Emergency medical rescue AI training drill assessment method

Abstract

The invention relates to an emergency medical rescue AI training drill evaluation method, and belongs to the technical field of computer vision image processing. The method comprises the following steps: acquiring an AI drilling scene and a drilling scene reference actual action optical flow frame, acquiring a drilling scene actual expression video sequence through a camera queue according to the AI drilling scene, and obtaining a drilling scene actual expression video optical flow frame through a preprocessing model according to the drilling scene actual expression video sequence, and according to the practice scene actual combat performance video optical flow frame, obtaining a rescue worker classification actual combat action optical flow frame through an action identification classification model, and according to the rescue worker classification actual combat action optical flow frame, obtaining rescue worker actual combat abnormal performance through an actual combat action evaluation model. And obtaining rescue worker actual combat performance evaluation through evaluation according to the rescue worker actual combat abnormal performance. Accurate, scientific and comprehensive evaluation of actual combat performance of rescue workers is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision image processing, and in particular relates to an emergency medical rescue AI training and drill evaluation method. Background Art

[0002] Through AI virtual emergency rescue scenarios such as flood disaster first aid, large-scale poisoning, mining trauma, fire rescue, subway tunnel disaster, large-scale heat stroke treatment, terrorist attacks and explosions, rescuers conduct actual combat drills in virtual emergency rescue scenarios to improve their ability to respond to sudden disaster events. It is necessary to conduct a detailed evaluation of the actual combat performance of rescuers in actual combat drills, but the existing drill evaluation methods have the problem of single drill scenarios and difficulty in comprehensively and accurately evaluating the actual combat performance of rescuers. Summary of the invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides an emergency medical rescue AI training and exercise evaluation method.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] An emergency medical rescue AI training and drill evaluation method, comprising:

[0006] Obtaining optical flow frames of AI drill scenes and benchmark actual combat actions of the drill scenes, and collecting actual combat performance video sequences of the drill scenes through camera queues according to the AI ​​drill scenes;

[0007] According to the video sequence of the actual combat performance of the drill scene, an optical flow frame of the actual combat performance video of the drill scene is obtained through a preprocessing model;

[0008] According to the actual combat performance video optical flow frames of the drill scene, the rescuer classification actual combat action optical flow frames are obtained through the action recognition classification model;

[0009] According to the rescuer's classified actual combat action optical flow frame, the actual combat action evaluation model is used to obtain the rescuer's abnormal performance in actual combat;

[0010] An actual combat performance evaluation of the rescuer is obtained through evaluation based on the abnormal actual combat performance of the rescuer.

[0011] Preferably, obtaining the video optical flow frames of the actual combat performance of the drill scene through the preprocessing model according to the actual combat performance video sequence of the drill scene includes:

[0012] According to the actual combat performance video sequence of the drill scene, a normalized sequence of the actual combat performance video of the drill scene is obtained by normalization processing;

[0013] According to the normalized sequence of the actual combat performance video of the drill scene, the optical flow frames of the actual combat performance video of the drill scene are extracted through Flownet2.

[0014] Preferably, obtaining the rescuer classification actual combat action optical flow frame through the action recognition classification model according to the actual combat performance video optical flow frame of the drill scene includes:

[0015] According to the actual combat performance video optical flow frame of the drill scene, the actual combat performance hidden state is obtained through LSTM;

[0016] According to the actual combat performance video optical flow frame of the drill scene, a real combat performance feature map is obtained through a double convolution pooling layer, and the double convolution pooling layer includes a convolution layer, a pooling layer, and a convolution layer;

[0017] Obtaining a combat performance pooling feature map through a global average pooling layer according to the combat performance feature map;

[0018] According to the actual performance pooling feature map, a normalized actual performance pooling feature map is obtained by batch normalization of one layer;

[0019] Obtaining an actual combat performance integrated feature vector through a fully connected layer according to the actual combat performance pooled normalized feature map and the actual combat performance hidden state;

[0020] According to the actual combat performance, the feature vector is integrated through the Softmax classifier to obtain the rescuer's classified actual combat action optical flow frame.

[0021] Preferably, the expression of the Softmax classifier is:

[0022]

[0023] Among them, S j represents the probability of the class to which the optical flow frame sample of the actual performance video of the current drill scene belongs, a is the integrated feature vector of the actual performance, N is the number of classes, and a j For practical performance, the j-th value of the feature vector is integrated, k is the k-th class, and e is the exponential operation.

