Unmanned injury assessment method and system

Abstract

The invention discloses an unmanned injury condition assessment method and system, and relates to the technical field of injury condition assessment, and the method comprises the following steps: collecting environment parameters and physiological state information of a wounded person; performing data integration and data normalization on the environmental parameters and the physiological state information to obtain an environmental feature sequence and a physiological feature sequence; the physiological feature sequence is input into an injury condition grade classification model, an injury condition grade classification result is obtained, and the injury condition grade classification model is composed of an Informer model and a GRU model; and inputting the environment feature sequence, the physiological feature sequence and the injury condition level classification result into a pre-constructed medical knowledge graph for retrieval matching to obtain a treatment measure set, and performing priority ranking on various treatment measures in the treatment measure set to obtain a life support decision scheme, thereby completing unmanned injury condition assessment. The problem that in the prior art, decision making depends on artificial experience under the condition that a data source is single can be solved.

Description

Technical Field

[0001] This invention relates to the field of injury assessment technology, specifically to an unmanned injury assessment method and system. Background Technology

[0002] In emergency rescue and other fields, rapid and accurate injury assessment of the wounded is a crucial step in improving survival rates. Currently, the technologies commonly used in this field mainly include portable medical monitoring devices, telemedicine consultation systems, and wearable physiological parameter sensors.

[0003] However, existing technologies still have significant limitations in practical applications: portable devices typically require operation by professionals, limiting their applicability in high-risk scenarios or where personnel cannot directly access them; telemedicine relies heavily on stable, high-quality communication links, making it difficult to guarantee real-time performance and reliability in complex environments such as earthquakes and battlefields; wearable monitoring devices require pre-wearing and have limited ability to identify concealed injuries or multiple injuries. These methods only provide a single-dimensional data source, resulting in fragmented injury information, and the assessment process still heavily relies on the human experience and judgment of experts. This isolation of the data source and the strong dependence of the decision-making process on human experience make it difficult for existing technologies to meet the urgent need for unmanned, contactless, and intelligent decision-making in high-risk, resource-constrained environments for injury assessment. Summary of the Invention

[0004] The purpose of this invention is to provide an unmanned injury assessment method and system to solve the problem of relying on human experience for decision-making in the case of a single data source in the existing technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, an unmanned injury assessment method includes the following steps: Collect environmental parameters and physiological status information of the wounded; The environmental parameters and physiological state information are integrated and normalized to obtain environmental feature sequences and physiological feature sequences. The physiological feature sequence is input into the injury severity classification model to obtain the injury severity classification result. The injury severity classification model consists of an Informer model and a GRU model. The environmental feature sequence, physiological feature sequence, and injury severity classification results are input into a pre-constructed medical knowledge graph for retrieval and matching to obtain a set of treatment measures. The treatment measures in the set are then prioritized to obtain a life support decision-making scheme, thus completing the unmanned injury assessment.

[0006] In some implementations, the Informer model includes a first input layer, a first feature extraction layer, and a first output layer connected in sequence, wherein the first feature extraction layer is configured with Prob sparse self-attention; The GRU model includes a second input layer, a second feature extraction layer, and a second output layer connected in sequence. The second feature extraction layer includes several layers of unidirectional GRU units.

[0007] In some embodiments, the step of inputting the physiological feature sequence into the injury severity classification model to obtain the injury severity classification result specifically includes: The physiological feature sequence is input into the first input layer, and after the long-term dependent features in the physiological feature sequence are extracted by the first feature extraction layer, the sequence is then subjected to nonlinear transformation by the first output layer to obtain the first feature vector. The physiological feature sequence is input into the second input layer, and the short-term temporal features in the physiological feature sequence are extracted by the second feature extraction layer. Then, the sequence is nonlinearly transformed by the second output layer to obtain the second feature vector. The first feature vector and the second feature vector are weighted and fused according to a preset weight to obtain a fused feature vector. The fused feature vector is then input into a classifier to obtain the injury level classification result.

[0008] In some implementations, the medical knowledge graph is pre-built through the following steps: Acquire medical knowledge data and identify and extract medical entities and their semantic relationships from the medical knowledge data; Create a node system, which includes: injury node, symptom node, vital sign status node, and treatment measure node, corresponding to injury, symptoms, vital sign status, and treatment measures, respectively; The medical entities are transformed into nodes of corresponding types as entity nodes. Relationship edges are established between the entity nodes according to the semantic relationships. The entity nodes and relationship edges are stored in a graph database to obtain a medical knowledge graph.

[0009] In some implementations, the steps of inputting the environmental feature sequence, physiological feature sequence, and injury severity classification results into a pre-constructed medical knowledge graph for retrieval and matching to obtain a set of treatment measures, and prioritizing the various treatment measures in the set to obtain a life support decision-making scheme, specifically include: The environmental feature sequence, physiological feature sequence, and injury severity classification results are mapped to entity nodes, and entity relationship retrieval is performed based on the relation edges to obtain a set of treatment measures; Based on a pre-defined clinical rule base, the execution priority of each treatment measure in the set of treatment measures is determined and sorted to obtain the treatment measures and their corresponding implementation order, which serves as a life support decision-making scheme.

