quantification unit, which integrates the respiratory wave frequency amplitude fluctuation rate and the number of skin conductance response amplitude mutations to construct a cognitive stability quantification matrix that reflects the autonomic nervous system activation state.

[0018] Preferably, the adaptive feedback module includes a scene control decision-maker, which triggers visual stimulus intensity adjustment commands and joystick force feedback coefficient adjustment commands based on the numerical threshold of the cognitive load index curve at key operation nodes in the preset flight phase.

[0019] The scene control decision-maker also receives data on the number of abrupt changes in the skin conductance response amplitude from the cognitive stability quantification matrix. When the number of abrupt changes exceeds the baseline threshold, the damping coefficient of the joystick force feedback coefficient is enhanced.

[0020] Preferably, the evaluation output module includes a neurophysiological coordination calculation unit, which performs normalized weighted fusion of the cognitive load index curve and the cognitive stability quantification matrix to generate a neurophysiological coordination score that reflects the degree of coordination between EEG features and physiological indicators.

[0021] The evaluation output module also includes an operation accuracy analysis unit, which performs dynamic time warping matching between simulated flight operation trajectory data and standard flight trajectory templates, calculates the mean heading angle deviation and altitude control error rate, and outputs an operation accuracy rating.

[0022] Preferably, the assessment output module further includes a comprehensive assessment report generation unit that links the neurophysiological coordination score with the operational accuracy rating. When the neurophysiological coordination score is below the critical value and the operational accuracy rating is unqualified, the flight training phase is marked as a high-risk state.

[0023] The comprehensive assessment report generation unit also extracts abnormal peak values ​​of respiratory wave frequency amplitude fluctuation rate from the cognitive stability quantification matrix and associates them with the time nodes when high-risk states occur.

[0024] Preferably, the system further includes a training optimization suggestion module, which receives a comprehensive training evaluation report generated by the evaluation output module and analyzes the correlation pattern between neurophysiological coordination score and operational accuracy rating;

[0025] The training optimization suggestion module generates specific training programs for visual attention allocation or physiological stress regulation based on the correlation pattern, and feeds the specific training programs back to the adaptive feedback module to update the visual stimulus intensity adjustment rules.

[0026] Preferably, the training optimization suggestion module includes a scheme customization engine, which generates a visual tracking training scheme to enhance context awareness when the neurophysiological coordination score is detected to be continuously lower than the baseline.

[0027] The customized engine also receives data on steep drops in the cognitive load index curve and generates anti-interference physiological adjustment training instructions based on the flight operation type corresponding to the steep drops.

[0028] Preferably, the system is equipped with a data persistence storage module, which archives the neurophysiological synchronization dataset of the physiological signal acquisition module, the cognitive stability quantification matrix of the cognitive state modeling module, and the comprehensive training evaluation report of the evaluation output module in a time sequence.

[0029] The persistent data storage module is also associated with the execution records of the special training programs generated by the training optimization suggestion module, forming an evolutionary tracking database of pilots' long-term neurophysiological adaptation.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] By integrating EEG features with multi-dimensional physiological indicators, a comprehensive characterization of pilots' cognitive states was achieved. Compared to single-indicator assessment methods, the integration of multimodal data can more accurately reflect the complex changes in the cognitive process. Time-frequency domain decomposition of EEG signals extracts energy entropy values ​​of different frequency bands, capturing the dynamic characteristics of the brain during information processing. Meanwhile, the heart rate variability feature vector generated from the RR interval sequence of ECG reflects the regulatory state of the autonomic nervous system. The correlation mapping established by the two through cross-modal attention mechanisms reveals the intrinsic connection between cognitive activities and physiological responses, making the assessment of cognitive states more in-depth and comprehensive.

[0032] The cognitive load index and cognitive stability quantification matrix generated by the cognitive state modeling module provide a quantitative basis for evaluation. The cognitive load index intuitively presents the pilot's level of concentration, while the quantification matrix constructed from the respiratory wave frequency amplitude fluctuation rate reflects the stability of the cognitive state. The combination of the two can comprehensively evaluate the pilot's cognitive performance in different mission scenarios. For example, when performing emergency evasive maneuvers, it can determine whether the pilot maintains a high level of focus and whether their cognitive state is stable, avoiding the one-sidedness of evaluation based on a single indicator.

[0033] The adaptive feedback module dynamically adjusts the simulated scenario parameters based on key operational nodes during the flight phase, enhancing the relevance and effectiveness of training. Different flight phases place different cognitive demands on pilots; takeoff and landing require higher levels of concentration, while cruise phases emphasize cognitive stability. Dynamically adjusting the intensity of visual stimuli and the joystick force feedback coefficient can simulate operational environments under varying cognitive loads, helping pilots adapt to various complex situations during training and improving their responsiveness in actual flight.

[0034] The assessment output module integrates data from multiple sources to generate a comprehensive report, including neurophysiological coordination scores and operational accuracy ratings, providing a comprehensive reference for evaluating training effectiveness. The neurophysiological coordination score reflects the synergy between cognitive state and physiological response, while the operational accuracy rating reflects actual operational performance. The combination of these two factors makes the assessment results more comprehensive, facilitating trainers to develop targeted training plans, optimize training content, and promote the overall improvement of pilot skills. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the working principle of the flight training assessment system that integrates EEG characteristics and physiological indicators as described in this invention.

