A cognitive function evaluation method and system suitable for a plateau environment

Abstract

The application relates to a cognitive function evaluation method and system suitable for a plateau environment, belonging to the field of cognitive function evaluation and physiological signal monitoring, and comprising the following steps: collecting original cognitive behavior data of a test subject during the execution of a preset cognitive task in a plateau environment, collecting original physiological data and respectively performing pretreatment to obtain standardized cognitive characteristic values and physiological characteristic values; performing weighted fusion on cognitive sub-indices and physiological sub-indices constructed based on the standardized cognitive characteristic values and the physiological characteristic values to obtain a comprehensive cognitive index; wherein the weights of the cognitive sub-indices and the physiological sub-indices are adjusted in real time by using environment parameters of the plateau environment acquired in real time; the standardized cognitive characteristic values and the physiological characteristic values in a time sequence window are combined into a multi-modal feature vector sequence which is input into a pre-trained individual GRU model of the test subject to obtain a cognitive function abnormality probability; and based on the cognitive function abnormality probability and the comprehensive cognitive index, a cognitive function evaluation result of the test subject in the plateau environment is obtained.

Description

Technical Field

[0001] This invention relates to the field of cognitive function assessment and physiological signal monitoring technology, and in particular to a cognitive function assessment method and system suitable for high-altitude environments. Background Technology

[0002] High altitudes and thin air in plateau regions significantly reduce oxygen partial pressure, impacting physiological and cognitive functions. Studies show that individuals migrating from plains to plateaus, as well as long-term residents, may experience varying degrees of decline in cognitive abilities such as attention, memory, and executive function. This decline not only reduces learning and work efficiency but can also lead to operational errors and safety hazards during high-intensity work. Therefore, developing cognitive function assessment technologies suitable for plateau environments is crucial for health monitoring of plateau workers, prevention of mental fatigue, and evaluation of the effectiveness of hypoxia-resistant interventions.

[0003] Current cognitive function assessments typically employ standardized psychological paradigms such as the Stroop color-word task and the N-back working memory task, quantifying cognitive dimensions like attentional control and working memory through behavioral indicators such as reaction time and accuracy. However, these assessment methods usually rely on independent testing terminals and offline data analysis processes. There is a lack of system integration between components for field applications, and a complete technical solution has not yet been developed for hardware and algorithm synergy and environmental adaptability design, making it difficult to operate independently and reliably in complex environments such as high altitudes.

[0004] On the other hand, existing physiological monitoring equipment can acquire blood oxygen saturation and pulse wave information through photoplethysmography (PPG) sensors, and heart rate and heart rate variability through electrocardiogram (ECG) sensors. These physiological signals are sensitive to changes in the body's state under hypoxic exposure, but relying solely on physiological parameters is insufficient to directly reflect changes in an individual's cognitive functions such as attention and memory, and it is also difficult to determine whether physiological abnormalities have affected performance on specific cognitive tasks. Therefore, a single physiological monitoring device is insufficient for a comprehensive assessment of cognitive function in high-altitude environments.

[0005] Some existing systems attempt to combine cognitive tasks with physiological monitoring, but they still have the following shortcomings in high-altitude field applications: First, the ability to analyze the correlation between cognitive behavioral data and physiological data is limited, usually only presenting the independent indicators of each, lacking end-side real-time fusion analysis methods for cognitive performance and physiological characteristics such as blood oxygen and heart rate variability; Second, data analysis mostly relies on offline statistics or upper-level computer processing after the test is completed, lacking the ability to process the identification of cognitive function abnormalities under hypoxia conditions in real time, making it difficult to trigger early warnings when abnormal indicators appear; Third, the overall power consumption management and environmental adaptability design of the equipment have not been optimized for high-altitude field scenarios, and there are still shortcomings in terms of long battery life and wide temperature range operation. Summary of the Invention

[0006] Based on the above analysis, the embodiments of the present invention aim to provide a cognitive function assessment method and system suitable for high-altitude environments, in order to solve the technical problem that the prior art lacks an integrated portable cognitive function assessment method and system specifically designed for high-altitude hypoxic environments, capable of simultaneously collecting cognitive behavior and multimodal physiological signals and performing real-time fusion analysis and low-power warning at the edge.

[0007] The objective of this invention is mainly achieved through the following technical solutions: This invention provides a cognitive function assessment method suitable for high-altitude environments, comprising the following steps: Raw cognitive behavioral data of subjects performing a pre-set cognitive task in a high-altitude environment were collected, and raw physiological data were collected simultaneously and preprocessed to obtain standardized cognitive and physiological characteristic values. The cognitive sub-index and physiological sub-index, constructed based on the standardized cognitive and physiological feature values, are weighted and fused to obtain a comprehensive cognitive index; wherein, the weights of the cognitive sub-index and physiological sub-index are adjusted in real time using environmental parameters of the plateau environment acquired in real time. The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into the pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction. Based on the probability of cognitive dysfunction and the comprehensive cognitive index, the cognitive function assessment results of the subjects in the high-altitude environment were obtained.

[0008] Furthermore, the preset cognitive tasks include the Stroop color-word interference task and the N-back working memory task; the raw cognitive behavioral data include reaction time, answer accuracy, error rate, hit rate, false negative rate, Stroop interference effect, and N-back working memory load index. The raw physiological data includes photoplethysmography (PPG) signals and electrocardiogram (ECG) signals; the standardized physiological features obtained after preprocessing include mean blood oxygen saturation, rate of change of blood oxygen saturation, mean and rate of change of heart rate, SDNN, RMSSD, and pNN50 heart rate variability index; wherein, the PPG signals are acquired using a PPG sensor.

[0009] Furthermore, the raw cognitive behavioral data and raw physiological data are preprocessed, including: The original cognitive behavioral data was filtered out by removing invalid trials that exceeded the individual baseline reaction time of the subjects by ±3 times the standard deviation or violated the task rules, to obtain the first cognitive behavioral data; The photoplethysmography (PPG) wave signal in the original physiological data is subjected to moving average filtering, and the electrocardiogram (ECG) signal is subjected to bandpass filtering and notch filtering to obtain the first physiological data. The first cognitive behavioral data and the first physiological data are timestamped and then Z-score standardized to obtain standardized cognitive feature values ​​and physiological feature values.

[0010] Furthermore, the cognitive sub-indices and physiological sub-indices constructed based on the standardized cognitive and physiological feature values ​​are weighted and fused to obtain a comprehensive cognitive index, including: The cognitive sub-index is calculated by weighted summation or weighted average of each standardized cognitive feature value. The physiological sub-index is calculated by weighted summation or weighted average of each standardized physiological characteristic value; The cognitive sub-index and the physiological sub-index are weighted and fused as follows:

[0011] in, As a comprehensive cognitive index, For cognitive sub-index, Physiological sub-index These are the weights of the cognitive sub-index and the physiological sub-index, respectively.

[0012] Furthermore, the environmental parameters of the plateau environment include ambient air pressure and ambient temperature; the ambient air pressure is obtained using an air pressure sensor, and the ambient temperature is obtained using a temperature sensor. The weights of the cognitive and physiological sub-indices are adjusted in real time using environmental parameters of the plateau environment acquired in real time, including: The current altitude is calculated based on the ambient air pressure, and the blood oxygen abnormality judgment threshold is corrected based on the altitude. The ambient temperature is obtained using a temperature sensor, and the LED light intensity and photodetector response drift of the PPG sensor are compensated based on the ambient temperature to ensure the accuracy of blood oxygen saturation measurement. The blood oxygen saturation of the subjects is detected in real time, and the blood oxygen saturation is compared with the blood oxygen abnormality judgment threshold after air pressure correction. If the blood oxygen saturation is lower than the blood oxygen abnormality judgment threshold after pressure correction, the weight of the physiological sub-index is increased based on the degree of deviation of blood oxygen saturation. At the same time, the weight of the cognitive sub-index is reduced accordingly. And the weight of the physiological sub-index The upper limit shall not exceed the weight threshold; If the blood oxygen saturation is greater than or equal to the blood oxygen abnormality judgment threshold after pressure correction, the weights of the physiological sub-index and the cognitive sub-index are maintained at the default values ​​and no weight adjustment is made.

