Method and system for evaluating injury condition of sick and wounded in battlefield environment

Abstract

The invention provides a method and a system for evaluating the injury condition of the sick and wounded in a battlefield environment. The method comprises the following steps: collecting instant physiological image data through portable medical imaging equipment, performing data cleaning and formatting processing, and generating standardized physiological image data; based on the standardized physiological image data, enhancing a key physiological region signal by applying a region attention mechanism algorithm, improving accuracy by adopting a transfer learning technology, and generating a preliminary injury level report; based on the preliminary injury condition grade report, a stacking generalization algorithm is applied, a plurality of prediction results are combined, and a life support measure scheme is generated by integrating visual and non-visual information through a secondary learner and adopting a multi-modal fusion technology; and based on the life support measure scheme, synchronizing to a remote medical command center in real time, and generating an injury condition assessment result of the sick and wounded. According to the technical scheme provided by the invention, the injury assessment speed and accuracy of the sick and wounded in a battlefield environment are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of military medical technology, and in particular to a method and system for assessing the injury of a sick or wounded person in a battlefield environment. Background Art

[0002] With the increasing complexity of modern warfare and emergencies, rapid and accurate assessment of the injuries of the wounded and sick in battlefield environments is crucial to improving the success rate of treatment. The application of portable medical imaging equipment makes it possible to collect physiological image data instantly on the front line, but these data often require preliminary data cleaning and formatting to generate standardized physiological image data. In addition, due to the particularity of the battlefield environment, the assessment system not only needs to have high accuracy, but also needs to be able to respond quickly under limited resources, and synchronize the assessment results to the remote medical command center in real time through a secure communication protocol, so that the rear team can make timely decisions.

[0003] Currently, battlefield medical treatment usually relies on traditional medical image analysis and manual evaluation. Some more advanced systems have begun to apply machine learning algorithms to assist diagnosis, such as using regional attention mechanisms to enhance signals in key physiological regions, and using transfer learning techniques to improve accuracy through the knowledge transfer capabilities of pre-trained models. However, the functions of these systems are relatively simple, mainly focusing on image analysis, and lacking comprehensive processing of multimodal information (such as visual and non-visual information).

[0004] Existing solutions have obvious deficiencies in practical applications. Traditional manual assessment methods are time-consuming and easily affected by human factors, resulting in insufficient speed and accuracy in injury assessment. Although some advanced systems have introduced machine learning algorithms, their adaptability and robustness are still limited in complex and changing battlefield environments. In addition, most existing systems fail to make full use of multiple information sources (such as physiological parameters, environmental conditions, etc.), which limits their ability to refine injury classification and formulate personalized life support measures. Therefore, there is an urgent need for a new method that can integrate multiple information sources and achieve rapid and accurate injury assessment. Summary of the invention

[0005] The embodiments of the present application provide a method and system for assessing the injury of the sick and wounded in a battlefield environment, so as to solve the problem of insufficient speed and accuracy of injury assessment in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for assessing the injury of a patient in a battlefield environment, comprising:

[0007] Through portable medical imaging equipment, real-time physiological image data of the wounded and sick are collected on the battlefield, and preliminary data cleaning and formatting are performed to generate standardized physiological image data;

[0008] Based on the standardized physiological image data, the regional attention mechanism algorithm is used to enhance the signals of key physiological regions, suppress irrelevant information, and transfer learning technology is used to improve the accuracy through the knowledge transfer capability of the pre-trained model to generate a preliminary injury level report;

[0009] Based on the preliminary injury grade report, a stacked generalization algorithm is used to combine the prediction results of multiple heterogeneous classifiers, the outputs of each primary classifier are integrated through a secondary learner, and multimodal fusion technology is used to integrate visual and non-visual information, refine the injury grade of the injured and sick, and generate a life support measure plan;

[0010] Based on the life support measures plan, the injury assessment results of the injured and sick are generated in real time through synchronization to the remote medical command center through a secure communication protocol.

[0011] Optionally, based on the standardized physiological image data, a regional attention mechanism algorithm is used to enhance the key physiological region signals, suppress irrelevant information, adopt transfer learning technology, improve accuracy through the knowledge transfer capability of the pre-trained model, and generate a preliminary injury level report, including:

[0012] Based on the standardized physiological image data, preliminarily identifying and marking each area in the image to generate image data labels;

[0013] Based on the image data labeling, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions in the image, suppress background and other non-related regional information, and generate focused image data;

[0014] Based on the focused image data, transfer learning technology is used to pre-train the model using a large-scale medical image data set, and the pre-trained model knowledge is transferred to the current task, so as to quickly adapt to the situation of the wounded and sick in a specific battlefield environment and generate knowledge transfer data;

[0015] Based on the knowledge transfer data, combined with clinical standards and empirical rules, the specific injury level of the injured and sick is further refined to generate a preliminary injury level report.

[0016] Optionally, based on the image data labeling, using a regional attention mechanism algorithm to enhance the key physiological region signals in the image, suppress background and other non-related region information, and generate focused image data, including:

[0017] Based on the image data labels, an attention map is constructed, weight values ​​are assigned to all pixel regions to reflect importance, and an initial attention distribution is generated;

[0018] Based on the initial attention distribution, the regional attention mechanism algorithm is used to adjust the feature representation in the original image, amplify the key physiological area features by weighted summation, weaken the influence of background and other non-related area information, and generate enhanced feature representation;

[0019] Based on the enhanced feature representation, convolutional neural network technology is used to further mine deep features, keep the spatial position of key physiological signals consistent, and generate deep feature maps;

[0020] Based on the deep feature map, fusion reconstruction is performed in combination with the original image information to ensure that key physiological areas are highlighted and generate focused image data.

[0021] Optionally, based on the focused image data, transfer learning technology is used to pre-train the model using a large-scale medical image data set, and the pre-trained model knowledge is transferred to the current task to quickly adapt to the situation of the wounded and sick in a specific battlefield environment, and generate knowledge transfer data, including:

[0022] Based on the focused image data, a large-scale medical image dataset is selected to pre-train the model to generate a basic pre-trained model;

[0023] Based on the basic pre-trained model, the model is fine-tuned using transfer learning technology, the bottom layer weights are frozen, and the top layer is retrained using the focused image data to generate a preliminary adaptation model;

[0024] Based on the preliminary adaptation model, the learning rate is adjusted and regularization is introduced to prevent overfitting, so as to ensure that the model accurately reflects the conditions of the wounded and sick in the battlefield environment and generate an optimized adaptation model;

[0025] Based on the optimized adaptation model, the focused image data is further processed, inheriting the generalization ability of the basic pre-trained model to generate knowledge transfer data.

[0026] Optionally, based on the preliminary injury level report, a stacked generalization algorithm is used, the prediction results of multiple heterogeneous classifiers are combined, the outputs of each primary classifier are integrated through a secondary learner, and multimodal fusion technology is used to integrate visual and non-visual information, refine the injury level of the injured and sick, and generate a life support measure plan, including:

[0027] Based on the preliminary injury level report, an integrated model is constructed to integrate multiple heterogeneous classifiers to independently predict the injury level of the injured and sick and to generate heterogeneous classifier prediction results;

[0028] Based on the prediction results of the heterogeneous classifiers, a stacked generalization algorithm is used to integrate the outputs of the primary classifiers through the secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result;

[0029] Based on the comprehensive prediction results, multimodal fusion technology is used to combine visual information with non-visual information to further analyze the specific conditions of the injured and sick and generate detailed injury classification;

[0030] Based on the detailed injury classification, the severity and type of the injury are comprehensively considered, the preset life support strategy template is matched, and a life support measure plan is generated.