[0036] Figure 2 A flowchart for the preprocessing and synchronization of physiological signal acquisition modules;

[0037] Figure 3 A flowchart for the generation and stability quantification of the cognitive state modeling module index;

[0038] Figure 4 A flowchart for evaluating the output module's score calculation and accuracy analysis;

[0039] Figure 5 A flowchart illustrating the generation and feedback of training optimization suggestion modules. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] See also Figure 1 This invention provides a flight training assessment system that integrates electroencephalogram (EEG) characteristics and physiological indicators, the system comprising:

[0042] By simultaneously acquiring pilots' electroencephalogram (EEG) signals and surface physiological indicators, and combining multimodal data fusion and cognitive state modeling techniques, the system achieves dynamic assessment and feedback regulation of pilots' training status. The system includes a physiological signal acquisition module, a multimodal fusion module, a cognitive state modeling module, an adaptive feedback module, and an assessment output module. The physiological signal acquisition module uses multi-channel EEG electrodes and surface physiological sensors to simultaneously capture raw EEG signals and physiological indicators such as ECG, respiratory waves, and skin conductance. The multimodal fusion module performs time-frequency domain decomposition on the EEG signals, extracting energy entropy values ​​in the Delta, Theta, Alpha, and Beta frequency bands. Simultaneously, it analyzes the ECG RR interval sequence to generate a heart rate variability feature vector and establishes a dynamic correlation mapping between EEG and physiological indicators through a cross-modal attention mechanism. The cognitive state modeling module generates a cognitive load index based on the dynamic correlation mapping and constructs a cognitive stability quantification matrix by combining respiratory wave frequency amplitude fluctuation rate. The adaptive feedback module dynamically adjusts the visual stimulus intensity and joystick force feedback coefficient of the simulated flight scenario according to cognitive state parameters. The evaluation output module integrates neurophysiological data and operational trajectory data to generate a comprehensive training evaluation report that includes neurophysiological coordination scores and operational accuracy ratings.

[0043] Example 1: See Figure 2 The EEG signal preprocessing unit of the physiological signal acquisition module employs a multi-level filtering architecture to process the raw EEG signal. In the initial stage, hardware-level analog filtering eliminates high-frequency noise. Subsequently, a 50Hz power frequency notch filter is implemented in the digital signal processing stage, using a bidirectional IIR filter design to avoid phase distortion. For eye movement artifact interference, the system combines independent component analysis and regression correction algorithms. First, the FastICA algorithm is used to separate independent components containing eye movement artifacts. Then, component identification is performed based on time-frequency feature templates of typical eye movement artifacts. Finally, the least squares method is used to construct an artifact regression model and remove it from the original signal. EMG interference suppression employs a wavelet thresholding denoising method, using the sym8 wavelet basis function for 8-level decomposition and soft thresholding of high-frequency coefficients. The preprocessed EEG signal retains the effective frequency band of 0.5-45Hz, resulting in a significant improvement in the signal-to-noise ratio.

[0044] The physiological signal synchronization unit employs a hierarchical time synchronization strategy. At the hardware level, an FPGA generates synchronization pulse signals accurate to the microsecond level, driving all acquisition devices to operate in master-slave mode. At the data level, a timestamp alignment mechanism based on the NTP protocol is used, embedding high-precision timestamps into each frame of data packets. For signal sources with different sampling rates, the system uses cubic spline interpolation to normalize the sampling rate, resampling ECG signals from 1000Hz and respiratory signals from 250Hz to the same 1000Hz sampling rate as EEG signals. During time alignment, the system continuously monitors the clock drift of each signal channel, automatically triggering a timestamp correction procedure when the cumulative deviation exceeds 1ms to maintain the temporal consistency of multimodal data.

[0045] The frequency domain decomposition engine of the multimodal fusion module uses an improved wavelet packet transform algorithm to process EEG signals. During the decomposition process, the db4 wavelet basis function is used for a 6-level recursive decomposition, forming 64 sub-bands covering 0.5-30Hz. For typical cognitive states during flight missions, the system focuses on extracting four characteristic frequency bands: Delta waves (0.5-4Hz) corresponding to deep relaxation, Theta waves (4-8Hz) associated with working memory load, Alpha waves (8-13Hz) reflecting resting arousal levels, and Beta waves (13-30Hz) characterizing cognitive activity. Energy entropy calculation employs a sliding time window analysis, using a Hamming window function. The window length of 2 seconds and the step size of 500ms balance the requirements of temporal resolution and spectral stability. The energy of each frequency band is normalized and converted into a probability distribution, and then the Shannon entropy value is calculated to form time series features.

[0046] The physiological feature analysis unit performs multi-scale feature extraction on the ECG RR interval sequence. Time-domain analysis includes conventional indicators such as standard deviation (SDNN) and root mean square (RMSSD) of the difference between adjacent RR intervals. Nonlinear dynamic analysis employs the sample entropy algorithm with an embedding dimension of 2 and a similarity tolerance of 0.2 times the time series standard deviation, effectively quantifying the complexity of the RR interval sequence. Fractal dimension calculation uses the Higuchi algorithm, setting the maximum subsequence length to 1 / 4 of the total number of RR intervals, and obtaining the fractal curve slope through least-squares fitting. Frequency-domain analysis estimates the high-frequency power to low-frequency power ratio using Lomb-Scargle periodograms, adapting to non-uniformly sampled RR interval sequences. The final generated 12-dimensional heart rate variability feature vector contains time-domain, frequency-domain, and nonlinear feature indicators, comprehensively reflecting the activity state of the autonomic nervous system.

[0047] The skin conductance response signal processing employs an adaptive baseline correction method. The system dynamically tracks the slow-varying trend components of the skin conductance signal, extracts the baseline using a moving average filter, and sets the window length to 30 seconds. For transient skin conductance response events, the detection algorithm comprehensively considers the signal slope change and amplitude threshold, defining a valid response as a rising edge slope exceeding 0.05 μS / s and an amplitude exceeding the baseline by 10%. Respiratory wave signal processing focuses on extracting features in two dimensions: frequency and amplitude. Instantaneous frequency estimation uses a zero-crossing detection method combined with median filtering, and the amplitude envelope is extracted using Hilbert transform. The system pays particular attention to the fluctuation characteristics of respiratory wave frequency, calculating the standard deviation of respiratory frequency per minute as an indicator of respiratory wave frequency-amplitude fluctuation rate.

[0048] The time alignment accuracy of multimodal data is ensured through a cross-validation mechanism. The system periodically injects standard test pulse signals to verify the accuracy of the time stamps for each channel. When an inter-channel delay exceeding 1ms is detected, a time compensation program is automatically initiated, adjusting the data buffer via a digital delay line. To address potential packet loss during signal transmission, the system employs a linear prediction method for data compensation, with a maximum tolerance for consecutive packet loss of 3 frames. The time synchronization status is displayed in real-time via a visual interface, including clock deviation curves for each channel and a synchronization error statistical histogram.