[0013] Furthermore, the standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into a pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction, including: A fixed-length sliding time window is set, and the window sliding step size is less than the total window length to achieve overlapping sampling of time series data; wherein, the time series window length and sliding step size are preset based on the data acquisition frequency; Based on the standardized cognitive and physiological feature values ​​collected synchronously within each time window, the dimensions are concatenated to generate a single-step multimodal feature vector; all single-step feature vectors within the window are arranged in chronological order of data collection time to form the first multimodal feature vector sequence. The first multimodal feature vector sequence is subjected to unified processing of dimension padding and truncation to obtain the second multimodal feature vector sequence, so that the dimension of the input sequence matches the input layer size of the pre-trained GRU model; The second multimodal feature vector sequence is input into the pre-trained GRU model to obtain the preliminary probability of cognitive function abnormality. The initial cognitive function abnormality probability output by the current window is taken and weighted averaged with the initial cognitive function abnormality probabilities output by the previous two consecutive time windows to eliminate probability jumps caused by fluctuations in single-frame data, and the smoothed cognitive function abnormality probability is obtained as the cognitive function abnormality probability of the subject.

[0014] Furthermore, based on the aforementioned probability of cognitive dysfunction and the comprehensive cognitive index, the cognitive function assessment results of the subjects in the high-altitude environment are obtained, including: The smoothed probability of cognitive dysfunction is fused with the comprehensive cognitive index for determination; among which... When one of the following conditions is met, the subject is judged to have cognitive dysfunction and a graded warning is triggered: Condition 1: The smoothed probability of cognitive function abnormality exceeds a preset probability threshold and continues to exceed a preset number of consecutive time windows; Condition 2: The comprehensive cognitive index is lower than the preset index threshold; Condition 3: The weighted combination of the smoothed cognitive function abnormality probability and the comprehensive cognitive index exceeds a preset comprehensive threshold; If none of the above conditions are met, the subject is judged to have normal cognitive function in a high-altitude environment.

[0015] Furthermore, based on the smoothed probability of cognitive dysfunction With comprehensive cognitive index The following rules are used to issue warnings for three levels of cognitive impairment: mild cognitive impairment, moderate cognitive impairment, and severe cognitive impairment: like And the duration exceeds A timing window, or ,or and weighted combination value and If so, a warning of mild cognitive abnormality will be issued; like And the duration exceeds A timing window, or ,or and weighted combination value and Or, a warning of mild cognitive abnormalities persisting for more than If the condition is not resolved within minutes, a warning for moderate cognitive abnormality will be issued. like And the duration exceeds A timing window, or ,or and weighted combination value Or, a moderate cognitive abnormality warning lasting longer than If the condition is not resolved within minutes, a warning of severe cognitive abnormality will be issued. If continuous Within each timing window, and and The current warning will be automatically lifted, and the system will return to normal cognitive status. in, , , , , All of these are preset threshold values.

[0016] Furthermore, the pre-trained individual GRU model of the subject is obtained through the following process; Step 1: Basic Model Pre-training A general training sample set was collected, including: raw cognitive behavioral data collected from multiple subjects in high-altitude and low-altitude environments during the execution of a preset cognitive task, and raw physiological data collected simultaneously and preprocessed separately to form a multimodal feature vector sequence; each sample is a multimodal feature vector sequence of 30 to 60 consecutive time windows, and the feature vector of each time window is composed of standardized cognitive feature values ​​and physiological feature values; Label each training sample with a cognitive abnormality tag; The GRU model is pre-trained using binary cross-entropy as the loss function and the Adam optimization algorithm with an initial learning rate to obtain the basic GRU model. Step 2: Individual adaptive fine-tuning of the subject based on the basic GRU model: Individual baseline data of the subjects were collected before the formal test. The individual baseline data included cognitive and behavioral data of the subjects in a preset number of practice trials and physiological data collected simultaneously. The individual baseline data is preprocessed and features are extracted to obtain individual standardized parameters and individual multimodal feature vector sequences for multiple time windows. The individual multimodal feature vector sequences are used to fine-tune the basic GRU model, freezing the underlying parameters of the basic GRU model and updating only the parameters of the top fully connected layer, or updating all parameters in a few rounds with a low learning rate to obtain an individual GRU model adapted to the current subject. Step 3: Deployment of Individual GRU Models The individual GRU model is pruned and quantized with INT8 and then deployed on an embedded device to output the subject's cognitive abnormality probability in real time.

[0017] This invention also discloses a cognitive function assessment system suitable for high-altitude environments, comprising: The cognitive monitoring module is used to collect raw cognitive behavioral data of subjects performing preset cognitive tasks in a high-altitude environment; The physiological monitoring module is used to collect raw physiological data from the subjects; Environmental sensors, including barometric pressure sensors and temperature sensors, are used to collect high-altitude environmental parameters in real time. The main control processing unit is responsible for task scheduling, multimodal data timestamp synchronization, raw data filtering and noise reduction, feature extraction, time-series window truncation, and model inference scheduling. The data analysis module is used to fuse cognitive behavioral data and physiological data and to identify intelligent anomalies. It includes a feature construction unit, a weighted fusion unit, and a lightweight temporal discrimination unit. The lightweight temporal discrimination unit adopts a GRU model after pruning and INT8 quantization, with 32 hidden units. Used to deploy individual GRU models of subjects after pruning and INT8 quantization, to complete the standardization, weighted fusion, comprehensive cognitive index calculation, temporal feature sequence inference, and cognitive function abnormality probability calculation of multimodal raw data; The alert module, including a buzzer and indicator lights, is used to issue an audible and visual alarm when a graded warning is triggered.

[0018] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. In terms of integrated portability and adaptability to high-altitude environments, this invention is the first to integrate standardized cognitive task presentation, simultaneous acquisition of multimodal physiological signals, edge-side intelligent fusion analysis, and dynamic power consumption management into a single portable hardware device, overcoming the shortcomings of existing technologies where components are separate and lack system integration. By using a barometric pressure sensor to correct the blood oxygen anomaly judgment threshold and a temperature sensor to compensate for PPG measurement drift, combined with a dual-material protective shell of polycarbonate reinforced alloy and aluminum-magnesium alloy, the device can operate stably for extended periods in harsh high-altitude environments ranging from -20℃ to +50℃, with low oxygen, strong light, and low temperatures. The entire device weighs less than 1000g and has a battery life exceeding 72 hours, truly achieving independent and reliable use in high-altitude environments. 2. In terms of intelligent fusion and real-time early warning, this invention constructs cognitive and physiological sub-indices and performs dynamic weighted fusion. When blood oxygen saturation is below the hypoxia threshold, the weight of the physiological sub-index is automatically increased (up to 0.75), enabling the comprehensive assessment to sensitively reflect the impact of hypoxia on cognitive function. Simultaneously, multimodal temporal feature sequences are input into a lightweight GRU model after pruning and INT8 quantization, outputting the probability of cognitive function abnormalities every 30 seconds. This probability is then combined with the comprehensive cognitive index for multi-condition fusion judgment and graded early warning. Compared to existing technologies that rely on offline statistics and independent indicator judgment, this invention achieves real-time fusion analysis and anomaly detection on the edge, triggering early warnings in the early stages of cognitive function abnormalities, significantly reducing the false negative and false positive rates. 3. Regarding individual adaptation and low-power deployment: This invention employs a strategy of "basic pre-training + individual fine-tuning" to obtain an individual GRU model adapted to the subject. Adaptive fine-tuning is performed using individual baseline data from practice trials, significantly improving the accuracy of individual cognitive function discrimination. Simultaneously, through dynamic voltage and frequency adjustment, time-sharing power supply, and deep sleep mechanisms, combined with pruning and INT8 quantization for lightweight model deployment, the device can achieve a battery life exceeding 72 hours under typical testing conditions. Compared with existing technologies, this invention ensures both low-power real-time inference capabilities on the embedded device and model adaptation to individual differences, providing an accurate, reliable, and portable solution for cognitive health monitoring of personnel working in high-altitude areas.