[0010] In some embodiments, the environmental parameters include: air pressure, temperature, humidity, light intensity, magnetic field strength, and sound intensity; The physiological state information includes: vital signs information, injury site information, and state of consciousness information. The vital signs information includes: respiratory rate, heart rate, and body temperature. The steps of integrating and normalizing the environmental parameters and physiological state information to obtain environmental feature sequences and physiological feature sequences specifically include: Obtain the maximum and minimum values ​​of the environmental parameters and physiological state information over time; Based on the maximum and minimum values, the environmental parameters and physiological state information are scaled to a preset range through normalization to obtain environmental feature sequences and physiological feature sequences.

[0011] Secondly, an unmanned injury assessment system includes: The environmental information and injury information sensing module is used to collect environmental parameters and physiological status information of the injured. The data processing module is used to integrate and normalize the environmental parameters and physiological state information to obtain environmental feature sequences and physiological feature sequences. The intelligent injury assessment module is used to input the physiological feature sequence into the injury level classification model to obtain the injury level classification result. The injury level classification model consists of an Informer model and a GRU model. The life support decision module is used to input the environmental feature sequence, physiological feature sequence and injury level classification results into a pre-constructed medical knowledge graph for retrieval and matching, obtain a set of treatment measures, prioritize each treatment measure in the set of treatment measures, obtain a life support decision scheme, and complete unmanned injury assessment.

[0012] Thirdly, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the unmanned injury assessment method.

[0013] Fourthly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the unmanned injury assessment method.

[0014] Fifthly, a computer program product comprising a computer program that, when executed by a processor, implements the steps of the unmanned injury assessment method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides an unmanned injury assessment method that simultaneously collects environmental parameters and physiological state information of the injured, solving the problems of single data sources and fragmented information in existing technologies. It employs an integrated algorithm based on Informer and GRU to analyze physiological feature sequences, simultaneously capturing both long-term dependencies and short-term temporal changes in vital signs. This intelligent assessment method replaces traditional subjective judgments relying on human experience, achieving automatic classification of injury levels, improving assessment accuracy, shortening assessment time, and meeting the timeliness requirements of the "golden hour" treatment window. Through a pre-constructed medical knowledge graph, it comprehensively retrieves and matches environmental feature sequences, physiological feature sequences, and injury level classification results to generate a personalized life support decision-making plan that includes treatment measures and their implementation sequence. This method fully considers the impact of environmental factors on treatment effectiveness and can provide the optimal treatment plan based on the actual situation on site, solving the problem of traditional rescue decision-making lacking environmental adaptability. Attached Figure Description

[0016] Figure 1 A flowchart of an unmanned injury assessment method provided in an embodiment of the present invention; Figure 2 A simplified structural diagram of an unmanned injury assessment system provided in an embodiment of the present invention; Figure 3 A detailed structural diagram of an unmanned injury assessment system provided in an embodiment of the present invention; Figure 4 This is a diagram illustrating the specific working process of the data processing module provided in an embodiment of the present invention. Figure 5 This is a diagram illustrating the specific working process of the intelligent injury assessment module provided in this embodiment of the invention. Figure 6 The injury assessment confusion matrix diagram provided in the embodiments of the present invention; Figure 7 A schematic diagram illustrating the injury assessment performance provided in an embodiment of the present invention; Figure 8 This is a diagram illustrating the specific working process of the life support decision-making module provided in an embodiment of the present invention. Figure 9 This is a diagram showing the problem settings of the voice interaction module provided in an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content described herein is for explanation rather than limitation of the present invention.

[0018] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this invention are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, systems, products, or devices.

[0019] Example 1 like Figure 1 As shown, this embodiment provides an unmanned injury assessment method, including the following steps: S1 collects environmental parameters and physiological status information of the injured person; environmental parameters include air pressure, temperature, humidity, light intensity, magnetic field strength, and sound intensity. Physiological status information includes vital signs, injury site information, and level of consciousness information; vital signs include respiratory rate, heart rate, and body temperature.

[0020] S2, the environmental parameters and physiological state information are integrated and normalized to obtain environmental feature sequences and physiological feature sequences; data integration is performed by continuously collecting data at a sampling rate of 0.5Hz for 100 seconds, and the static physiological indicators are copied and expanded into time series of the same length as the dynamic physiological indicators. Data normalization is performed by obtaining the maximum and minimum values ​​of the environmental parameters and physiological state information in the time series, and scaling each feature data to the [0,1] interval based on the maximum and minimum values.

[0021] S3, the physiological feature sequence is input into the injury severity classification model to obtain the injury severity classification result. The injury severity classification model consists of an Informer model and a GRU model. The physiological feature sequence is input into the Informer model and the GRU model respectively. The Informer model extracts long-term dependency features from the physiological feature sequence and outputs a first feature vector. The GRU model extracts short-term time-series features from the physiological feature sequence and outputs a second feature vector. The first feature vector and the second feature vector are weighted and fused according to preset weights to obtain a fused feature vector. The fused feature vector is then input into the classifier to output the injury severity classification result. The injury severity classification result includes four levels: minor injury, moderate injury, serious injury, and critical injury.