[0024] Preferably, the step of obtaining the abnormal performance of the rescuer in actual combat by using the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame comprises:

[0025] According to the rescuer classification actual combat action optical flow frame, a reconstruction model is enhanced by actual combat action feature to obtain a reconstructed rescuer classification actual combat action optical flow frame;

[0026] The abnormal performance of the rescuer in actual combat is obtained by calculating the pixel error based on the reconstructed rescuer classification actual combat action optical flow frame and the drill scene benchmark actual combat action optical flow frame.

[0027] Preferably, the step of reconstructing a model by enhancing the actual combat action features according to the rescuer classification actual combat action optical flow frame to obtain a reconstructed rescuer classification actual combat action optical flow frame comprises:

[0028] According to the rescuer classification actual combat action optical flow frame, an encoder is used to extract the rescuer classification actual combat action optical flow feature map;

[0029] According to the rescuer classification actual combat action optical flow feature map, the feature dimension is flattened by defining a latent space to obtain the rescuer classification actual combat action optical flow flattened feature, wherein the latent space carries spatial features;

[0030] The latent space is defined as:

[0031] M∈R D ×N

[0032] Among them, M represents the latent space, R represents the real number field, N represents the number of feature vectors contained in the latent space, and D represents the dimension of the feature item;

[0033] According to the flattened optical flow features of the rescuer classification actual combat actions, similarity of the rescuer classification actual combat action features is obtained by similarity calculation;

[0034] The similarity calculation expression is:

[0035]

[0036] Among them, d represents the similarity of the rescuer's classified actual combat action features, m i represents the spatial features of the i-th latent space, and T is the transposed calculation;

[0037] According to the similarity of the rescuer classification actual combat action feature, a similar feature item index of the rescuer classification actual combat action is obtained by querying similar feature items;

[0038] According to the similar feature item index of the rescuer classification actual combat action, normal sample feature items of the rescuer classification actual combat action are obtained by analysis;

[0039] According to the normal sample feature item of the rescuer's classified actual combat action, the normal sample embedding feature item of the rescuer's classified actual combat action is obtained by convolution action feature embedding;

[0040] The convolution action feature embedding expression is:

[0041]

[0042] in, represents the embedding feature item of the normal sample of the rescuer's classified actual combat action, represents the normal sample feature item of the rescuer's classified actual combat action, and f represents the convolution operation;

[0043] The reconstructed rescuer classification actual combat action optical flow frame is obtained by decoding the normal sample embedding feature items of the rescuer classification actual combat action through a decoder.

[0044] Preferably, the pixel error calculation expression is:

[0045]

[0046] Among them, MES is the pixel error value, n is the total number of pixels, i represents the i-th pixel, I pred (i) represents the value of the optical flow frame of the reconstructed rescuer classification actual combat action at pixel i, I true (i) represents the value of the optical flow frame of the benchmark actual combat action in the training scene at pixel point i.

[0047] An emergency medical rescue AI training and drill evaluation method system includes an information collection module, an information preprocessing module, an action classification module, an anomaly detection module, and an action evaluation module, including:

[0048] The information collection module is used to obtain the AI ​​drill scene and the actual combat action optical flow frames of the drill scene benchmark, and collect the actual combat performance video sequence of the drill scene through the camera queue according to the AI ​​drill scene;

[0049] The information preprocessing module is used to obtain the optical flow frame of the actual combat performance video of the drill scene through the preprocessing model according to the actual combat performance video sequence of the drill scene;

[0050] The action classification module is used to obtain the rescuer classification actual combat action optical flow frame through the action recognition classification model according to the actual combat performance video optical flow frame of the drill scene;

[0051] The anomaly detection module is used to obtain the abnormal performance of the rescuer in actual combat through the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame;

[0052] The action evaluation module is used to obtain an evaluation of the rescuer's actual combat performance through evaluation based on the rescuer's abnormal actual combat performance.

[0053] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned emergency medical rescue AI training and drill evaluation method when executing the program.

[0054] A storage medium containing computer executable instructions, which, when executed by a computer processor, are used to perform the above-mentioned emergency medical rescue AI training drill evaluation method.

[0055] The beneficial effects of the present invention are:

[0056] (1) Through AI technology virtual drill scenarios, the actual combat actions of rescuers in different drill scenarios can be effectively collected and analyzed, which improves the comprehensiveness and comprehensiveness of the evaluation of rescuers' actual combat performance;

[0057] (2) Obtaining the optical flow frames of the actual combat performance video of the drill scene through Flownet2 can accurately describe the movement speed and direction of each pixel point in the actual combat performance video sequence of the drill scene between two consecutive frames of images, thereby improving the accuracy of the evaluation of the actual combat performance of the rescue personnel;