[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0021] Figure 1 This is a flowchart of a cognitive function assessment method suitable for high-altitude environments, as described in an embodiment of the present invention. Figure 2a This is a front view diagram of the hardware device for cognitive function assessment in an embodiment of the present invention. Figure 2b This is a schematic diagram of the internal structure of the hardware device for cognitive function assessment in an embodiment of the present invention; Figure 3 This is a flowchart of the cognitive task in an embodiment of the present invention; Figure 4 This is a schematic diagram of the physiological detection module structure in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the multimodal data acquisition, fusion processing, and cognitive function assessment process in an embodiment of the present invention.

[0022] Figure label: 1- Protective housing for the hardware device for cognitive function assessment; 2- Touchscreen display; 21a, 21b- Yes / No buttons for answering cognitive tasks; 22a, 22b, 22c, 22d- Red / Yellow / Blue / Green buttons for answering cognitive tasks; 31- Microcontroller; 32- Storage unit; 33- Wireless communication unit; 4- Power management module; 51- PPG sensor; 52- Electrocardiogram sensor. Detailed Implementation

[0023] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0024] To address the aforementioned issues, current technologies lack a cognitive function assessment method and system suitable for high-altitude environments. A portable, integrated cognitive function testing system specifically designed for the complex high-altitude environment can integrate standardized cognitive task presentation, simultaneous acquisition of multimodal physiological signals, edge-side intelligent fusion analysis, and dynamic power consumption management. It can provide intelligent fusion analysis and real-time risk warning, and achieve cognitive function assessment and monitoring in high-altitude environments under conditions such as low oxygen, low temperature, and strong light. This will provide cognitive function assessment and safety assurance for high-altitude workers, researchers, and medical institutions.

[0025] Example 1: A specific embodiment of the present invention discloses a cognitive function assessment system suitable for high-altitude environments, comprising: The cognitive monitoring module is used to collect raw cognitive behavioral data of subjects performing preset cognitive tasks in a high-altitude environment; The physiological monitoring module is used to collect raw physiological data from the subjects; Environmental sensors, including barometric pressure sensors and temperature sensors, are used to collect high-altitude environmental parameters in real time. The main control processing unit, namely the microprocessor 31, is responsible for the task scheduling of the hardware device for cognitive function assessment, multimodal data timestamp synchronization, raw data filtering and noise reduction, feature extraction, time window truncation and model inference scheduling. The data analysis module is used to fuse cognitive behavioral data and physiological data and to identify intelligent anomalies. It includes a feature construction unit, a weighted fusion unit, and a lightweight temporal discrimination unit. The lightweight temporal discrimination unit adopts a GRU model after pruning and INT8 quantization, with 32 hidden units. This module is used to deploy individual GRU models of subjects after pruning and INT8 quantization to complete the standardization, weighted fusion, comprehensive cognitive index calculation, temporal feature sequence inference, and cognitive function abnormality probability calculation of multimodal raw data; the data analysis module is an algorithm program running on a microcontroller. The alert module, including a buzzer and indicator lights, is used to issue an audible and visual alarm when a graded warning is triggered.

[0026] The hardware device for cognitive function assessment adopts a portable, integrated design. Structurally, it integrates a cognitive testing module, a physiological monitoring module, a main control processing unit, a power management module, and a data analysis module. Combined with low-power control strategies and intelligent analysis algorithms, it realizes real-time assessment, anomaly detection, and result visualization of cognitive function in high-altitude environments.

[0027] The cognitive function assessment hardware device in this invention adopts a dual-material structure with a polycarbonate-reinforced alloy shell and an aluminum-magnesium alloy internal skeleton. The outer layer provides impact resistance, heat insulation, and electrical insulation, while the inner layer enhances structural strength, heat dissipation, and electromagnetic shielding capabilities. The overall dimensions are approximately 200×120×40 mm, and the weight does not exceed 1000 g. It can operate stably for extended periods in high-altitude environments ranging from -20°C to +50°C, and is characterized by portability, robustness, and strong environmental adaptability, making it suitable for use in high-altitude field operations.

[0028] like Figure 2a and Figure 2bAs shown, the hardware device for cognitive function assessment adopts a portable, integrated design, with internal components encapsulated in a protective housing 1. A touchscreen display 2 is embedded in the front of the housing 1 and connected to the main control circuit board via a display cable and touch signal cable, used to present cognitive tasks and transmit response signals. The main control circuit board is located inside the housing and integrates a microcontroller 31, a storage unit 32, and a wireless communication unit 33. The microcontroller 31 is connected to the touchscreen display 2, storage unit 32, wireless communication unit 33, power management module 4, and physiological sensors (including a PPG sensor 51 and an ECG sensor 52), providing unified control over task execution, data acquisition, timestamp synchronization, feature extraction, and result output. The power management module 4 supplies power to each component through a power management circuit and switches power supply states according to the main control commands.

[0029] The hardware device for cognitive function assessment also includes an environmental sensor and a prompting module connected to the main control processing unit. The environmental sensor is used to collect environmental parameters such as air pressure and temperature in the high-altitude environment; the prompting module includes a buzzer and an indicator light, which are used to issue an audible and visual alarm when a graded warning is triggered.

[0030] ① The cognitive monitoring module is used to present cognitive tasks to subjects and record behavioral response data. It has built-in Stroop color-word interference task and N-back working memory task. Stimulus information is presented through a 5-inch high-brightness touch screen 2. The buttons adopt a symmetrical layout, namely 21a, 21b, and 22a-22d. Subjects can answer while wearing gloves. The main control processing unit records the reaction time, accuracy and other indicators in real time to ensure stable operation under strong light and low temperature conditions.

[0031] The cognitive monitoring module is a combination of hardware and software. Its task operation logic and response recording are executed by the algorithm program on the microcontroller 31, while stimulus presentation and response input rely on the touch screen 2 and the hardware components of buttons 21a, 21b, 22a, 22b, 22c, and 22d.

[0032] ② Physiological monitoring module, used to collect the subject's pulse waveform and electrocardiogram signal, and calculate physiological parameters such as blood oxygen saturation, heart rate and heart rate variability based on these.

[0033] The physiological monitoring module includes a PPG sensor 51 and an electrocardiogram sensor 52, which are connected to the main control circuit board via the sensor interface. The pulse wave and electrocardiogram signal are transmitted to the microcontroller 31 and stored synchronously with the original cognitive behavioral data and original physiological data of the cognitive task in the storage unit 32 for data analysis.

[0034] The PPG sensor is a clip-on or finger sleeve structure, worn on the fingertip of the index, middle or ring finger of the subject. It uses a dual-wavelength light source of 660nm red light and 940nm infrared light and a photodetector to collect pulse wave signals, and calculates blood oxygen saturation based on the changes in the absorption intensity of red light and infrared light. The ECG sensor uses patch electrodes, which are worn on the chest or limb leads of the subject to collect electrocardiogram signals. It uses the ADS1292R ECG acquisition chip to acquire ECG signals at a sampling rate of 250-500Hz. Figure 4 The structure and signal processing flow of the physiological monitoring module of this invention are illustrated. Specifically, the PPG sensor 51 alternately illuminates the subject's fingertip with 660nm red light and 940nm infrared light from an LED. A photodetector converts the light absorption fluctuations caused by changes in blood volume into a pulse wave analog signal, which is used to calculate blood oxygen saturation. The ECG sensor 52 collects surface electrocardiogram signals through patch electrodes, which are then amplified and conditioned by the ADS1292R chip and output to extract the R wave and RR interval and calculate heart rate (HR) and heart rate variability (HRV).