[0031] Optionally, based on the prediction results of the heterogeneous classifiers, a stacked generalization algorithm is used to integrate the outputs of each primary classifier through a secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result, including:

[0032] Based on the prediction results of the heterogeneous classifier, the prediction results are input into the stacked generalization model as new features to generate stacked layer input data;

[0033] Based on the stacked layer input data, a stacked generalization algorithm is used to train the secondary learner, learn to integrate the prediction results of each primary classifier to capture the complementary information between different classifiers, and generate a secondary learner model;

[0034] Based on the secondary learner model, optimizing secondary learner parameters, minimizing overfitting risk, and generating an optimized secondary learner;

[0035] Based on the optimized secondary learner, new patient data is input for prediction to improve overall prediction accuracy and stability and generate comprehensive prediction results.

[0036] Optionally, the life support measure plan is synchronized to a remote medical command center in real time through a secure communication protocol to generate an injury assessment result of the injured or sick, including:

[0037] Based on the life support measure plan, perform structured processing, convert into a unified data format, add necessary metadata, and generate a standardized data package;

[0038] Based on the standardized data packet, a secure communication protocol is used for encryption processing to ensure confidentiality and integrity of the transmission process and generate an encrypted data packet;

[0039] Based on the encrypted data packet, the encrypted data packet is sent to the remote medical command center in real time through a reliable network connection to generate transmission confirmation information;

[0040] Based on the transmission confirmation information, the remote medical command center receives and decrypts the encrypted data packet to generate an injury assessment result of the injured or sick.

[0041] In a second aspect, an embodiment of the present application provides a system for assessing the condition of a wounded or sick person in a battlefield environment, comprising:

[0042] The collection module is used to collect the real-time physiological image data of the wounded and sick at the battlefield through portable medical imaging equipment, perform preliminary data cleaning and formatting, and generate standardized physiological image data;

[0043] An enhancement module is used to enhance the key physiological region signals based on the standardized physiological image data by using a regional attention mechanism algorithm, suppress irrelevant information, adopt transfer learning technology, improve accuracy through the knowledge transfer capability of the pre-trained model, and generate a preliminary injury level report;

[0044] A fusion module is used to use a stacked generalization algorithm based on the preliminary injury level report, combine the prediction results of multiple heterogeneous classifiers, integrate the outputs of each primary classifier through a secondary learner, and use multimodal fusion technology to integrate visual and non-visual information, refine the injury level of the injured and sick, and generate a life support measure plan;

[0045] A generation module is used to generate injury assessment results of the injured and sick based on the life support measures plan and synchronized to the remote medical command center in real time through a secure communication protocol.

[0046] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for assessing the injury of a sick and wounded person in a battlefield environment as described in one of the first aspects.

[0047] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for assessing the injury of a sick and wounded person in a battlefield environment as described in the first aspect.

[0048] In an embodiment of the present application, a portable medical imaging device is used to collect real-time physiological image data of the wounded and sick at the battlefield, and preliminary data cleaning and formatting processing are performed to generate standardized physiological image data; based on the standardized physiological image data, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions, suppress irrelevant information, and transfer learning technology is used to improve accuracy through the knowledge transfer capability of the pre-trained model to generate a preliminary injury level report; based on the preliminary injury level report, a stacked generalization algorithm is used, combined with the prediction results of multiple heterogeneous classifiers, the outputs of each primary classifier are integrated through a secondary learner, and multimodal fusion technology is used to integrate visual and non-visual information, refine the injury classification of the wounded and sick, and generate a life support measures plan; based on the life support measures plan, it is synchronized to the remote medical command center in real time through a secure communication protocol to generate the injury assessment results of the wounded and sick. Real-time physiological image data is collected through portable medical imaging equipment, and preliminary data cleaning and formatting are performed to ensure data quality and consistency; the regional attention mechanism algorithm is used to enhance signals in key physiological areas, while suppressing irrelevant information to ensure the accuracy of injury assessment; transfer learning technology is used to improve the accuracy of assessment of the condition of the wounded and sick in specific battlefield environments through the knowledge transfer capability of the pre-trained model; after generating a preliminary injury level report, the injury classification of the wounded and sick is further refined through the stacked generalization algorithm and multimodal fusion technology to ensure the comprehensiveness and accuracy of the assessment results; the assessment results are synchronized to the remote medical command center in real time through a secure communication protocol, realizing timely sharing of information, which is conducive to rapid decision-making and resource allocation.

[0049] Furthermore, by enhancing the signals of key physiological areas through the regional attention mechanism, important health information of the wounded can be captured more accurately, while suppressing the information of the background and other non-relevant areas reduces interference factors; the use of transfer learning technology enables the pre-trained model to quickly adapt to the specific battlefield environment, improving the efficiency and accuracy of injury assessment; the injury level is refined by combining clinical standards and empirical rules, and the generated preliminary injury level report provides an important reference basis for subsequent treatment, which helps to improve the speed and accuracy of diagnosis.

[0050] Furthermore, by building an integrated model to integrate multiple heterogeneous classifiers, we can fully utilize the advantages of different classifiers and improve the accuracy and stability of the overall prediction; by using the stacked generalization algorithm, the output of the primary classifier is integrated through the secondary learner, which effectively solves the possible deviation problem of a single model; the multimodal fusion technology is used to integrate visual and non-visual information, making the injury assessment more comprehensive and detailed. The life support measures plan finally generated can better match the specific conditions of the injured and the sick, thereby improving the effectiveness and pertinence of life support measures.

[0051] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A flowchart of a method for assessing the condition of a wounded or sick person in a battlefield environment provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of the structure of a system for assessing the condition of a wounded or sick person in a battlefield environment provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0057] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0059] Figure 1A flowchart of a method for assessing the condition of a wounded or sick person in a battlefield environment is provided for an embodiment of the present application. Figure 1 As shown, the method includes:

[0060] 101. Collect instant physiological image data of the wounded and sick at the battlefield through portable medical imaging equipment, perform preliminary data cleaning and formatting, and generate standardized physiological image data;

[0061] In this step, portable medical imaging equipment includes portable ultrasound machines, X-ray machines or thermal imagers, etc., which are used to instantly obtain physiological image data of the wounded and sick on the battlefield. These devices can quickly generate high-quality images in complex environments, covering various physiological characteristics from the skin surface to internal organs.

[0062] Physiological image data refers to images collected by the above-mentioned devices, such as X-rays, CT scans, ultrasound images and wound photos, which provide detailed information about the internal organ structure, external injuries and lesion areas, helping doctors to make more comprehensive diagnoses.

[0063] Preliminary data cleaning and formatting refers to the preprocessing of raw image data, removing noise and irrelevant information, and converting the data into a unified standard format. Specific operations include resizing, grayscale normalization, background removal, etc., to ensure that data from different sources are comparable and consistent.

[0064] Standardized physiological image data is pre-processed image data with consistent format and quality standards, and can be directly used in subsequent analysis and evaluation processes.