[0022] S4. The environmental feature sequence, physiological feature sequence, and injury level classification results are input into a pre-constructed medical knowledge graph for retrieval and matching to obtain a set of treatment measures. The treatment measures in this set are then prioritized to obtain a life support decision-making scheme, completing the unmanned injury assessment. The medical knowledge graph obtains medical knowledge data from a predetermined medical data source, identifies and extracts medical entities and their semantic relationships from the medical knowledge data, and creates a node system including injury, symptoms, vital signs, and treatment measures. Medical entities are converted into corresponding types of nodes, and relational edges are established between entity nodes based on semantic relationships. This information is then stored in a graph database. During decision-making, the environmental feature sequence, physiological feature sequence, and injury level classification results are mapped to entity nodes. An entity relationship retrieval based on relational edges is performed to obtain a set of treatment measures. Then, based on a pre-set clinical rule base, the execution priority of each treatment measure in the set is determined and ranked to obtain a life support decision-making scheme containing treatment measures and their corresponding implementation order.

[0023] Example 2 like Figure 2 As shown, this embodiment provides an unmanned injury assessment system, including: The environmental information and injury information sensing module is used to collect environmental parameters and physiological status information of the injured. The environmental information and injury information perception module includes an environmental information sensing module and an injury information perception module. The environmental information sensing module contains a barometric pressure sensor, a temperature sensor, a humidity sensor, a light sensor, a magnetic field sensor, and a sound intensity sensor. It is used to perceive environmental information of the injured person, such as barometric pressure, temperature, humidity, light intensity, magnetic field intensity, and sound intensity. This provides important reference for subsequent life support decisions, ensuring that rescue measures can fully consider the impact of environmental factors, thereby improving the scientific nature and effectiveness of the rescue.

[0024] Barometric Pressure Sensor: In this patent embodiment, the barometric pressure sensor module is used to acquire barometric pressure information near the injured person. Barometric pressure measurement is performed using a capacitive pressure sensor. The sensor's pressure-sensing element consists of two plates separated by a very small gap. Changes in ambient pressure cause changes in the gap size, resulting in changes in the capacitance between the two plates. The sensor converts these changes in capacitance into a digital signal, which is then used to calculate the barometric pressure value.

[0025] Temperature Sensor: In this patent embodiment, the temperature sensor module is used to acquire temperature information near the injured person. The temperature measurement principle of the temperature sensor module is based on the characteristics of an NTC thermistor. As the temperature rises, the resistance of the NTC decreases, and as the temperature falls, the resistance increases. By measuring the change in resistance, the ambient temperature is calculated. The sensor converts this analog signal into a digital signal and outputs the temperature data.

[0026] Humidity sensor: In this patent embodiment, the humidity sensor module is used to acquire humidity information near the injured person. Humidity measurement relies on the property that the dielectric constant of the polymer film on the humidity sensor changes with humidity. The dielectric constant of the film changes with the amount of water vapor adsorbed on the film, causing a corresponding change in the film's capacitance. The sensor calculates the relative humidity of the current environment by measuring this capacitance value.

[0027] Light sensor: In this patent embodiment, the light sensor module is used to acquire light information near the injured person. The light sensor internally consists of a photodiode, operational amplifier, ADC acquisition, crystal oscillator, etc. The photodiode converts the input light signal into an electrical signal through the photovoltaic effect. After being amplified by the operational amplifier circuit, the voltage is acquired by the ADC, and then converted into a 16-bit binary number by the logic circuit and stored in the internal register.

[0028] Magnetic field sensor: In this patent embodiment, the magnetic field sensor module is used to acquire magnetic field information near the injured person. The magnetic field sensor employs the magnetoresistive effect principle, acquiring magnetic field information by measuring the effect of the magnetic field on the magnetoresistive material. The magnetoresistive effect is based on the properties of magnetoresistive materials. When a magnetic field is applied to a magnetoresistive material, the magnetic field changes the resistance value of the magnetoresistive material. The sensor calculates the strength of the magnetic field by measuring this change in resistance.

[0029] Sound intensity sensor: In this patent embodiment, the sound intensity sensor module is used to acquire sound information near the injured person. Sound waves (changes in air pressure) strike the diaphragm of the microphone, causing the diaphragm to vibrate. In an electret microphone, the diaphragm and a fixed backplate form a capacitor. The vibration of the diaphragm changes the capacitance of this capacitor (change in the distance between the plates). The presence of the electret material (carrying a permanent charge) causes the change in capacitance to be converted into a change in voltage across the capacitor. The sensor calculates the sound intensity by measuring this voltage change.

[0030] The injury information perception module includes a bio-radar sensing module, an optical sensing module, and a voice interaction module, which are used for non-contact monitoring of the injured person's vital signs (respiratory rate, heart rate, and body temperature), identifying the injured person's injury site, and assessing the level of consciousness.