[0035] The two signals are amplified, filtered, and noise suppressed by the preamplifier circuit, and then input to the microcontroller 31 for digital sampling, downsampling, bandpass and notch filtering, and feature extraction. The output is... Physiological characteristics such as HR and HRV are recorded. The microcontroller 31 adds timestamps to all data, synchronizes and aligns the physiological characteristics, waveform data and behavioral data such as reaction time and accuracy generated by the cognitive test module, and stores them together in the storage unit 32, and uploads them to an external terminal or remote platform through the communication unit 33.

[0036] In the above structure, the PPG sensor 51, ECG sensor 52, and pre-amplifier circuit constitute the physiological signal acquisition and conditioning front end. The main control unit 31 is the core part of the main control processing unit. The storage unit 32 and communication unit 33 are responsible for data storage and transmission. The power management module 4 supplies power to each unit and controls the sampling rate and system sleep / wake-up according to the task status. The data analysis module retrieves... Physiological characteristics such as HR and HRV are integrated with cognitive characteristics for analysis and anomaly detection.

[0037] ③ Environmental sensor: An environmental sensor connected to the microcontroller 31 is also installed inside the protective housing 1. The environmental sensor is used to collect air pressure and temperature data in the high-altitude environment.

[0038] ④ The main control processing unit, namely the microprocessor 31, adopts a low-power microcontroller based on the ARM Cortex-M4F core. The operating frequency is dynamically adjustable between 20-120MHz. During the cognitive task operation and fusion analysis stage, it can operate at 80-120MHz to balance computing power and power consumption. It has no less than 1MB Flash and 256KB SRAM built-in for cognitive task program operation, physiological signal acquisition and control, data caching and feature extraction.

[0039] The main control processing unit supports a multi-channel ADC (Analog-to-Digital Converter) interface to receive PPG signals and ECG signals; and supports a USB-C interface and a wireless communication unit 33 for exporting or transmitting detection data, analysis results and device configuration parameters. The ECG signal undergoes 0.5-40Hz bandpass filtering and 50Hz notch filtering by the main control processing unit to reduce baseline drift, motion artifacts, electromyographic interference, and power line interference. After enhancing the QRS complex and R wave peak characteristics, R wave detection, RR interval sequence extraction, outlier removal, and artifact correction are performed. Then, heart rate (HR) and heart rate variability (HRV) parameters such as SDNN (Standard Deviation of Normal-to-Normal intervals), RMSSD (Root Mean Square of Successive Differences), and pNN50 (Proportion of NN50 divided by total number of NN intervals) are calculated. Meanwhile, the main control processing unit is connected to the power management module 4, and switches the working mode, reduces the main frequency, or enters sleep mode according to the task running status.

[0040] ⑤ The data analysis module is used for fusing cognitive behavioral data and physiological data and for intelligent anomaly detection. It includes a feature construction unit, a weighted fusion unit, and a lightweight temporal discrimination unit. The data analysis module is an algorithm program running on a microcontroller.

[0041] ⑥ Prompt module, including a buzzer or indicator light.

[0042] ⑦ Power management module 4 is equipped with a high-energy-density lithium battery, preferably with a capacity of 6000mAh, supports 5V DC charging and solar-assisted power supply, and supports external solar panels for auxiliary charging to meet the needs of field use. It features dynamic voltage and frequency regulation, time-sharing power supply, and sleep / wake-up mechanisms. Dynamic voltage and frequency regulation is hierarchically controlled based on the CPU utilization and task status of the main control unit: through Dynamic Voltage and Frequency Regulation (DVFS) technology, during cognitive task operation and fusion analysis, the main control unit's operating frequency can be dynamically adjusted within the range of 20-120MHz, with the main frequency reaching a maximum of 120MHz and the core voltage increasing to 1.3V; during physiological data acquisition and feature extraction, the main frequency is adjusted to 48-80MHz and the core voltage to 1.0-1.2V; during data storage and transmission, the main frequency drops to 20MHz and the core voltage drops to 0.9V.

[0043] The time-sharing power supply strategy adjusts the sensor sampling rate accordingly: during the cognitive testing phase, the PPG sampling rate is 50-250Hz, and the ECG sampling rate is 250-500Hz; during test intervals, the PPG sampling rate drops to 25-100Hz, and the ECG sampling rate drops to 125-250Hz; during idle periods, only low-frequency sampling is maintained or unnecessary sensors are turned off. Non-working peripherals are shut down via clock gating, and the device automatically enters deep sleep mode after more than 60 seconds of idle time, maintaining only the real-time clock, interrupt detection, and data retention; the cognitive function assessment hardware can be woken up within 50ms after sensor input or external interruption. This strategy enables the device to operate for over 72 hours in high-altitude environments.

[0044] The power management module integrates battery management and voltage regulation circuits to achieve charge and discharge management, overcharge / over-discharge protection, and maintain battery performance in low-temperature environments. To extend battery life, multiple low-power strategies are employed: after a test idle time exceeding a preset period (e.g., 60 seconds), the main control processing unit automatically enters deep sleep, reducing the overall deep sleep current to approximately 10µA, retaining only the real-time clock and interrupt wake-up functions; touchscreen operation or external interrupt signals can wake the system within tens of milliseconds.

[0045] The core voltage can be adjusted within the range of 0.9-1.3V to optimize power consumption according to real-time computing load; the peripheral power supply voltage can be set individually according to device requirements.

[0046] ⑧ Storage unit 32, with a capacity of 1MB Flash and 256KB SRAM, is used to store evaluation algorithm programs and cache cognitive and physiological data.

[0047] ⑨ Wireless communication unit 33 is equipped with a BLE (Bluetooth Low Energy) module for data synchronization and remote configuration.

[0048] ⑩ Touch screen 2 is located on the front of the protective cover 1. The touch screen 2 is about 5 inches in size, with a resolution of 800×480. It supports multi-touch and high brightness mode, and has a glove touch function to ensure clear operation in high-altitude and strong light environments or when wearing gloves.

[0049] The protective casing 1 employs a dual-material structure: a polycarbonate-reinforced alloy outer shell and an aluminum-magnesium alloy frame. The outer shell uses a polycarbonate-reinforced alloy to ensure excellent impact resistance, heat insulation, and electrical insulation properties. The internal load-bearing and heat dissipation frame uses an aluminum-magnesium alloy to enhance overall structural strength, heat dissipation, and electromagnetic shielding capabilities. This dual-material design allows for long-term stable operation in harsh high-altitude environments ranging from -20℃ to +50℃, balancing lightweight design, strength, and environmental adaptability. The overall dimensions of the cognitive function assessment hardware are approximately 200mm × 120mm × 40mm, with a weight controlled below 1000g, making it easy to hold with one hand or place on a desktop.

[0050] Example 2: A specific embodiment of the present invention, such as Figure 1 As shown, a cognitive function assessment method suitable for high-altitude environments is disclosed, comprising the following steps: Step S1: Collect raw cognitive behavioral data of the subjects during the performance of the preset cognitive task in the plateau environment, and simultaneously collect raw physiological data and preprocess them to obtain standardized cognitive feature values ​​and physiological feature values. Step S2: The cognitive sub-index and physiological sub-index constructed based on the standardized cognitive feature values ​​and physiological feature values ​​are weighted and fused to obtain a comprehensive cognitive index; wherein, the weights of the cognitive sub-index and physiological sub-index are adjusted in real time using environmental parameters of the plateau environment acquired in real time. Step S3: The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into the pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction. Step S4: Based on the probability of cognitive function abnormality and the comprehensive cognitive index, obtain the cognitive function assessment results of the subjects in the plateau environment.

[0051] Step S1, specifically.

