CN-122020284-A - Fatigue warning method integrating driver physiological signals and unmanned aerial vehicle flight state data

CN122020284ACN 122020284 ACN122020284 ACN 122020284ACN-122020284-A

Abstract

The fatigue warning method for fusing the physiological signals of the driver and the flight state data of the unmanned aerial vehicle is characterized in that the wearable device and the flight control system synchronously acquire the heart rate variability, the physiological signals such as brain electricity alpha waves, the control frequency, the flight state characteristics such as communication interruption and the like, a multi-mode fusion sample is generated by means of data alignment and cleaning, the physical complexity, decision pressure and time urgency of comprehensive load index modeling are utilized, the self-adaptive optimization of driving classification model parameters is realized through meta-learning, and the individual dynamic discrimination of the fatigue state is realized by combining with the historical response mode of the driver.

Inventors

HU JING
Yin Kengxiong

Assignees

广州骏耀信息科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260107

Claims (10)

1. The fatigue warning method for fusing the physiological signals of the driver and the flight state data of the unmanned aerial vehicle is characterized by comprising the following steps of: S1, acquiring physiological signal data of a driver and flight state data of an unmanned aerial vehicle, and performing alignment marking on the physiological signal data and the flight state data of the unmanned aerial vehicle according to a uniform time stamp; S2, calculating the comprehensive load index of the current task based on a preset task type knowledge base and the unmanned aerial vehicle flight state data; S3, inputting the comprehensive load index into a parameter tuning unit of a fatigue judgment model, wherein the parameter tuning unit is constructed based on a model-independent fine tuning idea under a meta-learning frame, obtains basic parameter space adaptation capacity through historical multitask-fatigue sample training, and generates a corresponding judgment boundary adjustment vector according to the comprehensive load index; s4, dynamically correcting the classification threshold value and the characteristic weight distribution of the original fatigue classifier based on the discrimination boundary adjustment vector to form a real-time fatigue judgment sub-model; s5, inputting the aligned physiological signal data and unmanned aerial vehicle flight state data into the real-time fatigue judgment sub-model, executing feature level fusion analysis, and outputting a preliminary fatigue score at the current moment; S6, calling an individual history response mode memory library of the driver, searching a physiological-behavior response characteristic mean value of the driver under the same or similar comprehensive load index, calculating a deviation value between the preliminary fatigue score and a history benchmark, and generating a personalized deviation compensation vector; And S7, carrying out weighted fusion on the personalized deviation compensation vector and the preliminary fatigue score, and outputting a final fatigue grade judging result.
2. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight status data according to claim 1, wherein the step S7 further comprises: And S8, judging whether the final fatigue grade judging result reaches a preset high fatigue warning threshold value, if so, triggering a multi-mode warning mechanism, and recording the judging context information to a system log.
3. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 1, wherein the step S1 specifically comprises: Based on the wearable physiological sensing equipment, the original bioelectric signals of the driver are collected, and electrocardio signals, electroencephalogram signals and eye movement signals are preprocessed by utilizing a band-pass filtering and baseline drift correction algorithm, so that effective heart rate variability, electroencephalogram alpha wave power and eye movement frequency characteristic parameter sequences are extracted; acquiring flight operation log data in real time through a bus interface of the unmanned aerial vehicle flight control system, analyzing key operation behavior records based on an event recognition algorithm, and generating a structured flight state event stream; A unified clock synchronization mechanism is deployed in the edge computing unit, a network time protocol is combined with a hardware pulse alignment method, and data streams from the physiological acquisition module and the flight state sensing module are marked according to millisecond unified time stamps, so that a multi-source data frame set after time alignment is obtained; performing missing value detection and outlier filtering processing on the time-aligned multi-source data frames, filling short signal loss and removing outliers which are remarkably deviated from a normal range by using a linear interpolation and an isolated forest algorithm respectively, and outputting a cleaned standardized physiological-flight combined data sequence; And packaging the characteristic vector matched with the time stamp by the cleaned standardized physiological-flight combined data sequence to generate a multidimensional fusion data sample.
4. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 3, wherein the wearable physiological sensing device is configured to synchronously acquire ECG, EEG and EOG signals at a sampling frequency of 1-500 Hz, and the unmanned aerial vehicle flight control system bus refreshing period is set to 10-100 ms.
5. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 1, wherein the step S2 specifically comprises: acquiring unmanned aerial vehicle flight state data, including control input frequency, route changing times, target locking quantity and communication interruption frequency, and forming a structured task behavior feature sequence based on uniform timestamp alignment marks; based on a preset task type knowledge base, performing pattern matching and semantic analysis on the structured task behavior feature sequence, and identifying a type tag of a current execution task; According to the identified task type labels and the corresponding complexity baseline weight vectors thereof, primary load component fusion is carried out on the control input frequency, the route changing times and the target locking quantity by utilizing a weighted summation algorithm, and a task physical complexity sub-index is generated; Taking the communication interruption frequency and the time sequence continuity breaking degree as input, extracting the time distribution characteristics of the sudden interference event through a long-short-term memory network model, and calculating a task time urgency sub-index by combining a task remaining time inverse ratio factor; and integrating the task physical complexity sub-index and the task time urgency sub-index based on a rule engine, introducing a target locking switching frequency as a decision pressure modulation item, executing three-dimensional coupling operation through a nonlinear activation function, and outputting a normalized comprehensive load index.
6. The fatigue warning method for fusing physiological signals of a driver and flight state data of an unmanned aerial vehicle according to claim 5, wherein the generated task physical complexity sub-index represents operation burden born by the driver on the space operation and action density level, and the task time urgency sub-index reflects psychological compression strength brought by acceleration of information processing rhythm under task cut-off constraint.
7. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 1, wherein the step S3 specifically comprises: Constructing a basic training set for training a meta-learning type parameter tuning unit based on a multi-task-fatigue labeling sample data set acquired by history; Performing meta training on the initialized fatigue judgment parameter tuning network by using a model-independent fine tuning algorithm framework, and performing inner loop rapid adaptation and outer loop gradient update on a plurality of task clusters; deploying the meta learning parameter tuning unit after training to a fusion analysis module, and butting an input interface of the meta learning parameter tuning unit with the output end of a task load modeling submodule; Carrying out embedded coding processing on the input comprehensive load index at the current moment, and mapping the comprehensive load index into a load representation vector in a high-dimensional semantic space by using a preset nonlinear transformation function; based on the load characterization vector, forward reasoning is executed in the meta-learning parameter tuning unit, and a discrimination boundary adjustment vector is generated.
8. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 7, wherein each sample in the basic training set for training the meta-learning parameter tuning unit comprises a comprehensive load index under a corresponding task scene, a physiological signal feature vector and a behavior state feature vector which are synchronously collected, and a fatigue grade label calibrated by an expert.
9. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 1, wherein the step S4 specifically comprises: Acquiring a discrimination boundary adjustment vector output by a parameter tuning unit, performing parameter mapping analysis operation based on the discrimination boundary adjustment vector, and respectively associating each dimension component to five types of adjustable parameter sets of classification threshold offset, heart rate variability characteristic weight gain coefficient, eye movement frequency sensitivity adjustment factor, electroencephalogram alpha wave power normalization baseline bias item and control input frequency fusion weight proportion by utilizing a predefined parameter index mapping table; Performing nonlinear transformation processing on the classification threshold offset, mapping the classification threshold offset into a preset dynamic adjustment interval, obtaining a classification boundary offset parameter after amplitude limiting, and performing pull-down correction on a fatigue state discrimination threshold of an original fatigue classifier based on the classification boundary offset parameter; Performing product coupling operation on the heart rate variability characteristic weight gain coefficient and the eye movement frequency sensitivity adjustment factor to generate a physiological-behavior bimodal characteristic enhancement factor, inputting the physiological-behavior bimodal characteristic enhancement factor into a characteristic weight distribution module, and performing affine transformation on original characteristic weight distribution to obtain an updated characteristic weight vector; Performing background drift compensation processing on original electroencephalogram signal characteristics based on the electroencephalogram alpha wave power normalization baseline bias term, calculating the deviation amount of current alpha wave band energy relative to an individual reference level, extracting trend components, overlapping the trend components to the bias term input end of a classifier, and dynamically correcting the neural fatigue baseline drift phenomenon caused by long-term task accumulation; and integrating all corrected classification threshold values, feature weighting vectors and bias term parameters to construct a real-time fatigue judgment sub-model with the current task context awareness capability.
10. The fatigue warning method for fusing driver physiological signals and unmanned aerial vehicle flight state data according to claim 1, wherein the step S5 further comprises performing time sequence bidirectional coding on the standardized feature vector set by adopting a bidirectional long-short-term memory network enhanced based on an attention mechanism, focusing on key fatigue related segments by using attention weights, outputting high discriminant time sequence characterization, outputting continuous fatigue confidence level of [0,1] intervals through full connection and Sigmoid transformation, and mapping the segments into low, medium and high-level intervals by referring to a historical verification data set segment.