[0058] (3) The temporal features of the optical flow frames of the actual combat performance video of the drill scene are extracted through LSTM, and the spatial features of the optical flow frames of the actual combat performance video of the drill scene are extracted through the CNN convolutional layer, thereby improving the accuracy of classification and thus improving the accuracy of the evaluation of the actual combat performance of the rescuers;

[0059] (4) The optical flow feature map of the rescuer's classified combat actions is strengthened and reconstructed through the combat action evaluation model, and the position and movement mode of the rescuer are considered in the time and space dimensions to improve the analysis of the dynamic changes of the rescuer in actual combat, thereby strengthening the detection of abnormal performance of the rescuer in actual combat. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0061] Figure 1 This invention provides an emergency medical rescue AI training and drill evaluation method. DETAILED DESCRIPTION

[0062] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0063] See also Figure 1 , an emergency medical rescue AI training drill evaluation method, including:

[0064] S1: Obtaining optical flow frames of AI drill scenes and benchmark actual combat actions of the drill scenes, and collecting actual combat performance video sequences of the drill scenes through camera queues according to the AI ​​drill scenes;

[0065] S2: obtaining an optical flow frame of the actual combat performance video of the drill scene through a preprocessing model according to the actual combat performance video sequence of the drill scene;

[0066] S3: Obtaining classified actual combat action optical flow frames of rescuers through an action recognition classification model according to the actual combat performance video optical flow frames of the drill scene;

[0067] S4: Obtaining abnormal performance of the rescuer in actual combat through an actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame;

[0068] S5: Obtaining an actual combat performance evaluation of the rescuer through evaluation according to the abnormal actual combat performance of the rescuer.

[0069] In this embodiment, in step S1, the AI ​​drill scenario includes actual combat of first aid for flood disasters, actual combat of fire rescue, and actual combat of trauma in mine disasters, and the actual combat performance video sequence of the drill scenario includes actual combat performance video sequence of first aid for flood disasters, actual combat performance video sequence of fire rescue, and actual combat performance video sequence of trauma in mine disasters.

[0070] In step S2, obtaining the video optical flow frame of the actual combat performance of the drill scene through the preprocessing model according to the actual combat performance video sequence of the drill scene includes:

[0071] S201: obtaining a normalized sequence of the actual combat performance video of the drill scene by normalizing the actual combat performance video sequence of the drill scene;

[0072] S202: Extracting optical flow frames of the actual combat performance video of the drill scene through Flownet2 according to the normalized sequence of the actual combat performance video of the drill scene.

[0073] In this embodiment, the optical flow frame of the actual combat performance video of the drill scene describes the movement speed and direction of each pixel point in the actual combat performance video sequence of the drill scene between two consecutive frames of images.

[0074] In step S3, obtaining the rescuer classified actual combat action optical flow frame through the action recognition classification model according to the actual combat performance video optical flow frame of the drill scene includes:

[0075] S301: Obtaining a hidden state of the actual combat performance through LSTM according to the optical flow frame of the actual combat performance video of the drill scene;

[0076] S302: Obtaining a combat performance feature map through a double convolution pooling layer according to the optical flow frame of the combat performance video of the drill scene, wherein the double convolution pooling layer includes a convolution layer, a pooling layer, and a convolution layer;

[0077] S303: Obtaining a combat performance pooling feature map through a global average pooling layer according to the combat performance feature map;

[0078] S304: obtaining a pooled normalized feature map of actual combat performance by batch normalization of one layer according to the pooled feature map of actual combat performance;

[0079] S305: Obtaining an actual combat performance integrated feature vector through a fully connected layer according to the actual combat performance pooling normalized feature map and the actual combat performance hidden state;

[0080] S306: Integrate the feature vectors according to the actual combat performance and obtain the rescuer's classified actual combat action optical flow frame through a Softmax classifier.

[0081] The expression of the Softmax classifier is:

[0082]

[0083] Among them, S j represents the probability of the class to which the optical flow frame sample of the actual performance video of the current drill scene belongs, a is the integrated feature vector of the actual performance, N is the number of classes, and a j For practical performance, the j-th value of the feature vector is integrated, k is the k-th class, and e is the exponential operation.

[0084] In this embodiment, the rescuer's classified actual combat action optical flow frame obtained according to step 3 is the classified rescuer's actual combat action optical flow frame.