[0052] The preset cognitive tasks include the Stroop color-word interference task and the N-back working memory task; the raw cognitive behavior data include reaction time, answer accuracy, error rate, hit rate, false negative rate, Stroop interference effect, and N-back working memory load index. The raw physiological data includes photoplethysmography (PPG) signals and electrocardiogram (ECG) signals; the standardized physiological features obtained after preprocessing include mean blood oxygen saturation, rate of change of blood oxygen saturation, mean and rate of change of heart rate, SDNN, RMSSD, and pNN50 heart rate variability index; wherein, the PPG signals are acquired using a PPG sensor.

[0053] like Figure 3 As shown, during the cognitive test, the touch screen 2 presents the corresponding test stimuli according to the current task flow. The answer buttons adopt a symmetrical layout, which makes it easy for the test subject to hold the hardware device for cognitive function assessment with both hands, and use the two thumbs to click the left and right buttons respectively to answer.

[0054] Taking the N-back task as an example, the central area of ​​the touchscreen display 2 sequentially displays the letter sequence to be memorized. The participant judges whether the current letter is the same as the letter presented in the nth position before it. When the participant thinks it "matches", they press button 21a (YES); otherwise, they press button 21b (NO). The system records the participant's reaction time and the correctness of their judgment after each stimulus presentation.

[0055] In the Stroop color-word task, touchscreen 2 presents colored text stimuli, such as displaying the word "green" in red. Participants must click on the corresponding option button (red, yellow, blue, green) from the red, yellow, blue, and green buttons (22a, 22b, 22c, 22d) at the bottom of the screen to respond. The main control processing unit records reaction time, accuracy, and other metrics in real time to ensure stable operation under conditions of strong light and low temperature at high altitudes. The touchscreen 2's interface design uses large font sizes and high-contrast colors for both text and buttons to ensure accurate recognition and operation even in high-altitude environments, under strong outdoor light, or when wearing gloves.

[0056] During cognitive behavioral monitoring, the physiological monitoring module simultaneously collects the subject's pulse waveform and electrocardiogram signal, and the microcontroller 31 calculates physiological parameters such as blood oxygen saturation, heart rate and heart rate variability based on these data.

[0057] We obtained the subjects' raw cognitive and behavioral data and raw physiological data.

[0058] Using a preset analysis window as a unit, one or more cognitive features are extracted from the cognitive behavioral data, including Stroop average reaction time, Stroop accuracy, Stroop error rate, Stroop hit rate, Stroop false negative rate, Stroop interference effect, N-back hit rate, and N-back working memory load index, as the subjects' raw cognitive behavioral data.

[0059] Simultaneously, blood oxygen saturation was extracted from physiological data. Mean, blood oxygen saturation One or more of the following physiological characteristics were used as raw physiological data of the subjects: rate of change, mean heart rate (HR), rate of change of heart rate (HR), SDNN, RMSSD, and pNN50.

[0060] The raw cognitive-behavioral data and raw physiological data are preprocessed, including: The original cognitive behavioral data was filtered out by removing invalid trials that exceeded the individual baseline reaction time of the subjects by ±3 times the standard deviation or violated the task rules, to obtain the first cognitive behavioral data; The photoplethysmography (PPG) wave signal in the original physiological data is subjected to moving average filtering, and the electrocardiogram (ECG) signal is subjected to bandpass filtering and notch filtering to obtain the first physiological data. The first cognitive behavioral data and the first physiological data are timestamped and then Z-score standardized to obtain standardized cognitive feature values ​​and physiological feature values.

[0061] Z-score normalization is performed on each feature as follows: Formula (1) in, For the first Raw cognitive behavioral data of each cognitive feature and These are the mean and standard deviation of the cognitive behavioral data, respectively. Standardized cognitive feature values; For the first Raw physiological data of each physiological characteristic and These are the mean and standard deviation of the physiological data, respectively. These are standardized physiological characteristic values.

[0062] The purpose of step S1 is to simultaneously collect multimodal behavioral and physiological raw data during the cognitive task execution process in a high-altitude environment, and to preprocess and standardize the data to provide high-quality, individualized calibrated feature inputs for subsequent cognitive function assessment.

[0063] Step S2, specifically.

[0064] The cognitive sub-indices and physiological sub-indices constructed based on the standardized cognitive and physiological feature values ​​are weighted and fused to obtain a comprehensive cognitive index, including: The cognitive sub-index is calculated by weighted summation or weighted average of each standardized cognitive feature value. The physiological sub-index is calculated by weighted summation or weighted average of each standardized physiological characteristic value; The cognitive sub-index and the physiological sub-index are weighted and fused as follows: Formula (2) in, As a comprehensive cognitive index, For cognitive sub-index, Physiological sub-index These are the weights of the cognitive sub-index and the physiological sub-index, respectively.

[0065] Cognitive Sub-index The physiological sub-index is obtained by weighted summation or weighted average of standardized cognitive feature values. It is obtained by weighted summation or weighted average of standardized physiological characteristic values.

[0066] Cognitive Sub-index The calculation is as follows: ①The weighted summation method is as follows: Formula (3) in, The number of cognitive features, For the first The weights of each cognitive feature satisfy... .

[0067] ② Weighted average method, as follows: Formula (4) when At that time, the two calculation methods are equivalent.

[0068] Similarly, physiological sub-index calculate: ①The weighted summation method is as follows: Formula (5) in, The number of cognitive features, For the first The weights of each cognitive feature satisfy... .

[0069] ② Weighted average method, as follows: Formula (6) The comprehensive cognitive function index is calculated as follows: Formula (7) For example, under normal conditions When blood oxygen saturation is below the preset high-altitude hypoxia threshold or shows a continuous downward trend, the concentration will be increased proportionally according to the degree of deviation. ,and The upper limit is no more than 0.75, in order to increase the weight of physiological state changes in the comprehensive assessment under hypoxic conditions.

[0070] For example, the preset high-altitude hypoxia threshold is 85%-88%, which can be dynamically adjusted according to altitude, air pressure compensation, and individual baseline.

[0071] The environmental parameters of the plateau environment include ambient air pressure and ambient temperature; the ambient air pressure is obtained using an air pressure sensor, and the ambient temperature is obtained using a temperature sensor. The weights of the cognitive and physiological sub-indices are adjusted in real time using environmental parameters of the plateau environment acquired in real time, including: The current altitude is calculated based on the ambient air pressure, and the blood oxygen abnormality judgment threshold is corrected based on the altitude. The LED light intensity and photodetector response drift of the PPG sensor are compensated based on the ambient temperature to ensure the accuracy of blood oxygen saturation measurement. The blood oxygen saturation of the subjects is detected in real time, and the blood oxygen saturation is compared with the blood oxygen abnormality judgment threshold after air pressure correction. If the blood oxygen saturation is lower than the blood oxygen abnormality judgment threshold after pressure correction, the weight of the physiological sub-index is increased based on the degree of deviation of blood oxygen saturation. At the same time, the weight of the cognitive sub-index is reduced accordingly. And the weight of the physiological sub-index The upper limit shall not exceed the weight threshold; If the blood oxygen saturation is greater than or equal to the blood oxygen abnormality judgment threshold after pressure correction, the weights of the physiological sub-index and the cognitive sub-index are maintained at the default values ​​and no weight adjustment is made.

[0072] For example, the weight threshold is 0.75.

[0073] The purpose of step S2 is to construct cognitive and physiological sub-indices by weighted fusion of standardized cognitive and physiological characteristics, and to dynamically adjust the weights of physiological sub-indices in real time using plateau environmental parameters (air pressure, temperature) to ultimately form a comprehensive cognitive index that is sensitive to the low oxygen state at high altitude.

[0074] Step S3 includes steps S31-S32.

[0075] Step S31: The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into the pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction.