[0065] In an embodiment of the present application, assuming a rescue site with a complex and ever-changing environment, first, the intelligent medical treatment box group is equipped with a variety of portable imaging devices, which can collect physiological image data of the injured and sick in real time; secondly, the system's built-in data processing module automatically identifies and removes noise points and irrelevant background information in the image; thirdly, all image data are uploaded to the central server for unified resizing and grayscale normalization processing; finally, the standardized image data are integrated into a unified database for subsequent analysis.

[0066] 102. Based on the standardized physiological image data, the regional attention mechanism algorithm is used to enhance the signals of key physiological regions, suppress irrelevant information, and transfer learning technology is used to improve the accuracy through the knowledge transfer capability of the pre-trained model to generate a preliminary injury level report;

[0067] In this step, the regional attention mechanism algorithm is a deep learning technology that aims to enhance the signal of specific areas in the image (such as injured areas or diseased tissue) while suppressing irrelevant information. This step improves the accuracy and efficiency of diagnosis by focusing on important areas.

[0068] Transfer learning technology uses the knowledge transfer capabilities of pre-trained models to enable the models to quickly adapt to specific battlefield environments. This method pre-trains models through large-scale medical image datasets and then applies these models to current tasks, thereby improving the accuracy of injury assessment.

[0069] Knowledge transfer data is data generated through transfer learning technology. It contains the results of applying the pre-trained model in a new environment, which helps to improve the generalization ability and adaptability of the model.

[0070] The preliminary injury level report is a brief report generated based on standardized image data and transfer learning results. It combines clinical standards and empirical rules to refine the specific injury level of the injured and guide initial treatment measures.

[0071] In an embodiment of the present application, assuming that at a front-line medical station with limited resources, first, the medical team uses a portable ultrasound device to quickly scan multiple wounded and sick people to obtain standardized physiological image data; second, the system's built-in regional attention mechanism algorithm automatically identifies and highlights key physiological areas, such as the heart and lungs; third, the application of transfer learning technology enables the pre-trained model to quickly adapt to battlefield conditions and optimize injury assessment; finally, the system automatically generates a detailed preliminary injury level report to provide timely guidance for on-site first aid.

[0072] Optionally, in step 102, based on the standardized physiological image data, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions, suppress irrelevant information, and transfer learning technology is used to improve the accuracy through the knowledge transfer capability of the pre-trained model to generate a preliminary injury level report, including: based on the standardized physiological image data, preliminary identification and labeling of each region in the image to generate image data labels; based on the image data labels, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions in the image, suppress background and other irrelevant regional information, and generate focused image data; based on the focused image data, transfer learning technology is used to pre-train a model with a large-scale medical image data set, and the knowledge of the pre-trained model is transferred to the current task to quickly adapt to the situation of the wounded and sick in a specific battlefield environment to generate knowledge transfer data; based on the knowledge transfer data, in combination with clinical standards and empirical rules, the specific injury level of the wounded and sick is further refined to generate a preliminary injury level report.

[0073] In this step, the standardized physiological image data refers to the consistent image data after preprocessing for further advanced processing.

[0074] Image data labeling is data generated by preliminary identification and labeling of each area in the image, which provides a basis for subsequent enhancement processing.

[0075] The focused image data is generated through a regional attention mechanism algorithm, which enhances the key area signal and suppresses the image data of background information.

[0076] Clinical standards and empirical rules refer to the experience and norms accumulated over a long period of time in medical practice, which provide a basis for refining the specific injury levels of the wounded and sick.

[0077] First, the regions in the standardized physiological image data are preliminarily identified and marked to generate detailed image data marks. Secondly, based on these marks, the regional attention mechanism algorithm is used to enhance the signals of key physiological regions in the image, suppress the background and information of other irrelevant regions, and generate focused image data. Thirdly, the model pre-trained with a large-scale medical image dataset is used to transfer the knowledge of the pre-trained model to the current task, quickly adapt to the situation of the wounded and sick in a specific battlefield environment, and generate knowledge transfer data. Finally, the specific injury level of the wounded and sick is further refined by combining clinical standards and empirical rules to generate a detailed preliminary injury level report.

[0078] Optionally, based on the image data labeling, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions in the image, suppress the background and other non-related regional information, and generate focused image data, including: based on the image data labeling, constructing an attention map, assigning weight values ​​to all pixel regions to reflect importance, and generating an initial attention distribution; based on the initial attention distribution, using the regional attention mechanism algorithm, adjusting the feature representation in the original image, amplifying the key physiological region features by weighted summation, weakening the background and other non-related regional information, and generating an enhanced feature representation; based on the enhanced feature representation, using convolutional neural network technology to further mine deep features, keep the spatial position of key physiological signals consistent, and generate a deep feature map; based on the deep feature map, combined with the original image information for fusion reconstruction, to ensure that the key physiological regions are highlighted, and generate focused image data.

[0079] Based on the focused image data, the transfer learning technology is used to pre-train the model through a large-scale medical image data set, and the pre-trained model knowledge is transferred to the current task to quickly adapt to the situation of the wounded and sick in a specific battlefield environment, and generate knowledge transfer data, including: based on the focused image data, a large-scale medical image data set is selected to pre-train the model to generate a basic pre-trained model; based on the basic pre-trained model, the transfer learning technology is used to fine-tune the model, freeze the bottom layer weights, and retrain the top layer using the focused image data to generate a preliminary adaptation model; based on the preliminary adaptation model, the learning rate is adjusted and regularization is introduced to prevent overfitting, so as to ensure that the model accurately reflects the situation of the wounded and sick in the battlefield environment, and an optimized adaptation model is generated; based on the optimized adaptation model, the focused image data is further processed to inherit the generalization ability of the basic pre-trained model to generate knowledge transfer data.

[0080] In this step, constructing the attention map is the process of assigning weight values ​​to all pixel areas based on the image data labels, mathematically reflecting their importance to the task. These weight values ​​help the system identify which areas are more critical.

[0081] The initial attention distribution is generated based on the attention map, which reflects the importance of each pixel and provides a basis for subsequent feature adjustment.

[0082] The enhanced feature representation is achieved by adjusting the features in the original image and using a weighted summation method to amplify the characteristics of key physiological areas while weakening the background and other irrelevant information.

[0083] In this process, convolutional neural network technology plays the role of mining deep-level features, generating deep feature maps, and ensuring the spatial consistency of key physiological signals.

[0084] Deep feature mapping is a high-level feature representation extracted by convolutional neural networks, which retains the spatial structural information of key physiological signals and enhances the model's ability to understand complex images.

[0085] Fusion reconstruction refers to combining the original image information with the deep feature map to ensure that the final output image not only highlights the key areas, but also retains the consistency with the original image information to generate focused image data.

[0086] The preliminary injury level report is a brief report generated based on standardized image data and transfer learning results. It combines clinical standards and empirical rules to refine the specific injury level of the injured and guide initial treatment measures.

[0087] In an embodiment of the present application, first, standardized physiological image data are received, and based on these data, an attention map reflecting the importance of each region is created; secondly, using the regional attention mechanism algorithm, the system adjusts the image feature representation to highlight the key physiological areas while reducing the impact of background information; thirdly, a convolutional neural network is used to deeply analyze the image, extract deep features, and maintain the consistency of these features with the original spatial position; finally, through the fusion and reconstruction process, the final focused image data is generated, and the model is optimized through transfer learning to make it more suitable for the injury assessment needs of the wounded and sick in a battlefield environment.