[0031] Bio-radar module: Used for non-contact monitoring of vital signs in wounded personnel, including respiratory rate and heart rate. Bio-radar transmits and receives millimeter-wave signals, penetrating obstacles such as clothing to accurately capture minute chest vibration signals, thus enabling real-time monitoring of respiratory and heartbeat signals. Bio-radar emits electromagnetic waves of a specific frequency towards the wounded. These waves are reflected upon contact with the human body. Due to breathing, heartbeat, and body movement, the frequency of the reflected electromagnetic waves shifts. Bio-radar receives and analyzes these reflected electromagnetic wave signals to extract information related to breathing, heart rate, and body movement, ultimately achieving non-contact monitoring of these vital signs.

[0032] The bio-radar in this embodiment adopts a common modular architecture, mainly including the following core modules. These modules work together to complete the non-contact vital sign monitoring function: Signal Transmission Module: This module mainly consists of a signal generator, a power amplifier, and a transmitting antenna. The signal generator generates a continuous wave signal at a preset frequency as the transmission signal of the bio-radar; the power amplifier amplifies the signal output from the signal generator to ensure that the transmitted signal can effectively reach the monitoring target and generate a detectable reflected signal; the transmitting antenna is responsible for radiating the amplified signal as electromagnetic waves towards the monitoring target. Signal Receiving Module: This module consists of a receiving antenna, a low-noise amplifier, and a mixer. The receiving antenna receives electromagnetic wave signals with Doppler frequency shift reflected from the monitoring target; the low-noise amplifier amplifies the weak reflected signal acquired by the receiving antenna while minimizing noise interference to avoid noise affecting subsequent signal analysis; the mixer mixes the signal output from the low-noise amplifier with the reference signal output from the signal generator in the signal transmission module to obtain an intermediate frequency signal containing Doppler frequency shift information, facilitating signal processing by subsequent modules. Signal processing module: The core of this module consists of a data acquisition unit and a digital signal processor. The data acquisition unit uses a common analog-to-digital converter to convert the analog intermediate frequency signal output from the signal receiving module into a digital signal. The digital signal processor performs basic processing on the digital signal, such as filtering (removing power frequency interference and environmental noise) and signal smoothing, to provide a high-quality processed signal for subsequent vital sign feature extraction.

[0033] The following is a detailed description of the acquisition of respiratory rate, heart rate and body movement information by bio-radar: (1) Respiratory rate and heart rate extraction: Filtering: High-frequency noise and low-frequency interference in the signal are removed by bandpass filter, and the characteristic frequency range of respiratory (0.2-0.5 Hz) and heartbeat (0.8-2 Hz) signals are retained. Adaptive filtering: Background noise and other interference signals are further removed by adaptive filtering algorithm, and the filtering parameters are dynamically adjusted to adapt to signal changes. Respiratory rate extraction and calculation: Variational mode decomposition is performed on the denoised radar signal, and the mode components with a frequency range of 0.2-0.5 Hz are extracted from the decomposition results as respiratory signals. Fast Fourier transform is performed on the respiratory signal, the power spectral density is calculated, the peak value corresponding to the respiratory frequency is found, and the respiratory rate (times / minute) is calculated by its reciprocal. Heart rate extraction and calculation: Mode components with a frequency range of 0.8-2 Hz are extracted from the mode decomposition results as heartbeat signals. Fast Fourier transform is performed on the heartbeat signal, the power spectral density is calculated, the peak value corresponding to the heartbeat frequency is found, and the heart rate (times / minute) is calculated by its reciprocal. (2) Body movement information extraction: The signal amplitude threshold detection method is adopted. When a human body moves, the amplitude of the reflected electromagnetic wave changes drastically, which in turn causes significant fluctuations in the amplitude of the signal received by the bio-radar. Based on this characteristic, a threshold for signal amplitude change is preset, and the amplitude of the processed signal output by the signal processing module is monitored in real time; when the change in signal amplitude exceeds the preset threshold, it is determined that the injured person has body movement; if the signal amplitude remains stable and the change is within the threshold range, it is determined that there is no body movement.

[0034] Optical sensing module: used to identify the injured parts of the wounded. The optical sensors are visible light cameras and infrared cameras, which can capture visible light images and thermal images of the wounded. By fusing the two image recognition technologies, the information of the injured parts is analyzed, and the body temperature information of the wounded is perceived through the thermal image of the wounded. This module includes a visible light camera and an infrared camera. By fusing the image data of these two cameras, the system can more accurately identify and locate the injured parts of the wounded. The following is a detailed description of the acquisition of the injured parts by the optical sensors, including the image processing process and the data fusion method. (1) Image processing: Visible light image processing: including denoising, contrast enhancement, edge detection, color segmentation, shape analysis, etc. Infrared image processing: including denoising, temperature calibration, temperature anomaly detection, region segmentation, etc. (2) Data fusion method: Image fusion: the visible light image and the infrared image are aligned and fused by feature point matching and image registration to generate feature vectors. Injured parts identification and localization: the basic architecture is built based on the improved YOLOv8 model, and the channel attention mechanism and spatial attention mechanism modules are introduced to enhance the feature extraction capability. The fused features are input into the model, and feature extraction and fusion are performed through convolutional layers and attention mechanism modules to identify the location and size of the bleeding area. The edge features and color gradient information of the bleeding area are further analyzed to determine the specific location of the injured part and output detailed information.