[0049] The calculation process for frequency band energy entropy incorporates a quality control mechanism. The system monitors the signal-to-noise ratio (SNR) of each frequency band in real time. When the energy in a specific frequency band is contaminated by noise exceeding a threshold, the entropy value for that period is automatically recalculated or marked as invalid data. For periods of signal distortion caused by motion artifacts, the system repairs the distortion through linear interpolation of the preceding and following valid data segments, with a maximum repair duration limited to 1 second. Before outputting the energy entropy value sequence, it undergoes moving median filtering for smoothing, with a window width set to 5 sampling points to eliminate the influence of isolated outliers.

[0050] Strict quality control is implemented during heart rate variability feature extraction. The R-wave detection algorithm combines amplitude thresholding with morphological template matching, and each detected R-wave is manually labeled and verified. When consecutive missed or false detections occur, the system automatically adjusts detection parameters or switches to a backup detection algorithm. For time periods containing arrhythmias such as premature atrial contractions, the system automatically excludes these data segments from feature calculation. After feature vector generation, a Mahalanobis distance test is performed to remove feature combinations that exceed the normal physiological range.

[0051] The signal preprocessing workflow implements multi-level quality monitoring. The raw signal is first displayed with a visual interface to show its waveform quality, and operators can manually mark abnormal periods. An automatic quality assessment algorithm calculates the signal-to-noise ratio, baseline drift, and peak-to-peak value for each period, generating a signal quality index curve. For periods where the quality index is below a threshold, the system automatically triggers a reacquisition process or prompts the operator to check electrode contact. The preprocessed signal must pass through an independent quality verification module, with verification criteria including the rationality of the frequency band energy distribution and the degree of artifact residue.

[0052] The time synchronization system employs a redundant design to ensure reliability. The master clock module is equipped with dual crystal oscillators for hot backup; it automatically switches to the backup oscillator when a frequency drift exceeding 1 ppm is detected in the master crystal oscillator. Timestamp recording uses a dual-track parallel approach of FPGA hardware timing and software timing, periodically comparing the differences between the two. The network time protocol service is configured in a multi-server redundancy mode, obtaining reference times from multiple time sources and using a majority voting mechanism to determine the final synchronization reference. The time synchronization status is incorporated into the system health monitoring system; any anomaly triggers a tiered alarm mechanism.

[0053] Example 2: See Figure 3 The cognitive load index generation unit of the cognitive state modeling module employs a deep convolutional neural network architecture to process multimodal physiological features. The network input layer receives a 76-dimensional composite feature consisting of a 64-dimensional EEG energy entropy sequence and a 12-dimensional heart rate variability feature vector. Zero-mean normalization preprocessing ensures that all feature dimensions are of the same scale. The main network structure contains three convolutional blocks, each consisting of a one-dimensional convolutional layer, a batch normalization layer, and a LeakyReLU activation layer in sequence. The convolutional kernel size is set to 5 sampling points along the time axis, and the number of output channels for each convolutional block is 32, 64, and 128, respectively. After the last convolutional block, the feature map is compressed into a 128-dimensional feature vector through a global average pooling layer. Finally, a cognitive load index in the range [0,1] is output after a fully connected layer and a Sigmoid activation function. The network training uses the Adam optimizer, with the loss function defined as the mean squared error between the predicted index and expert annotations. A dropout rate of 0.3 is applied during training to prevent overfitting.

[0054] The fluctuation rate of respiratory wave frequency amplitude is calculated using sliding window statistical analysis. The system calculates the standard deviation of the instantaneous frequency sequence of the respiratory wave signal with a window length of 10 seconds and a step size of 1 second. Instantaneous frequency estimation employs an improved peak detection algorithm. First, the respiratory wave signal is low-pass filtered at 4Hz to eliminate high-frequency noise. Then, the respiratory cycle boundary is determined by finding the zero-crossing point of the first derivative of the signal. The fluctuation rate within a single window is calculated using the following formula: in, This represents the frequency amplitude fluctuation rate of the respiratory wave in the current window. The number of complete respiratory cycles detected within the window. For the The instantaneous frequency per respiratory cycle, This represents the arithmetic mean of the respiratory rates within the window. The system applies median filtering to the calculated volatility sequence, setting the window width to 5 data points to eliminate the influence of transient abnormal fluctuations.

[0055] The detection of abrupt changes in skin conductance response amplitude employs a two-level criterion mechanism. The primary detection is based on the first-order difference of the signal; a potential event is triggered when the difference between three consecutive sampling points exceeds three standard deviations of the baseline noise level. The secondary verification combines amplitude change rate and duration criteria, requiring the effective abrupt change to have a rising edge slope lasting at least 0.5 seconds and an amplitude change exceeding 15% of the baseline value. The system establishes personalized baseline reference values ​​for each pilot, with baseline data derived from the statistical characteristics of skin conductance signals during the 5-minute resting state prior to training. The timestamp and amplitude change of the abrupt event are recorded as a two-dimensional feature vector, participating in the construction of the subsequent cognitive stability quantification matrix.

[0056] The cognitive stability quantification matrix organizes physiological indicators using a six-dimensional feature space. The rows of the matrix correspond to the time dimension, dividing the training process into equal-length time intervals of 30 seconds. The columns contain six feature dimensions: mean respiratory wave frequency fluctuation rate, number of skin conductance mutations, heart rate variability sample entropy, EEG Theta / Alpha power ratio, EEG global complexity index, and pupil diameter change rate. Each matrix cell stores the standardized values ​​of each feature within its corresponding time interval. The standardization process uses the z-score method, with parameters derived from the statistical distribution of a large-sample pilot database. Strict quality control is implemented during matrix completion; if the data missing rate for a given time interval exceeds 20%, that time interval is marked as invalid and compensated using interpolation from adjacent time intervals.

[0057] The adaptive feedback module's scene control decision-maker implements rule-based fuzzy logic control. Input variables include the first-order difference of the cognitive load index, the current row vector of the cognitive stability matrix, and the flight phase encoding. The fuzzification process converts continuous variables into three linguistic variables: cognitive load changes are categorized as "decreasing," "stable," and "increasing"; stability states are categorized as "disorderly," "fluctuating," and "stable." The knowledge base contains 27 production rules, such as "if cognitive load increases and stability is disordered, then enhance visual stimulation." Defuzzification uses the centroid method to calculate precise output values, ultimately generating a visual stimulation modulation coefficient in the range [0,1] and a force feedback scaling factor in the range [0.5,2].