[0088] Suppose during a military exercise in a remote area, first, the front-line medical team deployed a set of portable medical imaging equipment, which can quickly obtain high-quality imaging data of the wounded and sick. At the same time, this set of equipment automatically processes the images and creates an attention map that can distinguish key physiological areas from the background, making subsequent analysis more accurate; secondly, the system uses the regional attention mechanism algorithm to intelligently process the image, strengthen the characteristic performance of the injured part or diseased tissue, and reduce the interference of irrelevant information. The convolutional neural network integrated in the system further analyzes the image to ensure the complete retention of deep-level features and the consistency of spatial position, providing doctors with more detailed injury details; finally, by introducing transfer learning technology, the existing model can be quickly fine-tuned to adapt to battlefield conditions. The generated knowledge transfer data not only improves the generalization ability of the model, but also provides the front-line medical team with accurate preliminary injury level reports, helping them to formulate effective treatment plans in a timely manner.

[0089] This application takes into account the problems in the prior art of insufficiently accurate extraction of key physiological region features and serious interference from background and other non-related region information, so the invention embodiment proposes this optional solution, which introduces a regional attention mechanism algorithm and a complex mathematical formula design to solve the technical problem of accurately capturing key physiological signals in a complex image background. This method not only improves the accuracy of diagnosis, but also enhances the model's adaptability to different environments and conditions.

[0090] Optionally, based on the initial attention distribution, a regional attention mechanism algorithm is used to adjust the feature representation in the original image, amplify the key physiological region features by weighted summation, weaken the influence of background and other non-related region information, and generate enhanced feature representation, including:

[0091] Based on the initial attention distribution, extract the feature vector of each region through a convolutional neural network to obtain local information, which is used as a query vector and a key vector for linear transformation to enhance the expression ability and generate an attention score;

[0092] The attention score is calculated using the following formula:

[0093]

[0094] Among them, A ij represents the attention score between the i-th query vector and the j-th key vector; Q i is the i-th row vector in the query matrix; W q is the weight matrix of the query vector; b q is the bias term of the query vector; is the transpose of the jth column vector in the key matrix; W k is the weight matrix of the key vector; b k is the bias term of the key vector; V j is a value vector; W v is the weight matrix of the value vector; d k is the dimension of the key vector; b a is an additional bias term; σ represents the activation function softmax;

[0095] Based on the attention score, each regional attention score is multiplied by the corresponding value vector, and the key physiological regional features are highlighted by weighted summation, the influence of the background and other non-related regions is weakened, and a nonlinear activation function is introduced to enhance the discrimination of feature representation, so as to generate regional feature representation;

[0096] The regional feature representation is calculated using the following formula:

[0097]

[0098] Among them, R i represents the adjusted feature representation of the ith region; A ij is the calculated attention score; V j is the feature vector of each region in the original image; W v is the weight matrix of the value vector; b v is the bias term of the value vector; j is the index of the image region to be processed, from 1 to N; N is the total number of image regions to be processed; W 1 ,W 2 ,W 3 are different weight matrices, used to weight information at different levels; b 1 ,b 2 ,b 3 They are respectively 1 ,W 2 ,W 3 The corresponding bias term; α and β are proportional coefficients that control the influence of sine and cosine terms; tanh and sin, cos functions are used to introduce nonlinear transformations to enhance the model's expressiveness; b r is the bias term in the final feature adjustment process;

[0099] Based on the regional feature representation, multiple weight matrix and bias term combinations are introduced, combined with sine and cosine functions, to increase the diversity and expression ability of feature representation, retain key physiological area information to the greatest extent, and generate enhanced feature representation.

[0100] This method aims to adjust the feature representation in the original image based on the initial attention distribution using the regional attention mechanism algorithm. The features of key physiological areas are amplified by weighted summation, while the influence of background and other irrelevant information is weakened, thereby generating an enhanced feature representation. This method aims to ensure that the medical team can focus more on the key parts of the injured and provide more accurate diagnostic support. In addition, a convolutional neural network is introduced to extract the feature vector of each region, and combined with a linear transformation to generate an attention score, so that the model can more intelligently distinguish between important and non-important areas. Finally, through a series of carefully designed mathematical formulas, the optimization of feature representation is achieved, providing a solid foundation for subsequent life support measures.

[0101] In the attention score, the query vector transformation term Q i W q +b q : By performing a linear transformation on the query vector, its expressive power is enhanced, so that the model can better capture the features of the region of interest in the image; key vector transformation term Similarly, the key vector is linearly transformed so that the model can more accurately distinguish the importance of different regions; the value vector transformation term V j W v +b a : Linearly transform the value vector to enhance its contribution in the final feature representation; scaling factor term Used to stabilize the gradient, prevent the value from being too large or too small, and ensure the stability of the calculation;

[0102] Among them, the query matrix Weight matrix W q is the pre-trained weight matrix; the key matrix Weight matrix W k is a pre-trained weight matrix; the value vector Weight matrix W v is the pre-trained weight matrix; the bias term b q ,b k ,b a It is automatically learned during the training process; dimension d k It is set according to the dimension of the key vector, for example 64;

[0103] In the regional feature representation, the weighted sum The weighted sum method is used to highlight the key regional features, ensuring that the information of the important areas is fully amplified; nonlinear activation term The hyperbolic tangent function is introduced to enhance feature differentiation, so that the model can better capture subtle differences; the sine transform term The sine transform increases the diversity of feature representation and helps the model understand periodic changes; the cosine transform The cosine transform further enhances the diversity of feature representation and complements the effect of the sine transform; the bias term b r : Adjust the final feature representation to ensure that the output features are more stable;

[0104] Among them, the attention score A ij Calculated by the attention score formula; value vector Weight matrix W v is the pre-trained weight matrix; the bias term b v It is automatically learned during the training process; the weight matrix W 1 ,W 2 ,W 3 is the pre-trained weight matrix; the bias term b 1 ,b 2 ,b 3 is automatically learned during the training process; the proportional coefficients α and β are set according to experimental results; the bias term b r It is automatically learned during the training process;

[0105] Suppose a medical rescue team needs to process the injury data of a group of wounded and sick people in a battlefield environment; suppose the standardized initial eigenvector X = [0.1, 0.2, 0.3, 0.4, 0.5]; query matrix Q i =W Q X, key matrix Value vector V j =W V X, where is a pre-trained weight matrix; assuming the key vector dimension is d k =64, bias term b q =0.5, bias term b k =0.3, bias term b v =0.2, additional bias term b a =0.1; Assume that the weight matrix used to weight information at different levels and the corresponding bias term b 1 =0.5,b 2 =0.3,b 3 = 0.4 is also pre-trained; the proportional coefficient α = 0.7, β = 0.8, the bias term b in the final feature adjustment process r =0.1;

[0106]

[0107] Assuming the threshold is θ = [0.8, 0.8, 0.8, 0.8, 0.8], the final result R of the comprehensive feature representation is i = Each element of [0.95, 0.88, 0.94, 0.90, 0.93] is greater than the corresponding hypothetical threshold, which indicates that the current patient information performs well in terms of understanding and feature capture, and is suitable for further use or as the final information representation. Through the above steps, medical data can be effectively processed and high-quality feature representation can be extracted, thereby improving the accuracy and timeliness of the assessment of the health status of the patient.

[0108] 103. Based on the preliminary injury grade report, a stacked generalization algorithm is used to combine the prediction results of multiple heterogeneous classifiers, the outputs of each primary classifier are integrated through a secondary learner, and multimodal fusion technology is used to integrate visual and non-visual information, refine the injury grade of the injured and sick, and generate a life support measure plan;

[0109] In this step, the integrated model is a model that contains multiple heterogeneous classifiers, each of which independently predicts the injury of the injured and provides a unique perspective and prediction results.