[0076] The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into a pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction, including: A fixed-length sliding time window is set, and the window sliding step size is less than the total window length to achieve overlapping sampling of time series data; wherein, the time series window length and sliding step size are preset based on the data acquisition frequency; Based on the standardized cognitive and physiological feature values ​​collected synchronously within each time window, the dimensions are concatenated to generate a single-step multimodal feature vector; all single-step feature vectors within the window are arranged in chronological order of data collection time to form the first multimodal feature vector sequence. The first multimodal feature vector sequence is subjected to unified processing of dimension padding and truncation to obtain the second multimodal feature vector sequence, so that the dimension of the input sequence matches the input layer size of the pre-trained GRU model; The second multimodal feature vector sequence is input into the pre-trained GRU model to obtain the preliminary probability of cognitive function abnormality. The initial cognitive function abnormality probability output by the current window is taken and weighted averaged with the initial cognitive function abnormality probabilities output by the previous two consecutive time windows to eliminate probability jumps caused by fluctuations in single-frame data, and the smoothed cognitive function abnormality probability is obtained as the cognitive function abnormality probability of the subject.

[0077] The lightweight timing discrimination unit adopts a GRU (Gated Recurrent Unit) structure, with a preferred number of hidden units of 32. After pruning and INT8 quantization, it is deployed on the embedded end to reduce the number of model parameters, storage usage and inference power consumption.

[0078] The input to the GRU model is a sequence of 30-60 second multimodal feature vectors, with each step consisting of cognitive and physiological features within the corresponding window.

[0079] The GRU output is mapped to a cognitive function abnormality probability between 0 and 1 via a fully connected layer and a Sigmoid function, and is updated every 30 seconds. When the cognitive function abnormality probability reaches a preset threshold and persists for a preset time, the main control processing unit triggers a graded warning through the touch display screen 2 and a prompting module located inside the protective shell and connected to the main control processing unit. The prompting module includes a buzzer or indicator light, and records the abnormality probability, warning time, and related indicators to the storage unit 32.

[0080] This section describes a lightweight temporal deep learning model deployed on an embedded system for real-time assessment of whether a subject's cognitive function is abnormal. The overall process can be divided into the following six steps: Feature sequence construction → GRU model processing → Anomaly probability output → Probability update cycle → Anomaly determination → Hierarchical early warning and recording.

[0081] While pruning and INT8 quantization are existing technologies, the innovation of this invention lies in their first application to an embedded temporal discriminant model for assessing cognitive function in a high-altitude environment, and in the optimization of parameters (such as 32 hidden units, 30-60 steps window length, and INT8 per-channel quantization) for the GRU structure and real-time inference requirements.

[0082] The GRU hidden layer is preferably set to 32 units to extract cognitive and physiological temporal changes in a continuous window with low computational cost, making it suitable for embedded deployment.

[0083] The model training samples are multimodal time-series feature sequences collected in plateau and lowland environments, with labels including: mild anomaly, moderate anomaly, and severe anomaly; The loss function used is binary classification cross-entropy, the optimization algorithm is Adam, and the optimal learning rate is [missing value]. The maximum number of training rounds is 1000.

[0084] The early stopping strategy is to stop training when the validation set loss does not decrease for 20 consecutive rounds.

[0085] The features are updated and inference is performed every 30 seconds during runtime.

[0086] Preferably, when the anomaly probability is greater than 0.70 and two consecutive analysis windows exceed the threshold, the prompting module is triggered to issue a warning using a buzzer or indicator light.

[0087] Based on the aforementioned probability of cognitive dysfunction and comprehensive cognitive index, the cognitive function assessment results of the subjects in the high-altitude environment were obtained, including: The smoothed probability of cognitive dysfunction is fused with the comprehensive cognitive index for determination; among which... When one of the following conditions is met, the subject is judged to have cognitive dysfunction and a graded warning is triggered: Condition 1: The smoothed probability of cognitive function abnormality exceeds a preset probability threshold and continues to exceed a preset number of consecutive time windows; Condition 2: The comprehensive cognitive index is lower than the preset index threshold; Condition 3: The weighted combination of the smoothed cognitive function abnormality probability and the comprehensive cognitive index exceeds a preset comprehensive threshold; If none of the above conditions are met, the subject is judged to have normal cognitive function in a high-altitude environment.

[0088] Step S32: Training process of individual GRU model for subjects.

[0089] The pre-trained individual GRU model of the subject is obtained through the following process; Step 1: Basic Model Pre-training A general training sample set was collected, including: raw cognitive behavioral data collected from multiple subjects in high-altitude and low-altitude environments during the execution of a preset cognitive task, and raw physiological data collected simultaneously and preprocessed separately to form a multimodal feature vector sequence; each sample is a multimodal feature vector sequence of 30 to 60 consecutive time windows, and the feature vector of each time window is composed of standardized cognitive feature values ​​and physiological feature values; Label each training sample with a cognitive abnormality tag; The GRU model is pre-trained using binary cross-entropy as the loss function and the Adam optimization algorithm with an initial learning rate to obtain the basic GRU model. Step 2: Individual adaptive fine-tuning of the subject based on the basic GRU model: Individual baseline data of the subjects were collected before the formal test. The individual baseline data included cognitive and behavioral data of the subjects in a preset number of practice trials and physiological data collected simultaneously. The individual baseline data is preprocessed and features are extracted to obtain individual standardized parameters and individual multimodal feature vector sequences for multiple time windows. The individual multimodal feature vector sequences are used to fine-tune the basic GRU model, freezing the underlying parameters of the basic GRU model and updating only the parameters of the top fully connected layer, or updating all parameters in a few rounds with a low learning rate to obtain an individual GRU model adapted to the current subject. Step 3: Deployment of Individual GRU Models The individual GRU model is pruned and quantized with INT8 and then deployed on an embedded device to output the subject's cognitive abnormality probability in real time.

[0090] Individual GRU models are trained incrementally, for example, once a month or once a quarter.

[0091] Step S3 is used to construct a multimodal feature sequence using a sliding time window, and to obtain the smoothed probability of cognitive function abnormalities in real time through a lightweight GRU model with individual adaptive fine-tuning. This probability is then fused with the comprehensive cognitive index to achieve graded early warning of cognitive function in high-altitude environments.

[0092] Step S4, specifically.

[0093] Based on smoothed probability of cognitive dysfunction With comprehensive cognitive index The following rules are used to issue warnings for three levels of cognitive impairment: mild cognitive impairment, moderate cognitive impairment, and severe cognitive impairment: like And the duration exceeds A timing window, or ,or and weighted combination value and If so, a warning of mild cognitive abnormality will be issued; like And the duration exceeds A timing window, or ,or and weighted combination value and Or, a warning of mild cognitive abnormalities persisting for more than If the condition is not resolved within minutes, a warning for moderate cognitive abnormality will be issued. like And the duration exceeds A timing window, or ,or and weighted combination value Or, a moderate cognitive abnormality warning lasting longer than If the condition is not resolved within minutes, a warning of severe cognitive abnormality will be issued. If continuous Within each timing window, and and The current warning will be automatically lifted, and the system will return to normal cognitive status. in, , , , , All of these are preset threshold values.

[0094] These are the probability thresholds for the first, second, and third cognitive functions, respectively. These are the thresholds for the first, second, and third comprehensive cognitive indices, respectively. First, second and third respectively and The weighted combination threshold.

[0095] For example, the preset threshold is as follows: , , , , , , , , , It lasts for 10 minutes. It lasts for 5 minutes. There are two timing windows. There are 3 timing windows.

[0096] After the analysis is completed, a detailed evaluation report is generated and decision-making suggestions are provided.

[0097] The cognitive function report includes performance indicators for each cognitive task, curves showing changes in physiological parameters, the cognitive-physiological integrated state index F, and whether any abnormal warnings were detected.

[0098] If the test result is normal, the interface will display that the subject's current cognitive function is at a normal level; if an abnormality is detected, the device will issue an alarm through a pop-up window on the touch screen, prompt text, and a prompt module located inside the protective housing 1 and connected to the main control unit 31; the prompt module includes a buzzer or indicator light, and four cognitive states: mild cognitive abnormality, moderate cognitive abnormality, severe cognitive abnormality, and normal cognition, and the buzzer or light prompt can be customized.