[0058] Visual stimulus intensity adjustment employs multi-parameter coordinated control. The system adjusts three visual dimensions of the head-up display (HUD) symbols: brightness, contrast, and flicker frequency. Brightness adjustment follows the CIE 1931 luminance perception curve, with a base brightness set at 100 cd / m² and a maximum adjustment range of ±30%. Contrast enhancement utilizes a histogram stretching algorithm, aiming to achieve a contrast ratio of 10:1 or higher for key instruments. Flicker frequency is controlled within the 4-8 Hz range to avoid inducing photosensitive reactions. The combination of adjustment parameters is dynamically selected based on the flight phase, prioritizing brightness adjustment during cruise and contrast adjustment during landing. All adjustments are performed with a gradual transition, with each adjustment not exceeding 10% of the baseline value to avoid sudden shifts in attention.

[0059] Adjusting the joystick force feedback coefficient involves reconstructing dynamic parameters. The basic force feedback model is a second-order mass-spring-damped system, and the damping coefficient in its dynamic equations... and stiffness coefficient The system updates dynamically based on cognitive state. When the cognitive stability quantification matrix shows an increase in autonomic neural activation, the system adjusts parameters according to the following strategy: damping coefficient. The stiffness coefficient was increased from the default value of 0.4 N·s / m to 0.7 N·s / m, increasing the resistance to motion of the joystick; The force was reduced from 20 N / m to 15 N / m to decrease the return force at the center position. The parameter adjustment process employed critical damping constraints to prevent drastic changes in the system's dynamic response characteristics. Force feedback commands were implemented through real-time control loops with an update frequency of 1 kHz, ensuring the continuity and accuracy of force feedback.

[0060] The identification of critical operational nodes during the flight phase employs a dual-judgment mechanism. Static nodes are preset according to standard operating procedures, including standard moments such as the start of takeoff roll, the instant of liftoff, and altitude transitions. Dynamic nodes are determined through real-time analysis of flight parameters; a temporary critical node marker is triggered when the airspeed change rate exceeds 0.5 m / s² or the heading angle deviation exceeds 5°. The node time window is set to 10 seconds before and 20 seconds after the event, during which the sampling frequency of cognitive state parameters increases from 1 Hz to 4 Hz. The system maintains a priority queue to manage multiple concurrent nodes, ensuring that adjustment commands from high-priority nodes can be responded to in a timely manner.

[0061] The real-time visualization of cognitive state parameters employs a multi-layered presentation design. The main monitor displays the trend curve of the cognitive load index, with the background color gradually changing according to the numerical range: green (0-0.3), yellow (0.3-0.7), and red (0.7-1). The auxiliary panel displays the current state of the cognitive stability quantification matrix in the form of a radar chart, with the values ​​of the six feature dimensions connected to form an irregular hexagon. Emergency alarms use a three-level prompt: when the cognitive load index exceeds 0.8 for 30 consecutive seconds, the system triggers an audible alarm and flashes a red border; when the stability matrix shows anomalies in more than three dimensions, an orange warning sign is displayed in the corresponding flight phase marker area. The position and size of all visualization elements conform to ergonomic standards, ensuring that instructors do not need to frequently adjust their line of sight during monitoring.

[0062] The priority management of parameter adjustment commands employs dynamic weight allocation. The system assigns a base weight to each type of adjustment command: 0.6 for visual stimulus adjustment and 0.4 for force feedback adjustment. These weights are dynamically adjusted according to the flight phase, increasing to 0.7 for force feedback adjustment during takeoff and landing. When command conflicts are detected, the arbitration module calculates a comprehensive priority based on the flight safety margin; if the safety margin is below 30%, the force feedback enhancement command is forcibly executed first. A sliding time window limit is implemented in the command queue, ensuring that the interval between consecutive adjustment commands is no less than 2 seconds to prevent parameter oscillations. All adjustment operations are logged in detail, including timestamps, adjustment types, parameter changes, and decision-making basis, supporting post-event analysis and system optimization.

[0063] Example 3: See Figure 4 The neurophysiological coordination calculation unit of the output evaluation module uses a multi-dimensional feature fusion algorithm to process cognitive state parameters. Input data includes time-aligned cognitive load index curves and a cognitive stability quantification matrix. Both are first normalized to eliminate dimensional differences. The normalization process uses a dynamic range compression method, using the maximum and minimum values ​​from historical data as reference boundaries. The neurophysiological coordination score is calculated using the following formula: in, Indicates the neurophysiological coordination score. This represents the normalized cognitive load index for the current time period. It is a deviation index of the cognitive stability quantification matrix. and These are the weighting coefficients for cognitive load and stability, respectively, with default settings of 0.6 and 0.4. Deviation index The Mahalanobis distance between the six feature dimensions of the stability matrix and the ideal value is calculated to reflect the overall degree of deviation. The scoring is performed in a sliding 10-second time window with a 50% window overlap, ultimately generating a consistency curve with a time resolution of 5 seconds.

[0064] The operational accuracy analysis unit performs multi-level comparisons of flight trajectories. The standard flight trajectory template contains ideal value sequences for three dimensions: heading angle, altitude, and airspeed, with a sampling interval of 1 second. The dynamic time warping algorithm employs improved constraints, limiting the path search range to a time window of ±5 seconds near the diagonal. The heading angle deviation calculation converts the instantaneous difference between the actual heading and the template heading into an angle deviation using the arctangent function. Altitude control error is calculated using a relative error method to avoid scaling issues caused by absolute values. Accuracy ratings are divided into four levels: A (Excellent), B (Good), C (Acceptable), and D (Unacceptable). The grading thresholds are dynamically adjusted according to the flight phase; for example, the tolerance range during landing is 30% stricter than that during cruise.

[0065] The comprehensive assessment report generation unit establishes a state-performance correlation model. High-risk states are determined using a composite condition: a neurophysiological coordination score below 60 points for three consecutive sampling points, and a concurrent operational accuracy rating of D. While marking high-risk periods on the timeline, the system automatically retrieves abnormal physiological characteristic patterns within those periods. The detection of abnormal peak values ​​in respiratory wave frequency amplitude fluctuation rate employs a local extremum algorithm, defining fluctuations exceeding twice the standard deviation of the mean within a 60-second interval as valid anomalies. The report generation process utilizes automatic summarization technology, extracting key event nodes, duration distributions, and abnormal pattern features from hours of training data to form structured assessment conclusions.