[0035] Voice Interaction Module: The voice interaction module is used to assess the level of consciousness of the injured. This module outputs a binary logic question via the voice module. For injured persons with speech ability, it receives voice signals; for those without speech ability, it detects body movement signals via a bio-radar module. Based on the question-and-answer response, it outputs a consciousness score and partial injury information (whether there is difficulty breathing / severe pain / cold body). In this patent embodiment, the voice interaction module assesses the injured person's consciousness score (GCS) and obtains partial injury information by capturing the injured person's binary voice signal, providing important basis for injury assessment and subsequent life support decisions.

[0036] In this embodiment, the consciousness score is primarily obtained through voice interaction between the system and the injured person. The system asks the injured person seven questions, with a 10-second pause between each question. The system asks twice whether voice interaction was necessary to prevent misjudgment. Simultaneously, this module is compatible with bio-radar detection of body movement interaction, using body movement to assist in determining the state of consciousness when voice interaction is not possible.

[0037] The inquiry is as follows: (1) "Hello, may I ask if we can have voice interaction? Please answer 'yes' or 'no' within 10 seconds. If no answer is received within 10 seconds, it will be determined that voice interaction is not possible." This question is used to confirm whether the injured person can engage in voice interaction. It will be asked twice to prevent misjudgment. If not (no response is received within 10 seconds of the question being asked, indicating an inability to engage in voice interaction, with a language score of 5 in the consciousness assessment), then bioradar will be used for motion monitoring.

[0038] (2) "Can you handle it yourself? Please answer 'yes' or 'no' within 10 seconds." This question aims to assess the injured person's consciousness and cognitive abilities, determining whether they can understand and respond to questions about their own state. The injured person should answer based on their physical condition. If they can manage on their own (answer "yes"), they receive a score of 10 in the consciousness assessment, and the next injured person is then questioned. If they cannot manage on their own (answer "no"), the next question is triggered.

[0039] (3) "Can you open your eyes automatically? Please answer 'yes' or 'no' within 10 seconds." This question is used to detect the basic level of consciousness of the injured person. It is mainly used for assessing the state of consciousness. If the answer is "yes", the score for opening the eyes is 2; otherwise, it is 0.

[0040] (4) "Do you have free movement of your limbs? Please answer 'yes' or 'no' within 10 seconds." This question assesses whether the injured person has the ability to move physically, and is mainly used to assess consciousness. If the answer is "yes", the motor score is 3; otherwise, it is 0.

[0041] (5) "Are you having difficulty breathing? Please answer 'yes' or 'no' within 10 seconds." The question asks the injured person if they are experiencing difficulty breathing. If they are experiencing difficulty breathing (answer "yes"), a respiratory stimulant will be administered. If the answer is yes, the respiratory score in the consciousness assessment will be -5.

[0042] (6) "Do you feel severe pain? Please answer 'yes' or 'no' within 10 seconds." The question asks the injured person if they are experiencing severe pain. If the injured person is experiencing severe pain (answers "yes"), then pain management is administered, and the pain score in the consciousness assessment is -2.

[0043] (7) "Do you feel cold? Please answer 'yes' or 'no' within 10 seconds." The question asks about the injured person's body temperature. If the injured person feels cold (answer "yes"), then subsequent warming treatment will be carried out.

[0044] The above seven questions are answered with a 10-second pause between each pair, keeping the entire question-and-answer interaction time within four minutes. Six of these questions—"Can I interact via voice?", "Can I handle this myself?", "Can I open my eyes automatically?", "Can I move my limbs freely?", "Is there difficulty breathing?", and "Is there severe pain?"—are used to assess consciousness. The remaining three questions—"Is there difficulty breathing?", "Is there severe pain?", and "Is my body cold?"—are used for subsequent life support decisions. This simple binary logic question-and-answer design effectively improves the adaptability of human-computer interaction to complex on-site environments and types of injuries, while also meeting the timeliness requirements for subsequent injury assessment and life support.

[0045] In addition, if the injured person is able to manage their injuries on their own, the overall level of consciousness is rated as 15; if the injured person does not have the ability to interact verbally, the overall level of consciousness is rated as 1.

[0046] The data processing module integrates and normalizes the environmental parameters and physiological state information to obtain environmental feature sequences and physiological feature sequences. It also integrates and processes the multimodal data collected by the sensors in the injury information sensing module, ensuring data accuracy and consistency. Specific functions include: data reception: receiving raw data from different sensors in the injury information sensing module; data integration: integrating the multimodal data from different sensors to form a unified data format; and data normalization: normalizing the integrated data to unify feature scales. Through the processing of the data processing module, the system can transform complex multimodal data into standardized features that can be used for analysis and evaluation, laying the foundation for subsequent intelligent injury assessment and life support decision-making.