[0085] In step S4, obtaining the abnormal performance of the rescuer in actual combat by using the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame includes:

[0086] S401: Reconstructing a model by enhancing the actual combat action feature according to the rescuer classification actual combat action optical flow frame to obtain a reconstructed rescuer classification actual combat action optical flow frame;

[0087] S402: Obtaining the abnormal performance of the rescuer in actual combat by calculating pixel errors according to the reconstructed rescuer classification actual combat action optical flow frame and the drill scene benchmark actual combat action optical flow frame;

[0088] In this embodiment, the actual combat action feature enhanced reconstruction model includes:

[0089] S401-1: extracting the rescuer classification actual combat action optical flow feature map through an encoder according to the rescuer classification actual combat action optical flow frame;

[0090] S401-2: flattening the feature dimension according to the rescuer classification actual combat action optical flow feature map by defining a latent space to obtain the rescuer classification actual combat action optical flow flattened feature, wherein the latent space carries spatial features;

[0091] The latent space is defined as:

[0092] M∈R D ×N

[0093] Among them, M represents the latent space, R represents the real number field, N represents the number of feature vectors contained in the latent space, and D represents the dimension of the feature item;

[0094] The flattened representation of the feature dimension is:

[0095]

[0096] in, is the optical flow feature of the rescuer classification actual combat action, H represents the height of the optical flow feature map of the rescuer classification actual combat action, and W represents the width of the optical flow feature map of the rescuer classification actual combat action;

[0097] S401-3: obtaining the similarity of the rescuer classification actual combat action feature by similarity calculation according to the optical flow flattening feature of the rescuer classification actual combat action;

[0098] The similarity calculation expression is:

[0099]

[0100] Among them, d represents the similarity of the rescuer's classified actual combat action features, m i represents the spatial features of the i-th latent space, and T is the transposed calculation;

[0101] S401-4: obtaining a similar feature item index of the rescuer classification actual combat action according to the similarity of the rescuer classification actual combat action feature through similar feature item query;

[0102] S401-5: obtaining normal sample feature items of the rescuer's classified actual combat action by analyzing the similar feature item index of the rescuer's classified actual combat action;

[0103] S401-6: Obtaining a normal sample embedding feature item of the rescuer's classified actual combat action according to the normal sample feature item of the rescuer's classified actual combat action by convolution action feature embedding;

[0104] The convolution action feature embedding expression is:

[0105]

[0106] in, represents the embedding feature item of the normal sample of the rescuer's classified actual combat action, represents the normal sample feature item of the rescuer's classified actual combat action, and f represents the convolution operation;

[0107] S401-7: embedding feature items of normal samples of the rescuer classification actual combat action through decoding by a decoder to obtain the reconstructed rescuer classification actual combat action optical flow frame.

[0108] In this embodiment, the pixel error calculation expression is:

[0109]

[0110] Among them, MES is the pixel error value, n is the total number of pixels, i represents the i-th pixel, I pred (i) represents the value of the optical flow frame of the reconstructed rescuer classification actual combat action at pixel i, I true (i) represents the value of the optical flow frame of the benchmark actual combat action in the training scene at pixel point i.

[0111] It should be noted that, in step S401, the autoencoder to be used is a dual-stream autoencoder improved based on U-Net, and the dual-stream autoencoder improved based on U-Net includes an encoder and a decoder; in step S401-6, the convolution operation is a 1×1 convolution operation.

[0112] An emergency medical rescue AI training and drill evaluation method system includes an information collection module, an information preprocessing module, an action classification module, an anomaly detection module, and an action evaluation module, including:

[0113] The information collection module is used to obtain the AI ​​drill scene and the actual combat action optical flow frames of the drill scene benchmark, and collect the actual combat performance video sequence of the drill scene through the camera queue according to the AI ​​drill scene;

[0114] The information preprocessing module is used to obtain the optical flow frame of the actual combat performance video of the drill scene through the preprocessing model according to the actual combat performance video sequence of the drill scene;

[0115] The action classification module is used to obtain the rescuer classification actual combat action optical flow frame through the action recognition classification model according to the actual combat performance video optical flow frame of the drill scene;

[0116] The anomaly detection module is used to obtain the abnormal performance of the rescuer in actual combat through the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame;

[0117] The action evaluation module is used to obtain an evaluation of the rescuer's actual combat performance through evaluation based on the rescuer's abnormal actual combat performance.

[0118] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0119] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0120] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0121] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An emergency medical rescue AI training and exercise evaluation method, characterized in that: The following steps are involved: Obtaining optical flow frames of AI drill scenes and benchmark actual combat actions of the drill scenes, and collecting actual combat performance video sequences of the drill scenes through camera queues according to the AI ​​drill scenes; According to the video sequence of the actual combat performance of the drill scene, an optical flow frame of the actual combat performance video of the drill scene is obtained through a preprocessing model; According to the actual combat performance video optical flow frames of the drill scene, the rescuer classification actual combat action optical flow frames are obtained through the action recognition classification model; According to the rescuer's classified actual combat action optical flow frame, the actual combat action evaluation model is used to obtain the rescuer's abnormal performance in actual combat; An actual combat performance evaluation of the rescuer is obtained through evaluation based on the abnormal actual combat performance of the rescuer.