[0066] The data visualization subsystem employs layered rendering technology to present evaluation results. The main view displays a time-series graph overlaid with the coordination score curve and operational accuracy rating, using color coding to distinguish different flight phases. The detailed view allows users to click to query detailed data at any point in time, including the EEG spectrogram, physiological index values, and flight control inputs at that time. Abnormal events are highlighted with semi-transparent areas to maintain the visibility of the underlying data. All charts support dynamic zooming and panning, with time resolution adjustable from an overview of the entire training session to second-level detail. Visualization parameters are stored as style templates, allowing for quick switching of display modes according to different evaluation needs.

[0067] Subsequent analysis of high-risk states employs pattern mining techniques. The system automatically clusters high-risk events from historical training, extracting common temporal features, flight phase distributions, and physiological indicator combinations. Each risk pattern is associated with a typical scenario description, such as "insufficient instrument scanning" or "excessive control input." The pattern library supports similarity retrieval; newly emerging risk events are matched with existing patterns, providing possible causal analyses. Risk pattern visualization uses parallel coordinate graphs to display multi-dimensional feature relationships, highlighting key distinguishing dimensions.

[0068] The report export function supports multiple structured formats. Standard evaluation reports are stored in JSON-LD format, containing complete time-series data, evaluation conclusions, and metadata descriptions. Summary reports generate Word document templates, automatically populating key data tables and trend charts. Interactive reports are output in HTML5 format, with built-in dynamic visualization components supporting offline viewing. All reports include digital signatures and timestamps to ensure the authenticity and immutability of the evaluation results. The report distribution system implements fine-grained access control, with different user roles having access to different report content and levels of detail.

[0069] Personalized adaptation of assessment parameters employs a progressive calibration method. The system records each pilot's historical assessment data, gradually establishing an individualized scoring baseline. The first three assessments for new pilots are used for reference only; from the fourth assessment onwards, they are formally incorporated into their individual competency files. The calibration process considers differences in flight experience, introducing adjustments to the aircraft type's assessment parameters for pilots undergoing conversion training. Personalized parameters are stored in a separate configuration file, containing custom settings for scoring weights, grading thresholds, and risk assessment rules.

[0070] The system integration testing framework verifies the reliability of the evaluation process. The test case library contains hundreds of standard test scenarios, covering various typical combinations of cognitive states and operational behaviors. Automated test scripts simulate the complete evaluation process, verifying every step from raw data input to report generation. Performance testing monitors evaluation latency, ensuring real-time evaluation calculations are completed within 200ms. Compatibility testing verifies the system's support for different sampling rates and data acquisition devices from different vendors. Test results generate detailed coverage reports, identifying functional modules requiring enhanced verification.

[0071] The interpretable analysis of the evaluation results employs interpretable AI technology. The underlying reasons for significant score changes are revealed through feature importance analysis, showing which physiological indicators dominated the score changes. Decision path visualization demonstrates the key reasoning steps in score calculation, aiding in understanding the system's evaluation logic. Counterfactual analysis simulates score changes after improvements in specific indicators, quantifying the influence of each factor. The interpretable analysis results are presented using natural language descriptions combined with visual charts, lowering the technical barrier to entry.

[0072] The long-term trend analysis module tracks the evolution of pilots' capabilities. Evaluation results from each training session are stored in a time-series database, allowing for viewing of score changes at different granularities: weekly, monthly, and quarterly. Trend prediction employs an exponential smoothing algorithm, providing the expected capability development curve for the next three training sessions. Anomaly detection identifies sudden drops or sustained declines in scores, triggering special attention alerts. The trend report compares individual performance with the average level of pilots of the same aircraft type, identifying relative strengths and areas for improvement.

[0073] A multi-evaluator consistency verification mechanism enhances the reliability of results. The evaluation process for important training tasks incorporates independent scores from multiple instructors, calculating intra-group correlation coefficients to measure the consistency between human and system evaluations. Cases with discrepancies exceeding a threshold undergo expert review to analyze the sources of discrepancies and update evaluation rules. Consistency analysis results are fed back to the evaluation model optimization stage, continuously improving the system's evaluation accuracy. The verification process retains complete annotation information, recording the judgment basis and notes of each evaluator.

[0074] Example 4: See Figure 5 The training optimization suggestion module's customized solution engine generates targeted training plans by analyzing the correlation patterns between neurophysiological coordination scores and operational accuracy ratings. The system first establishes a mapping table between evaluation results and training elements.

[0075] Table 1: Establishing a mapping relationship between evaluation results and training elements.

[0076] The scheme generation process employs a decision tree algorithm. Input features include trend characteristics of neurophysiological coordination scores, distribution patterns of operational accuracy ratings, and anomaly patterns in the cognitive stability quantification matrix. Each branch node of the decision tree corresponds to a key judgment condition, such as "whether the coordination score has decreased by more than 15% in the last three training sessions." Leaf nodes are associated with specific training scheme templates, and the template parameters are dynamically adjusted based on historical optimization results.

[0077] The visual tracking enhancement program employs a dynamic optotype system. The initial optotype shape is a circle with a diameter of 2cm, and the color uses high-contrast yellow and black stripes. The motion trajectory algorithm includes three basic modes: straight-line saccade, sinusoidal curve tracking, and random jump, automatically adjusting the mode ratio according to training progress. When the deviation between the trainee's fixation point and the optotype center is consistently less than 1° of visual angle, the system gradually increases the optotype movement speed, with each increase not exceeding 5° / s. Eye movement data is recorded in real-time during training, and the fixation stability index and saccade accuracy are calculated as the basis for program adjustments.