[0047] The data processing module synchronously acquires data from each sensor at a sampling rate of 0.5Hz for 100 seconds, replicating and expanding the static physiological indicators into a sequence of 50 points to maintain consistent data length. Data integration converts the acquired body temperature, respiratory rate, heart rate, and GCS data into a 50×4 feature vector. Data normalization employs the min-max normalization method, calculating the maximum and minimum values ​​of each feature over time, and scaling the original data points to the interval [0,1] using these maximum and minimum values, as shown in the following formula:

[0048] in, For the normalized data features, Features of the original data The minimum value of the original data features in the time dimension. This represents the maximum value of the original data feature in the time dimension.

[0049] The intelligent injury assessment module is used to input the physiological feature sequence into an injury severity classification model to obtain the injury severity classification result. The injury severity classification model consists of an Informer model and a GRU model. The Informer model includes a first input layer, a first feature extraction layer, and a first output layer connected in sequence. The first input layer expands the normalized feature matrix into a 50×512-dimensional feature matrix after passing through an encoding layer. The first feature extraction layer uses a Prob sparse self-attention mechanism, where the sparsity factor is 5, the number of multi-head attention heads is 8, and the head dimension is 512. The first output layer obtains a 128×4 feature vector through nonlinear transformation. The GRU model includes a second input layer, a second feature extraction layer, and a second output layer connected in sequence. The second input layer expands the normalized feature matrix into a 50×512-dimensional feature matrix after passing through an encoding layer. The second feature extraction layer uses 4 layers of unidirectional GRU units, with a hidden layer dimension of 512. The second output layer obtains a 128×4 feature vector through nonlinear transformation. The intelligent injury assessment module weights and fuses the feature vectors output by the two models with a weight ratio of 0.8:0.2, and outputs injury assessment results of four levels: minor, moderate, serious, and critical injury through a classifier. Specifically, the Informer model's self-attention mechanism captures long-term dependencies in physiological feature sequences, thereby identifying global patterns of injury features changing over time. The GRU model, through its gating mechanism, models short-term temporal change patterns in injury feature sequences, thereby identifying local mutations in the injury feature sequences, such as short-term fluctuations in heart rate and respiratory rate. By integrating these two algorithms, the system can dynamically fuse spatiotemporal features, enhance feature representation using local information, and predict trends using global information, thus achieving injury severity assessment for the injured person.

[0050] When processing long-series multi-source data, the Informer model excels in the comprehensiveness of feature extraction and its ability to capture long-term dependencies, better integrating complex correlations among various vital signs of the injured. The GRU model, on the other hand, excels in handling short-term time-series fluctuations and rapidly responding to local data changes, thus supplementing the model's ability to capture short-term dynamic features. Based on the complementary advantages of these two individual models, the outputs of the Informer and GRU modules are weighted and fused to obtain the final evaluation result. The weights are determined through multiple optimizations to obtain the model with optimal performance. The fusion strategy expression is:

[0051] The model was trained multiple times, and the best performance was observed when the weight coefficient was 0.8. Therefore, the weight coefficient was ultimately determined to be 0.8.

[0052] During training, the massive amount of time-series data on the vital signs of the injured is first preprocessed, including data cleaning (removing outliers and filling in missing values) and data standardization (mapping the data to a uniform interval and eliminating the influence of units). Then, the preprocessed data is divided into training set, validation set and test set. The training set accounts for 60% and is used for iterative optimization of model parameters; the validation set accounts for 20% and is used to monitor overfitting during model training and adjust hyperparameters such as learning rate and batch size in a timely manner; the test set accounts for 20% and is used to evaluate the generalization ability of the model.

[0053] The intelligent assessment module outputs the final injury assessment results to the life support decision-making module, providing a basis for subsequent rescue decisions. Assessment results include: injury level: minor injury, moderate injury, serious injury, and critical injury.

[0054] Figure 6 The image shows the confusion matrix of the evaluation results of the ensemble learning model (i.e., the injury severity classification model). Figure 7 The accuracy, precision, recall, and F1 score of the model were all above 80%, indicating that the model has high predictive accuracy and stability, thus providing a reference for the implementation of subsequent treatment strategies.

[0055] like Figure 8 As shown, the life support decision module is used to input the environmental feature sequence, physiological feature sequence and injury level classification results into a pre-constructed medical knowledge graph for retrieval and matching, obtain a set of treatment measures, prioritize each treatment measure in the set of treatment measures, obtain a life support decision scheme, and complete unmanned injury assessment.

[0056] The medical knowledge graph was constructed from data sources including the A+ Medical Encyclopedia website, the Merck Handbook website, the Tactical Combat Wound Care Manual, and the Practical Pre-hospital Emergency Care Manual. The knowledge graph construction involved three steps: knowledge acquisition, knowledge processing, and knowledge storage. The Neo4j graph database was used to store nodes and relationships.