[0099] All cognitive and behavioral data, physiological data, and assessment results are stored in the internal storage unit 32 and can be exported via the USB-C port or wireless communication unit 33 for subsequent analysis or remote monitoring. The entire process, from the start of the assessment to the final output of the results, is real-time, typically completed within a few minutes, providing researchers, medical personnel, or those working at high altitudes with timely and accurate cognitive status assessments and safety alerts.

[0100] Step S4 is to determine the level of cognitive dysfunction based on the smoothed probability of cognitive dysfunction and the comprehensive cognitive index, output the three-level warning of mild, moderate and severe cognitive dysfunction or the assessment result of normal cognitive function, and provide a visual assessment report and real-time decision support.

[0101] Figure 5 This paper illustrates the data acquisition, fusion processing, and cognitive ability assessment process of the present invention. The entire analysis process includes the following steps: ① Initialization and calibration: After the device is powered on, the microcontroller 31 performs self-tests on the storage unit 32, the touch screen 2, the wireless communication unit 33, and the power management module 4 in sequence.

[0102] PPG sensor 51 turns off the LED to collect dark current reference, and ECG sensor 52 performs internal offset calibration. Environmental parameters are acquired through barometric pressure and temperature sensors. The blood oxygen anomaly detection threshold is corrected according to the altitude level corresponding to the barometric pressure in the high-altitude environment, and LED light intensity and photodetector response drift are compensated according to the temperature in the high-altitude environment. Subsequently, the subject completes 10 trials of the Stroop consistency condition and 15 trials of the 1-back task. Microcontroller 31 records the average reaction time of the correct trials as the baseline speed, calculates the overall accuracy rate as the baseline accuracy rate, and stores it in storage unit 32 for subsequent individualized evaluation.

[0103] ② Cognitive task execution and data collection: Participants completed the Stroop color-word task and the N-back memory task based on prompts on the touch screen 2; At the same time, the cognitive monitoring module records behavioral data such as reaction time and correctness of answers for each trial. The physiological monitoring module synchronously acquires blood oxygen saturation and electrocardiogram signals using the same time reference. The microcontroller 31 assigns a unified timestamp to each frame of behavioral events and physiological sampling points to ensure that the two types of data are precisely aligned on the timeline.

[0104] ③ Data preprocessing and feature construction: The microcontroller 31 performs parallel preprocessing on the collected raw cognitive behavioral data and raw physiological data.

[0105] ④ Cognitive Ability Assessment: The data analysis module performs weighted fusion of standardized cognitive and physiological characteristic values. It then calculates the cognitive sub-index. With physiological sub-index ;based on and Calculate the comprehensive state index .

[0106] ⑤ Output of cognitive function assessment results and decision support for participants: based on the probability of cognitive function abnormalities With comprehensive cognitive index The results indicate either mild cognitive impairment, moderate cognitive impairment, or severe cognitive impairment, or normal cognition.

[0107] Decision support is shown in Table 1.

[0108]

[0109] The raw cognitive function data, raw physiological data, standardized cognitive feature values, standardized physiological feature values, and cognitive assessment results are stored for subsequent training of individual GRU models for the subjects.

[0110] In summary, the cognitive function assessment method and system of this invention, applicable to high-altitude environments, has the following beneficial effects: 1. In terms of integrated portability and adaptability to high-altitude environments, this invention is the first to integrate standardized cognitive task presentation, simultaneous acquisition of multimodal physiological signals, edge-side intelligent fusion analysis, and dynamic power consumption management into a single portable hardware device, overcoming the shortcomings of existing technologies where components are separate and lack system integration. By using a barometric pressure sensor to correct the blood oxygen anomaly judgment threshold and a temperature sensor to compensate for PPG measurement drift, combined with a dual-material protective shell of polycarbonate reinforced alloy and aluminum-magnesium alloy, the device can operate stably for extended periods in harsh high-altitude environments ranging from -20℃ to +50℃, with low oxygen, strong light, and low temperatures. The entire device weighs less than 1000g and has a battery life exceeding 72 hours, truly achieving independent and reliable use in high-altitude environments. 2. In terms of intelligent fusion and real-time early warning, this invention constructs cognitive and physiological sub-indices and performs dynamic weighted fusion. When blood oxygen saturation is below the hypoxia threshold, the weight of the physiological sub-index is automatically increased (up to 0.75), enabling the comprehensive assessment to sensitively reflect the impact of hypoxia on cognitive function. Simultaneously, multimodal temporal feature sequences are input into a lightweight GRU model after pruning and INT8 quantization, outputting the probability of cognitive function abnormalities every 30 seconds. This probability is then combined with the comprehensive cognitive index for multi-condition fusion judgment and graded early warning. Compared to existing technologies that rely on offline statistics and independent indicator judgment, this invention achieves real-time fusion analysis and anomaly detection on the edge, triggering early warnings in the early stages of cognitive function abnormalities, significantly reducing the false negative and false positive rates. 3. Regarding individual adaptation and low-power deployment: This invention employs a strategy of "basic pre-training + individual fine-tuning" to obtain an individual GRU model adapted to the subject. Adaptive fine-tuning is performed using individual baseline data from practice trials, significantly improving the accuracy of individual cognitive function discrimination. Simultaneously, through dynamic voltage and frequency adjustment, time-sharing power supply, and deep sleep mechanisms, combined with pruning and INT8 quantization for lightweight model deployment, the device can achieve a battery life exceeding 72 hours under typical testing conditions. Compared with existing technologies, this invention ensures both low-power real-time inference capabilities on the embedded device and model adaptation to individual differences, providing an accurate, reliable, and portable solution for cognitive health monitoring of personnel working in high-altitude areas.

[0111] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0112] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A cognitive function assessment method suitable for high-altitude environments, characterized in that, include: Raw cognitive behavioral data of subjects performing a pre-set cognitive task in a high-altitude environment were collected, and raw physiological data were collected simultaneously and preprocessed to obtain standardized cognitive and physiological characteristic values. The cognitive sub-index and physiological sub-index, constructed based on the standardized cognitive and physiological feature values, are weighted and fused to obtain a comprehensive cognitive index; wherein, the weights of the cognitive sub-index and physiological sub-index are adjusted in real time using environmental parameters of the plateau environment acquired in real time. The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into the pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction. Based on the probability of cognitive dysfunction and the comprehensive cognitive index, the cognitive function assessment results of the subjects in the high-altitude environment were obtained.

2. The cognitive function assessment method suitable for plateau environments according to claim 1, characterized in that, The preset cognitive tasks include the Stroop color-word interference task and the N-back working memory task; the raw cognitive behavior data include reaction time, answer accuracy, error rate, hit rate, false negative rate, Stroop interference effect, and N-back working memory load index. The raw physiological data includes photoplethysmography (PPG) signals and electrocardiogram (ECG) signals. The standardized physiological characteristic values ​​obtained after preprocessing include mean blood oxygen saturation, rate of change of blood oxygen saturation, mean and rate of change of heart rate, SDNN, RMSSD, and pNN50 heart rate variability index; wherein, the photoplethysmography (PPG) signal is acquired using a PPG sensor.

3. The cognitive function assessment method suitable for plateau environments according to claim 2, characterized in that, The raw cognitive-behavioral data and raw physiological data are preprocessed, including: The original cognitive behavioral data was filtered out by removing invalid trials that exceeded the individual baseline reaction time of the subjects by ±3 times the standard deviation or violated the task rules, to obtain the first cognitive behavioral data; The photoplethysmography (PPG) wave signal in the original physiological data is subjected to moving average filtering, and the electrocardiogram (ECG) signal is subjected to bandpass filtering and notch filtering to obtain the first physiological data. The first cognitive behavioral data and the first physiological data are timestamped and then Z-score standardized to obtain standardized cognitive feature values ​​and physiological feature values.