[0078] The task priority training program designs multi-task parallel scenarios. The system simultaneously displays the main flight instruments, navigation map, and system alarm information on the head-up display, requiring trainees to process each type of information according to its urgency. The task generator dynamically adjusts the task density based on trainee performance, with an initial task interval of 12 seconds and each task lasting 3-5 seconds. Prioritization accuracy and response latency are used as core evaluation metrics. When the accuracy of five consecutive tasks reaches 90% or higher, the system reduces the task interval by 10%. The training scenario includes typical interference factors, such as simulated radio communication and sudden system alarms, to cultivate trainees' task switching capabilities.

[0079] The breathing regulation training program integrates biofeedback technology. The system overlays a breathing guidance indicator onto the flight scenario, displaying the target breathing rhythm as a dynamic light strip. The trainee's real-time breathing waveform is acquired via a chest strap sensor and compared with the target waveform. The feedback interface displays two dimensions: breathing rate deviation and tidal volume consistency. An audible alert is triggered when the deviation exceeds a threshold. The training program employs a progressive difficulty design, initially requiring only the maintenance of a stable breathing rate, and later adding operational tasks such as altitude control to create dual-task pressure. After each training session, a breathing parameter improvement curve is generated, showing the improvement in control stability at each stage.

[0080] The anti-interference training program constructs a multimodal interference environment. Visual interference sources include a randomly flashing high-intensity light simulator outside the cockpit, with four intensity levels corresponding to different brightness variations. Auditory interference uses acoustically processed radio noise, with the signal-to-noise ratio gradually reduced from +10dB to -5dB. Tactile interference is achieved through high-frequency micro-vibrations of the control stick, with the vibration frequency randomly varying within the 20-50Hz range. The temporal relationship between interference events and critical flight operations is carefully designed to avoid predictable patterns. Training effectiveness is evaluated by examining the standard deviation of operational parameters under interference conditions, reflecting the trainee's improved anti-interference capabilities.

[0081] The training program is implemented with closed-loop optimization. Before each training session, the system loads a basic program template and fine-tunes initial parameters based on the trainees' recent performance. During training, key physiological indicators and operational performance are monitored in real time. When an expected improvement trend is detected, the training difficulty level is automatically increased; if maladaptation occurs, the level is downgraded or the training mode is switched. Detailed logs of program adjustments are recorded, including the adjustment time, decision basis, and parameter changes, forming a complete optimization trajectory. After training, the system compares the actual results with the expected goals, calculates the program effectiveness index, and uses it to update the parameter mapping relationships of the decision tree nodes.

[0082] Personalized training program recommendations employ a collaborative filtering algorithm. The system maintains a database containing training records from hundreds of pilots. When a new trainee's evaluation characteristics are similar to a certain group in the database, training programs that have received good responses from that group are prioritized. Similarity calculation considers three dimensions: the distribution pattern of neurophysiological characteristics, years of flight experience, and aircraft adaptation curve. The recommendation results present three alternative programs, each emphasizing different optimization directions, for instructors to choose from. The actual results after program implementation are fed back to the recommendation system to continuously optimize the similarity calculation model.

[0083] Example 5: The persistent data storage module employs a hierarchical storage architecture to manage the multimodal data generated by the flight training and evaluation system. Raw neurophysiological data is stored in binary format, with each data packet containing an acquisition timestamp, device identifier, and signal quality marker. EEG signals are processed using a lossy compression algorithm, reducing the data volume to 30% of its original size while preserving effective frequency band information. ECG and respiratory signals store RR interval sequences and respiratory wave envelope data, rather than complete waveform records. Electrodermal response data records baseline values ​​and abrupt event parameters, avoiding the storage of continuously changing raw signals. All physiological data is indexed in a three-level system by pilot ID, training date, and mission phase, supporting millisecond-level time precision retrieval.

[0084] Cognitive state modeling data is persistently stored in matrix format. The cognitive stability quantification matrix is ​​compressed using sparse matrix techniques, recording only non-zero elements and their position indices. The cognitive load index curve stores key feature points rather than the complete sampling sequence, and the original curve is reconstructed during retrieval using an interpolation algorithm. An incremental update strategy is implemented during storage, appending only new data segments with each training iteration to avoid repeatedly storing unmodified historical data. Data version control uses snapshot technology to retain the complete state of important evaluation nodes, and only records the differences in routine modifications. Metadata description files record in detail the physical meaning and numerical units of the matrix dimensions, ensuring long-term readability.

[0085] The assessment report data is stored using a hybrid approach of structured documents and associated databases. The text content of the comprehensive training assessment report is stored in Markdown format, preserving formatting marks while reducing storage space. Chart data is stored separately; vector graphics are saved as SVG format, and bitmaps are compressed using WebP lossy compression. The raw comparison data for operational accuracy ratings is stored in a time-series database format, supporting efficient range queries and similarity retrieval. High-risk status markers are associated with binary event description files, including triggering conditions, duration, and associated evidence chains. The reference relationships between reports are maintained through a graph database, forming a network structure of assessment knowledge.

[0086] The execution records of specialized training programs are stored using an event sourcing model. The execution process of each training program is decomposed into a series of immutable event objects, appended to a read-only log in chronological order. Each event object contains a complete context, including execution time, parameter settings, operational behavior, and result feedback. During retrieval, the training state at any given time point is reconstructed by replaying the event sequence. Program adjustment records store the reasoning chain of the decision-making process, including considered alternative programs and reasons for rejection. A two-way reference is established between execution performance data and evaluation reports, supporting tracing from training programs to evaluation results, or retrieving relevant training records from evaluation deficiencies.

[0087] The Long Term Evolutionary Tracking Database implements a spatiotemporal multidimensional index. The time dimension supports three query methods: natural time, training cycle, and circadian rhythm, allowing data analysis by calendar date, training batch, or circadian clock cycle. The spatial dimension records training environment characteristics, including scene parameters such as simulator model, cockpit layout, and ambient noise level. The physiological adaptation dimension tracks neural plasticity indicators, such as the long-term drift trend of EEG frequency band energy distribution. The index structure uses a hybrid layout of B+ trees and R trees to optimize query efficiency across different dimensions. A data aging strategy automatically converts detailed raw data exceeding five years into aggregated statistical form for storage, preserving trend information while freeing up storage space.