[0057] The life support decision-making module maps input data to entity nodes in a knowledge graph, retrieves a set of treatment measures based on relational edges, and then performs priority reasoning using a predefined clinical rule base. Finally, it outputs a personalized life support decision plan containing treatment measures and their implementation order. The decision-making process adopts a three-layer architecture: "data input layer - retrieval and reasoning layer - decision output layer." The data input layer is responsible for multimodal data fusion; the retrieval and reasoning layer queries the knowledge base using Cypher language and performs logical reasoning using a rule engine; and the decision output layer integrates various information to output the final life support decision plan, as detailed below: (1) Knowledge acquisition Before constructing the knowledge graph, relevant knowledge needs to be crawled from the aforementioned medical resources. This was implemented using a Python environment for web scraping. During the data extraction phase, the Requests library was used to send HTTP requests and retrieve webpage content. In the data parsing phase, the Lxml library was used to convert Requests objects into objects that could be parsed using XPath. The parsed data was then converted using the Pandas library and stored in a CSV file. During the crawling process, some websites have anti-scraping mechanisms; the Webdriver Selenium library can be used to simulate browser operations and bypass some simple anti-scraping strategies.

[0058] (2) Knowledge processing Data annotation was performed using the Sprite Annotation Assistant, and annotation standards were developed with reference to the "Medical Entity Standard Annotation for Medical Text Processing." The annotation process followed the principles of simplicity, ease of operation, and consistency. Annotation primarily focused on entities and relationships. Entities included injuries, symptoms, vital signs, and treatment measures, while relationships included <injury, symptoms>, <injury, vital signs>, <injury, treatment measures>, <symptoms, treatment measures>, and <vital signs, treatment measures>.

[0059] (3) Knowledge storage After defining entities and relationships, the crawled data is structured into a knowledge graph according to the rules and imported into the database. The data is then stored using the Neo4j graph database.

[0060] The specific steps are as follows: 1) First, connect to the graph database and create a node for each input parameter, namely: injury, symptoms, vital signs, and treatment measures.

[0061] 2) Check if the node already exists in the graph database. If it does not exist, create the node and add it to the graph database.

[0062] 3) Creating a Relationship: a. The injury affects vital signs.

[0063] b. The injury manifests as symptoms.

[0064] c. Symptoms indicate treatment measures.

[0065] d. Treatment of injuries.

[0066] 4) Check if the relationship already exists in the graph database. If it does not exist, create the relationship between the corresponding nodes and add it to the graph database.

[0067] 5) After creating the nodes and relationships, read the crawled CSV file and then start inserting the data into the created nodes.

[0068] Specific implementation process of the life support decision-making module: The module receives injury severity, vital signs, bleeding site information, and voice information from the environmental information sensing module and the injury information perception module. First, the real-time data is converted into entities in a knowledge graph. Using knowledge graph technology, a three-layer architecture of "data input layer - retrieval and reasoning layer - decision output layer" is constructed, as follows: (1) Data input layer: The input layer is responsible for fusing multimodal data, including injury severity (3 for critical injury, 2 for serious injury, 1 for moderate injury, and 0 for minor injury), vital signs (respiratory rate, heart rate, and body temperature, with corresponding values), bleeding site information (bleeding from the head, limbs, and trunk), and voice information (whether there is difficulty breathing, severe pain, or coldness; 1 for yes, 0 for no). The input layer receives raw data in different formats transmitted by the communication module, decodes it according to the protocol specifications, extracts information from each field, and uses a dictionary structure for data mapping.

[0069] (2) Decision-making reasoning layer: The Neo4j knowledge base is queried using the Cypher language. Based on the input information, relevant structural information is searched and the resulting treatment measures are stored in the variable required. If other entity nodes are activated during the query process, they are saved in the variable other.

[0070] The rule engine uses vital signs, voice information, and pathological mechanisms in other variables, combined with a predefined clinical rule base (in which the priority of measures is defined by integrating various physiological indicators and the severity of the injury; for example, if it is a critical injury, the priority of administering anti-shock injection is the highest), to trigger chain-like logical reasoning and thus output a conclusion.

[0071] (3) Decision output layer: By integrating information on the injuries, binary voice information, and the severity of injuries, and by linking multiple dimensions of information such as pathological mechanisms, vital signs, and treatment plans through a knowledge graph, the rule engine matches treatment measures and adjusts their priorities, thereby providing a reasonable life support decision plan (injecting anti-shock, injecting sedatives, injecting respiratory stimulants, injecting analgesics, spraying hemostatic agents, and spraying warming agents).

[0072] Experimental Verification: After administering life support medication, the injured person's condition was continuously monitored, and an intelligent injury severity classification was performed again. The injury severity was categorized into four levels: 0, 1, 2, and 3, with higher values ​​indicating more severe injuries. If the difference in severity level before and after treatment was greater than or equal to 0, it indicated that the injured person's condition was stabilizing; if the difference was less than or equal to -2, it indicated that the injured person's condition was rapidly deteriorating; otherwise, it was considered a slow deterioration.