[0110] The prediction results of heterogeneous classifiers are the prediction outputs generated by various classifiers based on different algorithms or data sources. They each provide different information, enhancing the diversity and accuracy of the overall prediction.

[0111] The stacked generalization algorithm is a machine learning method that integrates the outputs of primary classifiers through secondary learners, captures complementary information between different classifiers, and improves the accuracy and stability of the overall prediction.

[0112] The secondary learner is a model that relearns the prediction results of the primary classifier. It generates a comprehensive prediction result by learning the prediction results of multiple primary classifiers.

[0113] Multimodal fusion technology is a method that combines visual and non-visual information, making injury assessment more comprehensive and detailed. The life support measures ultimately generated can better match the specific conditions of the injured.

[0114] Refined injury classification is a more accurate injury assessment result generated through comprehensive analysis using multimodal fusion technology, providing a basis for formulating personalized treatment plans.

[0115] The life support measures plan is a specific treatment plan generated based on the detailed injury classification, ensuring that the most appropriate treatment and support can be provided according to the actual needs of the injured and sick.

[0116] In the embodiment of the present application, assuming that in a complex military exercise scenario, first, the medical command center deploys an advanced data analysis platform to integrate data from multiple sensors and imaging devices; secondly, the platform constructs an integrated model to combine the prediction results of multiple heterogeneous classifiers to independently predict the injury of each injured person; thirdly, through secondary learners and stacked generalization algorithms, the outputs of each primary classifier are integrated to generate a more accurate comprehensive prediction result; finally, multimodal fusion technology is used to integrate visual and non-visual information to generate detailed injury classification, and personalized life support measures are formulated accordingly.

[0117] Optionally, the method in step 103 is based on the preliminary injury level report, uses a stacked generalization algorithm, combines the prediction results of multiple heterogeneous classifiers, integrates the outputs of each primary classifier through a secondary learner, and uses multimodal fusion technology to integrate visual and non-visual information, refines the injury grade of the injured and generates a life support measure plan, including: based on the preliminary injury level report, constructs an integrated model to integrate multiple heterogeneous classifiers, and independently predicts the injuries of the injured and generates heterogeneous classifier prediction results; based on the heterogeneous classifier prediction results, uses a stacked generalization algorithm to integrate the outputs of each primary classifier through a secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result; based on the comprehensive prediction result, uses multimodal fusion technology to combine visual information with non-visual information, further analyzes the specific conditions of the injured and generates a refined injury grade; based on the refined injury grade, comprehensively considers the severity and type of the injury, matches the preset life support strategy template, and generates a life support measure plan.

[0118] Among them, based on the prediction results of the heterogeneous classifiers, the stacked generalization algorithm is used to integrate the outputs of each primary classifier through the secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result, including: based on the prediction results of the heterogeneous classifiers, as new features are input into the stacked generalization model to generate stacked layer input data; based on the stacked layer input data, the stacked generalization algorithm is used to train the secondary learner to learn and integrate the prediction results of each primary classifier to capture the complementary information between different classifiers and generate a secondary learner model; based on the secondary learner model, the secondary learner parameters are optimized to minimize the risk of overfitting and generate an optimized secondary learner; based on the optimized secondary learner, new injured and sick data are input for prediction to improve the overall prediction accuracy and stability and generate a comprehensive prediction result.

[0119] In this step, stacking layer input data refers to inputting the prediction results of the primary classifier as new features into the stacked generalization model, providing a rich source of information for the secondary learner.

[0120] The secondary learner model is trained by the secondary learner and the stacked generalization algorithm is used to integrate the prediction results of the primary classifier to capture the complementary information between different classifiers. The optimized model can better adapt to specific tasks.

[0121] Preset life support strategy templates are a series of predefined treatment plans that provide standardized medical response measures for different types of injuries and severity. These templates are designed by medical experts to quickly guide on-site medical teams to take appropriate first aid and follow-up treatment actions.

[0122] The risk of overfitting refers to the situation where the model performs too well on the training data, so that it cannot maintain the same performance on new, unseen data. When optimizing the secondary learner, it is crucial to reduce the risk of overfitting, which can be achieved through regularization, cross-validation and other techniques.

[0123] The comprehensive prediction result is generated by integrating the output of the primary classifier through the secondary learner. It comprehensively considers the information of all primary classifiers and improves the accuracy and stability of the prediction. This result not only reflects the current injury status of the injured, but also provides a solid foundation for subsequent treatment decisions.

[0124] In the embodiments of the present application, first, an integrated model is constructed based on the preliminary injury level report, and multiple heterogeneous classifiers are integrated to independently predict the injuries of the injured and sick. The prediction results of these classifiers are used as new features to input into the stacked generalization model to generate stacked layer input data; secondly, the stacked generalization algorithm is used to train the secondary learner to learn and integrate the prediction results of each primary classifier, capture the complementary information between different classifiers, and generate a secondary learner model; thirdly, by optimizing the parameters of the secondary learner, the risk of overfitting is minimized, and new data of the injured and sick are used for prediction, the overall prediction accuracy and stability are improved, and a comprehensive prediction result is generated; finally, multimodal fusion technology is used to combine visual and non-visual information to further refine the injury classification, and the preset life support strategy template is matched according to the refined results to generate a life support measure plan.

[0125] Suppose in an international peacekeeping mission, the medical team faces complex treatment needs of the wounded and sick from multiple countries. First, the medical command center receives the preliminary injury level reports sent back from the front line and quickly deploys a comprehensive assessment system. This system integrates multiple heterogeneous classifiers provided by medical institutions from different countries to independently predict the injuries of each wounded and sick person. Second, the prediction results of these classifiers are used as new features and input into a stacking generalization model to generate detailed input data for the stacking layer, ensuring the effective combination of data from different sources. Third, a secondary learner is trained through the stacking generalization algorithm to learn and integrate the outputs of the primary classifiers, capture the complementary information between different classifiers, generate an optimized secondary learner model, and optimize the parameters through strict cross-validation to reduce the risk of overfitting. Finally, the system uses new wounded and sick data for prediction to generate comprehensive prediction results. On this basis, multi-modal fusion technology is adopted, combining visual and non-visual information (such as physiological parameters, environmental factors), further analyzing the specific conditions of the wounded and sick, generating refined injury grading, and matching the preset life support strategy template according to the refined results to generate personalized life support measure plans, ensuring that the multinational medical team can implement efficient treatment measures in a coordinated manner.

[0126] This application considers that in the prior art, due to the problems of insufficient generalization ability of a single model and insufficient utilization of complementary information between different classifiers when evaluating the injuries of the wounded and sick, the invention embodiment proposes this alternative solution. By introducing the stacking generalization algorithm to train the secondary learner, the complementary information between different primary classifiers can be captured. This method not only improves the accuracy of diagnosis but also enhances the adaptability of the model to complex and changing environments.