4. The cognitive function assessment method suitable for plateau environments according to claim 2, characterized in that, The cognitive sub-indices and physiological sub-indices constructed based on the standardized cognitive and physiological feature values ​​are weighted and fused to obtain a comprehensive cognitive index, including: The cognitive sub-index is calculated by weighted summation or weighted average of each standardized cognitive feature value. The physiological sub-index is calculated by weighted summation or weighted average of each standardized physiological characteristic value; The cognitive sub-index and the physiological sub-index are weighted and fused as follows: in, As a comprehensive cognitive index, For cognitive sub-index, Physiological sub-index These are the weights of the cognitive sub-index and the physiological sub-index, respectively.

5. The cognitive function assessment method suitable for plateau environments according to claim 4, characterized in that, The environmental parameters of the plateau environment include ambient air pressure and ambient temperature; the ambient air pressure is obtained using an air pressure sensor, and the ambient temperature is obtained using a temperature sensor. The weights of the cognitive and physiological sub-indices are adjusted in real time using environmental parameters of the plateau environment acquired in real time, including: The current altitude is calculated based on the ambient air pressure, and the blood oxygen abnormality judgment threshold is corrected based on the altitude. The ambient temperature is obtained using a temperature sensor, and the LED light intensity and photodetector response drift of the PPG sensor are compensated based on the ambient temperature to ensure the accuracy of blood oxygen saturation measurement. The blood oxygen saturation of the subjects is detected in real time, and the blood oxygen saturation is compared with the blood oxygen abnormality judgment threshold after air pressure correction. If the blood oxygen saturation is lower than the blood oxygen abnormality judgment threshold after pressure correction, the weight of the physiological sub-index is increased based on the degree of deviation of blood oxygen saturation. At the same time, the weight of the cognitive sub-index is reduced accordingly. And the weight of the physiological sub-index The upper limit shall not exceed the weight threshold; If the blood oxygen saturation is greater than or equal to the blood oxygen abnormality judgment threshold after pressure correction, the weights of the physiological sub-index and the cognitive sub-index are maintained at the default values ​​and no weight adjustment is made.

6. The cognitive function assessment method suitable for plateau environments according to claim 1, characterized in that, The standardized cognitive and physiological feature values ​​within the time window are combined to form a multimodal feature vector sequence, which is then input into a pre-trained individual GRU model of the subject to obtain the probability of cognitive dysfunction, including: A fixed-length sliding time window is set, and the window sliding step size is less than the total window length to achieve overlapping sampling of time series data; wherein, the time series window length and sliding step size are preset based on the data acquisition frequency; Based on the standardized cognitive and physiological feature values ​​collected synchronously within each time window, the dimensions are concatenated to generate a single-step multimodal feature vector; all single-step feature vectors within the window are arranged in chronological order of data collection time to form the first multimodal feature vector sequence. The first multimodal feature vector sequence is subjected to unified processing of dimension padding and truncation to obtain the second multimodal feature vector sequence, so that the dimension of the input sequence matches the input layer size of the pre-trained GRU model; The second multimodal feature vector sequence is input into the pre-trained GRU model to obtain the preliminary probability of cognitive function abnormality. The initial cognitive function abnormality probability output by the current window is taken and weighted averaged with the initial cognitive function abnormality probabilities output by the previous two consecutive time windows to eliminate probability jumps caused by fluctuations in single-frame data, and the smoothed cognitive function abnormality probability is obtained as the cognitive function abnormality probability of the subject.

7. The cognitive function assessment method suitable for plateau environments according to claim 6, characterized in that, Based on the aforementioned probability of cognitive dysfunction and comprehensive cognitive index, the cognitive function assessment results of the subjects in the high-altitude environment were obtained, including: The smoothed probability of cognitive dysfunction is fused with the comprehensive cognitive index for determination; among which... When one of the following conditions is met, the subject is judged to have cognitive dysfunction and a graded warning is triggered: Condition 1: The smoothed probability of cognitive function abnormality exceeds a preset probability threshold and continues to exceed a preset number of consecutive time windows; Condition 2: The comprehensive cognitive index is lower than the preset index threshold; Condition 3: The weighted combination of the smoothed cognitive function abnormality probability and the comprehensive cognitive index exceeds a preset comprehensive threshold; If none of the above conditions are met, the subject is judged to have normal cognitive function in a high-altitude environment.

8. The cognitive function assessment method suitable for plateau environments according to claim 7, characterized in that, Based on smoothed probability of cognitive dysfunction With comprehensive cognitive index The following rules are used to issue warnings for three levels of cognitive impairment: mild cognitive impairment, moderate cognitive impairment, and severe cognitive impairment: like And the duration exceeds A timing window, or ,or and weighted combination value and If so, a warning of mild cognitive abnormality will be issued; like And the duration exceeds A timing window, or ,or and weighted combination value and Or, a warning of mild cognitive abnormalities persisting for more than If the condition is not resolved within minutes, a warning for moderate cognitive abnormality will be issued. like And the duration exceeds A timing window, or ,or and weighted combination value Or, a moderate cognitive abnormality warning lasting longer than If the condition is not resolved within minutes, a warning of severe cognitive abnormality will be issued. If continuous Within each timing window, and and The current warning will be automatically lifted, and the system will return to normal cognitive status. in, , , , , All of these are preset threshold values.

9. The cognitive function assessment method suitable for high-altitude environments according to any one of claims 1-8, characterized in that, The pre-trained individual GRU model of the subject is obtained through the following process; Step 1: Basic Model Pre-training A general training sample set was collected, including: raw cognitive behavioral data collected from multiple subjects in high-altitude and low-altitude environments during the execution of a preset cognitive task, and raw physiological data collected simultaneously and preprocessed separately to form a multimodal feature vector sequence; each sample is a multimodal feature vector sequence of 30 to 60 consecutive time windows, and the feature vector of each time window is composed of standardized cognitive feature values ​​and physiological feature values; Label each training sample with a cognitive abnormality tag; The GRU model is pre-trained using binary cross-entropy as the loss function and the Adam optimization algorithm with an initial learning rate to obtain the basic GRU model. Step 2: Individual adaptive fine-tuning of the subject based on the basic GRU model: Individual baseline data of the subjects were collected before the formal test. The individual baseline data included cognitive and behavioral data of the subjects in a preset number of practice trials and physiological data collected simultaneously. The individual baseline data is preprocessed and features are extracted to obtain individual standardized parameters and individual multimodal feature vector sequences for multiple time windows. The individual multimodal feature vector sequences are used to fine-tune the basic GRU model, freezing the underlying parameters of the basic GRU model and updating only the parameters of the top fully connected layer, or updating all parameters in a few rounds with a low learning rate to obtain an individual GRU model adapted to the current subject. Step 3: Deployment of Individual GRU Models The individual GRU model is pruned and quantized with INT8 and then deployed on an embedded device to output the subject's cognitive abnormality probability in real time.

10. A cognitive function assessment system suitable for high-altitude environments, characterized in that, include: The cognitive monitoring module is used to collect raw cognitive behavioral data of subjects performing preset cognitive tasks in a high-altitude environment; The physiological monitoring module is used to collect raw physiological data from the subjects; Environmental sensors, including barometric pressure sensors and temperature sensors, are used to collect high-altitude environmental parameters in real time. The main control processing unit is responsible for task scheduling, multimodal data timestamp synchronization, raw data filtering and noise reduction, feature extraction, time-series window truncation, and model inference scheduling. The data analysis module is used to fuse cognitive and behavioral data with physiological data and to intelligently identify anomalies. It includes a feature construction unit, a weighted fusion unit, and a lightweight temporal discrimination unit; the lightweight temporal discrimination unit adopts a GRU model after pruning and INT8 quantization, with 32 hidden units; Used to deploy individual GRU models of subjects after pruning and INT8 quantization, to complete the standardization, weighted fusion, comprehensive cognitive index calculation, temporal feature sequence inference, and cognitive function abnormality probability calculation of multimodal raw data; The alert module, including a buzzer and indicator lights, is used to issue an audible and visual alarm when a graded warning is triggered.

Images