[0073] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0074] This embodiment also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes a computing component and an iterative component, capable of model calculation and model updating). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor described in this embodiment can be used for the operation of an unmanned injury assessment method.

[0075] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the unmanned injury assessment method in the above embodiment.

[0076] This embodiment also provides a computer program product, which includes a computer program that, when executed by a processor, implements the corresponding steps of the unmanned injury assessment method described in the above embodiment.

[0077] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. An unmanned injury assessment method, characterized in that, Includes the following steps: Collect environmental parameters and physiological status information of the wounded; The environmental parameters and physiological state information are integrated and normalized to obtain environmental feature sequences and physiological feature sequences. The physiological feature sequence is input into the injury severity classification model to obtain the injury severity classification result. The injury severity classification model consists of an Informer model and a GRU model. The environmental feature sequence, physiological feature sequence, and injury severity classification results are input into a pre-constructed medical knowledge graph for retrieval and matching to obtain a set of treatment measures. The treatment measures in the set are then prioritized to obtain a life support decision-making scheme, thus completing the unmanned injury assessment.

2. The unmanned injury assessment method according to claim 1, characterized in that, The Informer model includes a first input layer, a first feature extraction layer, and a first output layer connected in sequence. The first feature extraction layer is configured with Prob sparse self-attention. The GRU model includes a second input layer, a second feature extraction layer, and a second output layer connected in sequence. The second feature extraction layer includes several layers of unidirectional GRU units.

3. The unmanned injury assessment method according to claim 2, characterized in that, The step of inputting the physiological feature sequence into the injury severity classification model to obtain the injury severity classification result specifically includes: The physiological feature sequence is input into the first input layer, and after the long-term dependent features in the physiological feature sequence are extracted by the first feature extraction layer, the sequence is then subjected to nonlinear transformation by the first output layer to obtain the first feature vector. The physiological feature sequence is input into the second input layer, and the short-term temporal features in the physiological feature sequence are extracted by the second feature extraction layer. Then, the sequence is nonlinearly transformed by the second output layer to obtain the second feature vector. The first feature vector and the second feature vector are weighted and fused according to a preset weight to obtain a fused feature vector. The fused feature vector is then input into a classifier to obtain the injury level classification result.

4. The unmanned injury assessment method according to claim 1, characterized in that, The medical knowledge graph is pre-constructed through the following steps: Acquire medical knowledge data and identify and extract medical entities and their semantic relationships from the medical knowledge data; Create a node system, which includes: injury nodes, symptom nodes, vital sign status nodes, and treatment measure nodes; The medical entities are transformed into nodes of corresponding types as entity nodes. Relationship edges are established between the entity nodes according to the semantic relationships. The entity nodes and relationship edges are stored in a graph database to obtain a medical knowledge graph.

5. The unmanned injury assessment method according to claim 4, characterized in that, The steps of inputting the environmental feature sequence, physiological feature sequence, and injury severity classification results into a pre-constructed medical knowledge graph for retrieval and matching to obtain a set of treatment measures, and prioritizing the various treatment measures in the set to obtain a life support decision-making scheme, specifically include: The environmental feature sequence, physiological feature sequence, and injury severity classification results are mapped to entity nodes, and entity relationship retrieval is performed based on the relation edges to obtain a set of treatment measures; Based on a pre-defined clinical rule base, the execution priority of each treatment measure in the set of treatment measures is determined and sorted to obtain the treatment measures and their corresponding implementation order, which serves as a life support decision-making scheme.

6. The unmanned injury assessment method according to claim 1, characterized in that, The environmental parameters include: air pressure, temperature, humidity, light intensity, magnetic field strength, and sound intensity; The physiological state information includes: vital signs information, injury site information, and state of consciousness information. The vital signs information includes: respiratory rate, heart rate, and body temperature. The steps of integrating and normalizing the environmental parameters and physiological state information to obtain environmental feature sequences and physiological feature sequences specifically include: Obtain the maximum and minimum values ​​of the environmental parameters and physiological state information over time; Based on the maximum and minimum values, the environmental parameters and physiological state information are scaled to a preset range through normalization to obtain environmental feature sequences and physiological feature sequences.

7. An unmanned injury assessment system, characterized in that, include: The environmental information and injury information sensing module is used to collect environmental parameters and physiological status information of the injured. The data processing module is used to integrate and normalize the environmental parameters and physiological state information to obtain environmental feature sequences and physiological feature sequences. The intelligent injury assessment module is used to input the physiological feature sequence into the injury level classification model to obtain the injury level classification result. The injury level classification model consists of an Informer model and a GRU model. The life support decision module is used to input the environmental feature sequence, physiological feature sequence and injury level classification results into a pre-constructed medical knowledge graph for retrieval and matching, obtain a set of treatment measures, prioritize each treatment measure in the set of treatment measures, obtain a life support decision scheme, and complete unmanned injury assessment.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the unmanned injury assessment method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the unmanned injury assessment method according to any one of claims 1 to 6.

10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the unmanned injury assessment method according to any one of claims 1 to 6.

Images