[0127] Optionally, based on the input data of the stacking layer, the stacking generalization algorithm is used to train the secondary learner to learn and integrate the prediction results of each primary classifier to capture the complementary information between different classifiers and generate a secondary learner model, including:

[0128] Based on the input data of the stacking layer, high-dimensional feature vectors of each region are extracted;

[0129] Preprocessing is performed through the L2 normalization method to ensure the same scale between different high-dimensional feature vectors, and then input into multiple different primary classifiers for prediction to generate prediction results;

[0130] The prediction results are calculated through the following formula:

[0131]

[0132] where, is the prediction result of the i-th sample; w j is the weight of the j-th primary classifier; γ jis the scaling factor of the jth primary classifier, which is used to adjust the weight of the prediction result; f j (x i ) is the prediction result of the jth primary classifier for the i-th sample; j is the nonlinear adjustment factor of the jth primary classifier, which is used to further adjust the weight of the prediction result; j is the index of the primary classifier, ranging from 1 to N; N is the number of primary classifiers;

[0133] Based on the prediction results, weights and scaling factors are assigned to each primary classifier prediction result, nonlinear characteristics are further enhanced by exponential and logarithmic functions, and the weighted prediction results are normalized using a soft maximization function to ensure that the sum of all weights is 1 to generate an intermediate representation;

[0134] The intermediate representation is calculated using the following formula:

[0135]

[0136] Among them, H i is the intermediate representation of the i-th sample by the secondary learner; σ is the activation function Sigmoid or ReLU, which is used to introduce nonlinearity; α j is the weight of the jth primary classifier prediction result after nonlinear transformation; β j is the scaling factor of the jth primary classifier prediction result; γ j is the scaling factor of the jth primary classifier, used to adjust the weight of the prediction result; j is the nonlinear adjustment factor of the jth primary classifier, which is used to further adjust the weight of the prediction result; δ j is the bias term of the jth primary classifier prediction result; tanh is the hyperbolic tangent function, which is used to further introduce nonlinearity; θ is the weight of the sine term; η k is the weight of the kth auxiliary function; is the prediction result of the kth auxiliary function for the ith sample; sin is a sine function used to introduce periodic changes; φ is the bias term of the final output; f j (x i ) is the prediction result of the jth primary classifier for the i-th sample; is the weighted nonlinear integration prediction result of the i-th sample; N is the number of primary classifiers; M is the number of auxiliary functions;

[0137] Based on the intermediate representation, complex nonlinear mapping is performed on the intermediate representation through multiple layers of nonlinear functions to capture complementary information between different primary classifiers. After each layer of nonlinear transformation, bias terms and weight matrices are added to adjust the output. Nonlinear characteristics are further introduced through activation functions to generate a secondary learner model.

[0138] This method aims to train the secondary learner based on the stacked layer input data using the stacked generalization algorithm, integrate the prediction results of each primary classifier to capture the complementary information between different classifiers, and generate a secondary learner model. Specifically, we first extract the high-dimensional feature vector of each region and preprocess it through the L2 normalization method to ensure the consistency of scale between different high-dimensional feature vectors. Then, these feature vectors are input into multiple different primary classifiers for prediction to generate prediction results. Next, the nonlinear characteristics are further enhanced by exponential and logarithmic functions, and the weighted prediction results are normalized using the soft maximization function to ensure that the sum of all weights is 1 to generate an intermediate representation. Finally, the intermediate representation is subjected to complex nonlinear mapping through multiple layers of nonlinear functions. After each layer of nonlinear transformation, bias terms and weight matrices are added to adjust the output. Nonlinear characteristics are further introduced through activation functions to finally generate a secondary learner model.

[0139] In the prediction results, the weighted exponential function term Used to weight the prediction results of each primary classifier and enhance the nonlinear characteristics through the exponential function so that important prediction results get greater weight; normalize the denominator Ensure that the sum of the output prediction results is 1, ensure the probability distribution property, and make the prediction results more interpretable;

[0140] Among them, the primary classifier weight w j It is automatically learned during the training process and reflects the importance of each primary classifier; the scaling factor γ j and the nonlinear adjustment factor λ j Usually, when training the secondary learner, it is optimized and determined by methods such as cross-validation or grid search; the prediction result f j (x i ) is directly provided by the primary classifier, without the need for additional learning; the exponential function exp and the logarithmic function log are fixed mathematical functions;

[0141] In the intermediate representation, the hyperbolic tangent transform term tanh(β j exp(γ j ·f j (x i )+λ j ·log(1+|f j (x i )|))+δ j ): Introducing stronger nonlinear characteristics to help the model better capture the complex relationship between features, scaling factor β j Adjust the scale of the internal expression of the hyperbolic tangent function, the bias term δ jAdjust the offset of the final output; weighted summation term The prediction results of all primary classifiers are weighted and summed to ensure that the contribution of each classifier is reasonably considered. The weight of the primary classifier α j It is automatically learned during the training process, reflecting the importance of each primary classifier; the sine transform term Enhance feature representation diversity, help the model understand periodic changes, auxiliary function weight η k Controls the influence of auxiliary functions. Provide additional information sources, such as time series data or other external factors; Final output bias term φ: adjust the offset of the final output;

[0142] Among them, the primary classifier weight α j , scaling factor β j , the bias term δ j Automatically learned during training; sine term weight θ, auxiliary function weight η k Set according to experimental results to control the influence of different components; auxiliary function Provided by external data sources, such as time series data of the injured and sick; the final output bias term φ is automatically learned during the training process;

[0143] Assume that in a battlefield environment, a frontline medical station receives a group of wounded and sick people from different conflict zones and needs to quickly and accurately assess their injuries in order to prioritize and treat them; Assume that x i =[0.1, 0.2, 0.3, 0.4, 0.5]; Assume that the number of primary classifiers used for training is N = 5, and the weight w j =[0.1,0.2,0.3,0.2,0.2], scaling factor γ j =[0.4,0.5,0.6,0.7,0.8], nonlinear adjustment factor λ j =[0.2,0.3,0.4,0.5,0.6], prediction result f j (x i ) = [0.3, 0.4, 0.5, 0.6, 0.7];

[0144] Assume that the weight α after nonlinear transformation j =[0.2,0.2,0.3,0.1,0.2], scaling factor β j =[0.5,0.6,0.7,0.8,0.9], bias term δ j =[0.1, 0.2, 0.3, 0.4, 0.5], sine term weight θ = 0.6, auxiliary function weight ηk =[0.4,0.3,0.3], auxiliary function prediction results The final output bias term φ = 0.2;

[0145]

[0146] Assume that the threshold is θ = [0.85, 0.85, 0.85, 0.85, 0.85], since the final result H of the comprehensive feature representation i = Each element of [0.87, 0.89, 0.91, 0.93, 0.95] is greater than the corresponding hypothetical threshold, which indicates that the current patient information performs well in terms of understanding and feature capture, and is suitable for further use or as the final information representation. Through the above steps, medical data can be effectively processed and high-quality feature representations can be extracted, thereby improving the accuracy and timeliness of the assessment of the health status of the patient.

[0147] 104. Based on the life support measures plan, the information is synchronized to the remote medical command center in real time through a secure communication protocol to generate an injury assessment result for the injured or sick.

[0148] In this step, the secure communication protocol is a set of communication specifications that ensure confidentiality and integrity during data transmission, such as TLS / SSL, to ensure the security of data during transmission.

[0149] Real-time synchronization means that life support measures can be quickly sent to the remote medical command center through a reliable network connection (such as 4G / 5G or satellite communication) to ensure that the information reaches the destination in time.

[0150] The telemedicine command center is a platform for centralized management and dispatch of medical resources. It is responsible for receiving and processing injury assessment information of the wounded and sick sent back from the front line, and coordinating the medical team to make corresponding decisions and allocate resources.