[0088] Data security is protected through a multi-layered encryption mechanism. Static data is encrypted using the AES-256 algorithm, and the key management system stores encryption keys and encrypted data separately. Transmission is protected using the TLS 1.3 protocol, and certificates are issued by an internal CA. Access control is based on attribute-based encryption technology, dynamically generating fine-grained data access permissions. Audit logs record all data access activities, including query content, result set size, and access time. Data de-identification removes direct personal information, and pseudo-name authentication maintains data consistency. Secure erasure ensures that data on decommissioned storage media is unrecoverable.

[0089] Data consistency maintenance employs a distributed transaction protocol. Cross-module data updates ensure atomicity through a two-phase commit, meaning all relevant storage nodes either all update successfully or all are rolled back. Concurrency control utilizes multi-version concurrency technology, ensuring read operations do not block write operations, and vice versa. A conflict detection mechanism identifies concurrent modifications, and operation transformation algorithms resolve editing conflicts for complex objects such as evaluation reports. Periodic consistency checks scan the entire database, comparing the degree of matching between the index and the actual data, and repairing any discovered anomalies or inconsistencies. Data repair tools support selective recovery, allowing data within a specific time frame to be rolled back to a known good state.

[0090] Storage system performance optimization employs a tiered caching strategy. Hot data, including recent training records and frequently used evaluation templates, is stored in an in-memory database. Warm data, covering detailed training data from the past three months, is stored on high-speed solid-state drives (SSDs). Cold data is migrated to a hard disk drive (HDD) array to store historical archive information. The cache replacement algorithm considers factors such as data access frequency, relational query patterns, and future training plans. A prefetch mechanism loads relevant pilots' historical data in advance based on the training schedule. Batch operations are scheduled for execution during system idle periods, such as data compression and index rebuilding tasks performed at night.

[0091] Data migration and compatibility assurance are achieved through a standardized intermediate format. Cross-version migration uses intermediate dump files in JSON format, containing complete data schema and value range descriptions. Format converters handle changes in storage structure, such as field splitting or encoding changes. A backward compatibility layer allows new systems to read data in the old format and performs real-time conversion as needed. Data validation tools check the completeness and accuracy of the migration results, sampling and comparing the consistency between the original and migrated data. The migration report records detailed processing logs, conversion rules, and anomalies for subsequent analysis and reference.

[0092] Data lifecycle management employs an automated strategy. Raw physiological data, retaining full accuracy for three months, is downsampled and stored, retaining only statistical features per channel per second. Intermediate evaluation results are converted to read-only mode after one year, prohibiting further modification. Detailed training records older than five years are automatically generated into aggregate summaries, replacing the original detailed data. The destruction process follows secure erasure standards, overwriting storage blocks multiple times. The lifecycle strategy can be temporarily adjusted for specific research projects, extending the retention period of critical datasets. The strategy engine supports conditional rules, such as "permanently retain raw data related to high-risk events."

[0093] The data retrieval interface offers multilingual support. The core query API supports a subset of SQL and is compatible with common data analysis tools. Domain-specific queries are encapsulated into high-level interfaces, such as "get a pilot's cognitive load fluctuations during landing." The natural language query parser understands conversational requests such as "show all periods last week where coordination was below 60." The streaming interface allows subscription to data updates and real-time push of newly generated assessment results. Batch export functionality generates data packages conforming to aviation medicine research standards, including complete metadata description documents.

[0094] The data visualization subsystem supports historical trend analysis. A timeline view overlays long-term changes in cognitive state parameters and operational performance. A heatmap matrix presents the distribution of ability characteristics across different training stages. A radar chart compares current evaluation results with historical best performance. Visualization configurations can be saved as templates, allowing for quick reproduction of specific analytical perspectives. Interactive exploration allows drill-down to view the detailed context of any data point. The export function generates interactive HTML reports, including dynamic filtering and detailed hints.

[0095] Data quality control is implemented through end-to-end monitoring. At the acquisition end, signal quality indicators are verified and low-confidence data segments are flagged. The stored procedure verifies data integrity hash values ​​and detects transmission corruption. During data retrieval, timestamp continuity and numerical reasonableness are checked. Periodic data cleaning identifies and repairs outliers, such as drift caused by sensor failure. Quality reports summarize data loss rates and anomaly occurrence rates at each stage, guiding system improvement. Repair tools allow selective recalculation of derived data, such as regenerating evaluation indicators from the original signal.

[0096] Data sharing and collaboration features enable secure data exchange. Sharing policies precisely control which data fields can be accessed by which roles; for example, instructors can see performance scores but not raw physiological signals. Collaborative annotation allows multiple experts to add comments to the same assessment results, forming discussion threads. Version tracking records the modification history of shared data and supports rollback to any shared version. Watermarking technology embeds invisible identification information into sensitive data, tracing the source of leakage. Auditing features record all data sharing activities, including recipients and access times.

[0097] Backup and disaster recovery systems ensure persistent data availability. Real-time synchronization replicates new data to an off-site backup center with latency controlled to within 5 seconds. Daily full backups are saved as read-only snapshots, retaining a 30-day rolling history. Disaster recovery drills are conducted quarterly to test the ability to restore a complete system from backups. Storage hardware uses a RAID-6 configuration, tolerating the simultaneous failure of two disks. A power protection system provides sufficient time for write operations to complete during power outages. Critical data is backed up with air gaps on physically isolated storage media to protect against cyberattacks.

[0098] Storage system monitoring and maintenance are automated. A health dashboard displays key metrics such as storage utilization, I / O latency, and error rate in real time. Predictive maintenance, based on device SMART data, proactively replaces potentially faulty disks. Performance tuning automatically adjusts database parameters to adapt to different load patterns. Capacity planning tools predict future storage needs and recommend expansion timing and scale. An alarm system provides tiered responses to anomalies, ranging from email notifications to automatic failover to backup systems. Maintenance logs record all configuration changes and system interventions, creating a complete operational history.

[0099] The data storage format is designed with long-term readability in mind. Core data uses open standard formats, such as HDF5 for scientific data and Parquet for tabular data. Self-describing formats embed metadata to explain field meanings and encoding methods. Documents are stored in PDF / A format to ensure long-term rendering consistency. Deprecated format detection tools regularly scan the repository and mark outdated format data that needs conversion. The format conversion service automatically migrates old data to currently supported formats, preserving the integrity of the original information.