Description

Fatigue warning method integrating driver physiological signals and unmanned aerial vehicle flight state data Technical Field The invention relates to the technical field of intelligent man-machine interaction and flight task state sensing, in particular to a fatigue warning method for fusing physiological signals of a driver and flight state data of an unmanned aerial vehicle. Background At present, the fatigue monitoring and alarming technology of the unmanned aerial vehicle driver mainly surrounds physiological signal analysis and operation behavior feature extraction, and fatigue states of the driver are automatically identified and classified through a multi-mode fusion model. The main flow scheme generally adopts a fatigue judgment model set by static parameters, namely, a judgment threshold value and key feature weight are set according to a set task scene and a general physiological response mode of a driver group when the system is deployed. For example, part of technologies are used for establishing a fixed fusion model based on a support vector machine, a random forest or a multi-layer neural network by collecting heart rate variability, eye movement frequency and statistical characteristics of brain electrical signals and combining flight operation frequency and behavior event count, so that preliminary classification of low, medium and high fatigue grades is realized. The scheme has a certain identification accuracy under the tasks of standardization and high consistency, and has been applied to the fields of intelligent man-machine interaction, flight safety monitoring and the like. In recent years, with the rapid increase of the task complexity and diversity of unmanned aerial vehicles, the development trend of technology gradually tends to merge task load dynamic changes and individual difference responses into a multi-level fatigue judgment system. There have been studies to quantitatively model factors such as task type (e.g., scout, hit, formation), task urgency, operational density, etc., and attempt to introduce them as auxiliary criteria into the fatigue recognition process. The static modeling is still used as a core in a few technical schemes, and real-time adjustment of model parameters according to task states and load changes is difficult to realize, so that the problems of fatigue misjudgment, missed judgment, reduced adaptability and the like often occur in complex scenes; The typical prior art generally adopts the following modes that a fatigue judgment logic rule is preset, physiological signal abnormal threshold values and operation behavior weights are statically configured, and part of schemes support simple compensation based on a historical behavior mode, but the whole framework does not have task load sensitivity and on-line parameter tuning capability. The technology is suitable for standardized flight scenes and collective driver average models, and is difficult to meet the requirement for individual fatigue dynamic identification under a changeable task state. When the unmanned aerial vehicle executes a high-intensity or time-efficiency burst task, the physiological and behavioral response modes of the driver may deviate from the basic model significantly, and the fixed threshold judgment mechanism cannot capture the risk state in time, so that the robustness and safety of the monitoring system are affected. Disclosure of Invention The invention aims to solve the technical problems and provides a fatigue warning method for fusing physiological signals of a driver and flight state data of an unmanned aerial vehicle. The technical scheme of the invention is realized by a fatigue warning method integrating physiological signals of a driver and flight state data of an unmanned aerial vehicle, comprising the following steps of: S1, acquiring physiological signal data of a driver and unmanned aerial vehicle flight state data, wherein the physiological signal data comprise heart rate variability, eye movement frequency and electroencephalogram alpha wave power, the flight state data comprise control input frequency, route change times, target locking quantity and communication interruption frequency, and all the data are aligned and marked according to a uniform time stamp; S2, calculating the comprehensive load index of the current task by utilizing a rule engine in combination with a lightweight neural network model based on a preset task type knowledge base and flight state data, wherein the index reflects the superposition effect of the task in three dimensions of physical complexity, decision pressure and time urgency and is output as a dynamically optimized regulating factor; S3, inputting the comprehensive load index into a parameter tuning unit of a fatigue judgment model, constructing the tuning unit based on a model-independent fine tuning idea under a meta-learning framework, obtaining basic parameter space adaptive capacity through historical multitask-fatigue sa