[0151] The injury assessment results of the wounded and sick are the final assessment reports generated based on the life support measures plan, which provide the rear medical team with an understanding of the specific conditions of the wounded and sick, and make corresponding decisions and resource allocations accordingly.

[0152] In an embodiment of the present application, assuming that in a multinational joint rescue operation, first, the front-line medical team uses encrypted communication equipment to encode the life support measures plan into a standardized data packet; second, the data packet is encrypted through a secure communication protocol (such as TLS / SSL) to ensure the security of transmission; third, the encrypted data packet is sent in real time to a remote medical command center located in different countries using a satellite communication link; finally, the command center decrypts and verifies the integrity of the data after receiving it, and generates a final injury assessment result for the injured and sick, so that medical teams from various countries can work together, understand the condition of the injured and sick in a timely manner, and make corresponding decisions and resource allocation.

[0153] Optionally, the life support measures plan in step 104 is synchronized to the remote medical command center in real time through a secure communication protocol to generate an injury assessment result of the injured or sick, including: based on the life support measures plan, structured processing is performed, converted into a unified data format, necessary metadata is added, and a standardized data packet is generated; based on the standardized data packet, encryption processing is performed using a secure communication protocol to ensure confidentiality and integrity of the transmission process to generate an encrypted data packet; based on the encrypted data packet, sending it to the remote medical command center in real time through a reliable network connection to generate transmission confirmation information; based on the transmission confirmation information, it is received by the remote medical command center, and the encrypted data packet is decrypted to generate an injury assessment result of the injured or sick.

[0154] In this step, structuring means formatting and organizing the information in the life support measure plan to facilitate subsequent conversion and transmission. This process ensures the consistency and readability of the data.

[0155] A unified data format is a standardized way of representing data to ensure interoperability between different systems. By converting data into a unified format, data processing and parsing can be simplified.

[0156] Metadata refers to data that describes data, such as timestamps, source identifiers, etc. It provides additional information about the content of the data packet, which helps the receiver to correctly understand and use the data.

[0157] A standardized data package is a structured data collection that contains necessary metadata and uses a unified data format. This data package facilitates secure transmission and efficient processing.

[0158] Encryption is the process of applying encryption algorithms to data to ensure that only authorized users can decrypt and access the data.

[0159] An encrypted data packet is a standardized data packet that has been encrypted to ensure that its content cannot be read or tampered with by unauthorized third parties during transmission.

[0160] A reliable network connection refers to a communication link with high stability and low latency characteristics, such as 4G / 5G or satellite communication, which ensures that data can be transmitted to its destination quickly and accurately.

[0161] The transmission confirmation message is a confirmation message received by the sender, indicating that the data has successfully reached the receiver. This step ensures the reliability of data transmission.

[0162] Decryption refers to the process of decrypting the received encrypted data packet and restoring the original data. This process requires the use of the same key and algorithm as the sender.

[0163] In the embodiment of the present application, the life support measures plan is first structured and converted into a unified data format, and necessary metadata is added to generate a standardized data packet; secondly, the standardized data packet is encrypted using a secure communication protocol to ensure confidentiality and integrity during transmission, and an encrypted data packet is generated; thirdly, the encrypted data packet is sent to the telemedicine command center in real time through a reliable network connection, and transmission confirmation information is generated; finally, after receiving the transmission confirmation information, the telemedicine command center decrypts the received encrypted data packet and generates an injury assessment result of the wounded and sick, ensuring information synchronization between the front-line and rear medical teams.

[0164] Suppose that in an international disaster relief operation, first the on-site medical team completes the formulation of life support measures for the wounded and sick, and performs structured processing, converts it into a unified data format, adds necessary metadata, and generates standardized data packets; secondly, these data packets are encrypted through a secure communication protocol to ensure confidentiality and integrity during transmission, and generate encrypted data packets; thirdly, using satellite communication links, the encrypted data packets are sent in real time to telemedicine command centers located in different countries through reliable network connections, and transmission confirmation information is generated; finally, after receiving the transmission confirmation information, the telemedicine command center decrypts the received encrypted data packets and generates injury assessment results for the wounded and sick, ensuring that medical teams from various countries can work together, understand the conditions of the wounded and sick in a timely manner, and make corresponding decisions and resource allocation.

[0165] In summary, steps 101 to 104 cover the complete process from the collection and preprocessing of multimodal data of the wounded and sick, the extraction of high-dimensional feature vectors, the generation of prediction results through multiple primary classifiers, to the use of stacked generalization algorithms to train secondary learners to integrate the prediction results of different classifiers. The aim is to provide an efficient and accurate solution for the injury assessment of the wounded and sick to meet the needs of rapid and accurate medical decision-making in battlefield environments.

[0166] Example

[0167] Figure 2The present application provides a schematic diagram of a system for assessing the condition of a wounded or sick person in a battlefield environment. Figure 2 As shown, the device comprises:

[0168] The collection module 21 is used to collect the real-time physiological image data of the wounded and sick at the battlefield through the portable medical imaging equipment, perform preliminary data cleaning and formatting processing, and generate standardized physiological image data;

[0169] Enhancement module 22, used to enhance the key physiological region signals based on the standardized physiological image data by using the regional attention mechanism algorithm, suppress irrelevant information, adopt transfer learning technology, improve accuracy through the knowledge transfer capability of the pre-trained model, and generate a preliminary injury level report;

[0170] A fusion module 23 is used to use a stacked generalization algorithm based on the preliminary injury level report, combine the prediction results of multiple heterogeneous classifiers, integrate the outputs of each primary classifier through a secondary learner, and use multimodal fusion technology to integrate visual and non-visual information, refine the injury level of the injured and sick, and generate a life support measure plan;

[0171] The generation module 24 is used to generate injury assessment results of the injured and sick based on the life support measure plan and synchronize it to the remote medical command center in real time through a secure communication protocol.

[0172] Figure 2 The injury assessment system for the wounded and sick in a battlefield environment can be executed Figure 1 The implementation principle and technical effect of the method for assessing the condition of the wounded and sick in a battlefield environment described in the embodiment shown are not described in detail. The specific manner in which each module and unit performs operations in the above embodiment of the system for assessing the condition of the wounded and sick in a battlefield environment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0173] In one possible design, Figure 2 The illustrated embodiment of the present invention is a system for assessing the condition of a wounded or sick person in a battlefield environment, which can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0174] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0175] The processing component 32 is used to: collect instant physiological image data of the wounded and sick at the battlefield through portable medical imaging equipment, perform preliminary data cleaning and formatting processing, and generate standardized physiological image data; based on the standardized physiological image data, use the regional attention mechanism algorithm to enhance the signals of key physiological regions, suppress irrelevant information, use transfer learning technology to improve accuracy through the knowledge transfer ability of the pre-trained model, and generate a preliminary injury level report; based on the preliminary injury level report, use the stacked generalization algorithm, combine the prediction results of multiple heterogeneous classifiers, integrate the outputs of each primary classifier through a secondary learner, use multimodal fusion technology to integrate visual and non-visual information, refine the injury classification of the wounded and sick, and generate a life support measure plan; based on the life support measure plan, synchronize it to the remote medical command center in real time through a secure communication protocol to generate an injury assessment result of the wounded and sick.

[0176] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0177] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0178] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0179] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0180] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0181] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0182] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a method for assessing the condition of a wounded or sick soldier in a battlefield environment.