[0100] The storage system architecture supports horizontal scaling. The distributed file system shards data across multiple nodes, using a consistent hashing algorithm to locate data blocks. The compute-storage separation architecture allows for independent scaling of compute resources and storage capacity. The stateless service design enables components to be replaced or upgraded at any time. The service discovery mechanism automatically identifies newly added storage nodes and adds them to the resource pool. Elastic scaling dynamically adjusts the number of active nodes based on load changes, balancing performance and energy consumption. Data rebalancing automatically adjusts data distribution when nodes are added or removed, maintaining load balance.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A flight training assessment system integrating electroencephalogram (EEG) characteristics and physiological indicators, characterized in that, include: The physiological signal acquisition module is used to simultaneously capture multi-channel raw EEG signals and surface physiological index data of the pilot during simulated flight missions. The surface physiological index data includes at least the ECG RR interval sequence, respiratory wave frequency amplitude, and skin conductance response amplitude. The multimodal fusion module is used to perform time-frequency domain decomposition processing on the raw EEG signal to extract energy entropy values ​​of different frequency bands, and at the same time, analyze the ECG RR interval sequence to generate heart rate variability feature vectors, and establish a dynamic correlation mapping between EEG entropy values ​​and physiological feature vectors through a cross-modal attention mechanism. The cognitive state modeling module is used to generate a cognitive load index reflecting the pilot's attention concentration based on the dynamic correlation mapping, and to construct a cognitive stability quantification matrix by combining the respiratory wave frequency amplitude fluctuation rate. An adaptive feedback module is used to receive the cognitive load index and cognitive stability quantification matrix, and dynamically adjust the visual stimulus intensity and joystick force feedback coefficient of the simulated flight scenario according to preset key operation nodes of the flight phase. The evaluation output module integrates cognitive load index, cognitive stability quantification matrix and simulated flight operation trajectory data to generate a comprehensive training evaluation report that includes neurophysiological coordination score and operation accuracy rating. The assessment output module includes a neurophysiological coordination calculation unit, which performs normalized weighted fusion of the cognitive load index curve and the cognitive stability quantification matrix to generate a neurophysiological coordination score that reflects the degree of coordination between EEG features and physiological indicators. The evaluation output module also includes an operation accuracy analysis unit, which dynamically time-warped and matched the simulated flight operation trajectory data with the standard flight trajectory template, calculates the mean heading angle deviation and altitude control error rate, and outputs the operation accuracy rating. The assessment output module further includes a comprehensive assessment report generation unit, which links the neurophysiological coordination score with the operational accuracy rating. When the neurophysiological coordination score is below the critical value and the operational accuracy rating is unqualified, the flight training phase is marked as a high-risk state. The comprehensive assessment report generation unit also extracts abnormal peak values ​​of respiratory wave frequency amplitude fluctuation rate from the cognitive stability quantification matrix and associates them with the time nodes when high-risk states occur.

2. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 1, characterized in that: The physiological signal acquisition module includes an EEG signal preprocessing unit, which is used to perform power frequency noise filtering and eye movement artifact correction on the multi-channel raw EEG signals, and to separate the effective EEG components using independent component analysis. The physiological signal acquisition module also includes a physiological signal synchronization unit, which binds the corrected EEG components, ECG RR interval sequences, and skin conductance response amplitudes to the same time coordinate system through a timestamp alignment mechanism, forming a time-domain aligned neurophysiological synchronization dataset.

3. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 2, characterized in that: The multimodal fusion module includes a frequency domain decomposition engine, which performs wavelet packet transform on the EEG components to divide them into Delta, Theta, Alpha, and Beta frequency bands, and calculates the energy entropy value sequence for each frequency band. The multimodal fusion module also includes a physiological feature analysis unit, which analyzes the sample entropy and fractal dimension of the ECG RR interval sequence through a nonlinear dynamics algorithm to generate a heart rate variability feature vector.

4. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 3, characterized in that: The cognitive state modeling module includes a cognitive load index generation unit, which receives a frequency band energy entropy value sequence and a heart rate variability feature vector, uses a convolutional neural network model to learn the nonlinear relationship between the energy entropy value and the heart rate variability feature vector, and outputs a cognitive load index curve that varies with flight mission time. The cognitive state modeling module also includes a stability quantification unit, which integrates the respiratory wave frequency amplitude fluctuation rate and the number of skin conductance response amplitude mutations to construct a cognitive stability quantification matrix that reflects the autonomic nervous system activation state.

5. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 4, characterized in that: The adaptive feedback module includes a scene control decision-maker, which triggers visual stimulus intensity adjustment commands and joystick force feedback coefficient adjustment commands based on the numerical threshold of the cognitive load index curve at key operation nodes in the preset flight phase. The scene control decision-maker also receives data on the number of abrupt changes in the skin conductance response amplitude from the cognitive stability quantification matrix. When the number of abrupt changes exceeds the baseline threshold, the damping coefficient of the joystick force feedback coefficient is enhanced.

6. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 1, characterized in that: The system also includes a training optimization suggestion module, which receives a comprehensive training evaluation report generated by the evaluation output module and analyzes the correlation pattern between neurophysiological coordination score and operational accuracy rating. The training optimization suggestion module generates specific training programs for visual attention allocation or physiological stress regulation based on the correlation pattern, and feeds the specific training programs back to the adaptive feedback module to update the visual stimulus intensity adjustment rules.

7. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 6, characterized in that: The training optimization suggestion module includes a scheme customization engine, which generates a visual tracking training scheme to enhance context awareness when the neurophysiological coordination score is detected to be continuously lower than the baseline. The customized engine also receives data on steep drops in the cognitive load index curve and generates anti-interference physiological adjustment training instructions based on the flight operation type corresponding to the steep drops.

8. The flight training assessment system integrating EEG characteristics and physiological indicators according to claim 7, characterized in that: The system is equipped with a persistent data storage module that archives the neurophysiological synchronization dataset from the physiological signal acquisition module, the cognitive stability quantification matrix from the cognitive state modeling module, and the comprehensive training evaluation report from the evaluation output module in a time sequence. The persistent data storage module is also associated with the execution records of the special training programs generated by the training optimization suggestion module, forming an evolutionary tracking database of pilots' long-term neurophysiological adaptation.

Images