[0183] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0184] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0185] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for assessing the condition of a wounded or sick person in a battlefield environment, characterized in that: include: Through portable medical imaging equipment, real-time physiological image data of the wounded and sick are collected on the battlefield, and preliminary data cleaning and formatting are performed to generate standardized physiological image data; Based on the standardized physiological image data, the regional attention mechanism algorithm is used to enhance the signals of key physiological regions, suppress irrelevant information, and transfer learning technology is used to improve the accuracy through the knowledge transfer capability of the pre-trained model to generate a preliminary injury level report; Based on the preliminary injury grade report, a stacked generalization algorithm is used to combine the prediction results of multiple heterogeneous classifiers, the outputs of each primary classifier are integrated through a secondary learner, and multimodal fusion technology is used to integrate visual and non-visual information, refine the injury grade of the injured and sick, and generate a life support measure plan; Based on the life support measures plan, the injury assessment results of the injured and sick are generated in real time through synchronization to the remote medical command center through a secure communication protocol.

2. The method according to claim 1, characterized in that Based on the standardized physiological image data, the regional attention mechanism algorithm is used to enhance the key physiological area signals, suppress irrelevant information, adopt transfer learning technology, improve accuracy through the knowledge transfer ability of the pre-training model, and generate a preliminary injury level report, including: Based on the standardized physiological image data, preliminarily identifying and marking each area in the image to generate image data labels; Based on the image data labeling, a regional attention mechanism algorithm is used to enhance the signals of key physiological regions in the image, suppress background and other non-related regional information, and generate focused image data; Based on the focused image data, transfer learning technology is used to pre-train the model using a large-scale medical image data set, and the pre-trained model knowledge is transferred to the current task, so as to quickly adapt to the situation of the wounded and sick in a specific battlefield environment and generate knowledge transfer data; Based on the knowledge transfer data, combined with clinical standards and empirical rules, the specific injury level of the injured and sick is further refined to generate a preliminary injury level report.

3. The method according to claim 2, characterized in that Based on the image data labeling, the regional attention mechanism algorithm is used to enhance the key physiological area signals in the image, suppress the background and other non-related area information, and generate focused image data, including: Based on the image data labels, an attention map is constructed, weight values ​​are assigned to all pixel regions to reflect importance, and an initial attention distribution is generated; Based on the initial attention distribution, the regional attention mechanism algorithm is used to adjust the feature representation in the original image, amplify the key physiological area features by weighted summation, weaken the influence of background and other non-related area information, and generate enhanced feature representation; Based on the enhanced feature representation, convolutional neural network technology is used to further mine deep features, keep the spatial position of key physiological signals consistent, and generate deep feature maps; Based on the deep feature map, fusion reconstruction is performed in combination with the original image information to ensure that key physiological areas are highlighted and generate focused image data.

4. The method according to claim 2, characterized in that: Based on the focused image data, the transfer learning technology is used to pre-train the model through a large-scale medical image data set, and the pre-trained model knowledge is transferred to the current task, so as to quickly adapt to the situation of the wounded and sick in a specific battlefield environment and generate knowledge transfer data, including: Based on the focused image data, a large-scale medical image dataset is selected to pre-train the model to generate a basic pre-trained model; Based on the basic pre-trained model, the model is fine-tuned using transfer learning technology, the bottom layer weights are frozen, and the top layer is retrained using the focused image data to generate a preliminary adaptation model; Based on the preliminary adaptation model, the learning rate is adjusted and regularization is introduced to prevent overfitting, so as to ensure that the model accurately reflects the conditions of the wounded and sick in the battlefield environment and generate an optimized adaptation model; Based on the optimized adaptation model, the focused image data is further processed, inheriting the generalization ability of the basic pre-trained model to generate knowledge transfer data.

5. The method according to claim 1, characterized in that Based on the preliminary injury grade report, the stacked generalization algorithm is used to combine the prediction results of multiple heterogeneous classifiers, the outputs of each primary classifier are integrated through the secondary learner, and the multimodal fusion technology is used to integrate visual and non-visual information, refine the injury grade of the injured and sick, and generate a life support measure plan, including: Based on the preliminary injury level report, an integrated model is constructed to integrate multiple heterogeneous classifiers to independently predict the injury level of the injured and sick and to generate heterogeneous classifier prediction results; Based on the prediction results of the heterogeneous classifiers, a stacked generalization algorithm is used to integrate the outputs of the primary classifiers through the secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result; Based on the comprehensive prediction results, multimodal fusion technology is used to combine visual information with non-visual information to further analyze the specific conditions of the injured and sick and generate detailed injury classification; Based on the detailed injury classification, the severity and type of the injury are comprehensively considered, the preset life support strategy template is matched, and a life support measure plan is generated.

6. The method according to claim 5, characterized in that Based on the prediction results of the heterogeneous classifiers, the stacked generalization algorithm is used to integrate the outputs of the primary classifiers through the secondary learner to improve the overall prediction accuracy and stability and generate a comprehensive prediction result, including: Based on the prediction results of the heterogeneous classifier, the prediction results are input into the stacked generalization model as new features to generate stacked layer input data; Based on the stacked layer input data, a stacked generalization algorithm is used to train the secondary learner, learn to integrate the prediction results of each primary classifier to capture the complementary information between different classifiers, and generate a secondary learner model; Based on the secondary learner model, optimizing secondary learner parameters, minimizing overfitting risk, and generating an optimized secondary learner; Based on the optimized secondary learner, new patient data is input for prediction to improve overall prediction accuracy and stability and generate comprehensive prediction results.

7. The method according to claim 1, characterized in that The life support measure plan is synchronized to the remote medical command center in real time through a secure communication protocol to generate an injury assessment result of the injured and sick, including: Based on the life support measure plan, perform structured processing, convert into a unified data format, add necessary metadata, and generate a standardized data package; Based on the standardized data packet, a secure communication protocol is used for encryption processing to ensure confidentiality and integrity of the transmission process and generate an encrypted data packet; Based on the encrypted data packet, the encrypted data packet is sent to the remote medical command center in real time through a reliable network connection to generate transmission confirmation information; Based on the transmission confirmation information, the remote medical command center receives and decrypts the encrypted data packet to generate an injury assessment result of the injured or sick.

8. A system for assessing the condition of the wounded and sick in a battlefield environment, characterized in that: include: The collection module is used to collect the real-time physiological image data of the wounded and sick at the battlefield through portable medical imaging equipment, perform preliminary data cleaning and formatting, and generate standardized physiological image data; An enhancement module is used to enhance the key physiological region signals based on the standardized physiological image data by using a regional attention mechanism algorithm, suppress irrelevant information, adopt transfer learning technology, improve accuracy through the knowledge transfer capability of the pre-trained model, and generate a preliminary injury level report; A fusion module is used to use a stacked generalization algorithm based on the preliminary injury level report, combine the prediction results of multiple heterogeneous classifiers, integrate the outputs of each primary classifier through a secondary learner, and use multimodal fusion technology to integrate visual and non-visual information, refine the injury level of the injured and sick, and generate a life support measure plan; A generation module is used to generate injury assessment results of the injured and sick based on the life support measures plan and synchronized to the remote medical command center in real time through a secure communication protocol.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for assessing the injury of a sick and wounded person in a battlefield environment as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a method for assessing the injury of a sick or wounded person in a battlefield environment as claimed in any one of claims 1 to 7 is implemented.

Images