2. The emergency medical rescue AI training and exercise evaluation method according to claim 1 is characterized in that: The optical flow frames of the actual combat performance video of the drill scene obtained by the preprocessing model according to the actual combat performance video sequence of the drill scene include: According to the actual combat performance video sequence of the drill scene, a normalized sequence of the actual combat performance video of the drill scene is obtained by normalization processing; According to the normalized sequence of the actual combat performance video of the drill scene, the optical flow frames of the actual combat performance video of the drill scene are extracted through Flownet2.

3. The emergency medical rescue AI training and exercise evaluation method according to claim 1 is characterized in that: The obtaining of the rescuer classification actual combat action optical flow frames according to the actual combat performance video optical flow frames of the drill scene by using the action recognition classification model includes: According to the actual combat performance video optical flow frame of the drill scene, the actual combat performance hidden state is obtained through LSTM; According to the actual combat performance video optical flow frame of the drill scene, a real combat performance feature map is obtained through a double convolution pooling layer, and the double convolution pooling layer includes a convolution layer, a pooling layer, and a convolution layer; Obtaining a combat performance pooling feature map through a global average pooling layer according to the combat performance feature map; According to the actual performance pooling feature map, a normalized actual performance pooling feature map is obtained by batch normalization of one layer; Obtaining an actual combat performance integrated feature vector through a fully connected layer according to the actual combat performance pooled normalized feature map and the actual combat performance hidden state; According to the actual combat performance, the feature vector is integrated through the Softmax classifier to obtain the rescuer's classified actual combat action optical flow frame.

4. The emergency medical rescue AI training and exercise evaluation method according to claim 3 is characterized in that: The expression of the Softmax classifier is: Among them, S j represents the probability of the class to which the optical flow frame sample of the actual performance video of the current drill scene belongs, a is the integrated feature vector of the actual performance, N is the number of classes, and a j For practical performance, the j-th value of the feature vector is integrated, k is the k-th class, and e is the exponential operation.

5. The emergency medical rescue AI training and exercise evaluation method according to claim 1, characterized in that: The abnormal performance of the rescuer in actual combat is obtained by using the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame, including: According to the rescuer classification actual combat action optical flow frame, a reconstruction model is enhanced by actual combat action feature to obtain a reconstructed rescuer classification actual combat action optical flow frame; The abnormal performance of the rescuer in actual combat is obtained by calculating the pixel error based on the reconstructed rescuer classification actual combat action optical flow frame and the drill scene benchmark actual combat action optical flow frame.

6. The emergency medical rescue AI training and exercise evaluation method according to claim 5 is characterized in that: The step of reconstructing the rescuer classification actual combat action optical flow frame by enhancing the reconstruction model with actual combat action features according to the rescuer classification actual combat action optical flow frame comprises: According to the rescuer classification actual combat action optical flow frame, an encoder is used to extract the rescuer classification actual combat action optical flow feature map; According to the rescuer classification actual combat action optical flow feature map, the feature dimension is flattened by defining a latent space to obtain the rescuer classification actual combat action optical flow flattened feature, wherein the latent space carries spatial features; The latent space is defined as: M∈R D×N , Among them, M represents the latent space, R represents the real number field, N represents the number of feature vectors contained in the latent space, and D represents the dimension of the feature item; According to the flattened optical flow features of the rescuer classification actual combat actions, similarity of the rescuer classification actual combat action features is obtained by similarity calculation; The similarity calculation expression is: Among them, d represents the similarity of the rescuer's classified actual combat action features, m i represents the spatial features of the i-th latent space, and T is the transposed calculation; According to the similarity of the rescuer classification actual combat action feature, a similar feature item index of the rescuer classification actual combat action is obtained by querying similar feature items; According to the similar feature item index of the rescuer classification actual combat action, normal sample feature items of the rescuer classification actual combat action are obtained by analysis; According to the normal sample feature item of the rescuer's classified actual combat action, the normal sample embedding feature item of the rescuer's classified actual combat action is obtained by convolution action feature embedding; The convolution action feature embedding expression is: in, represents the embedding feature item of the normal sample of the rescuer's classified actual combat action, represents the normal sample feature item of the rescuer's classified actual combat action, and f represents the convolution operation; The reconstructed rescuer classification actual combat action optical flow frame is obtained by decoding the normal sample embedding feature items of the rescuer classification actual combat action through a decoder.

7. The emergency medical rescue AI training and exercise evaluation method according to claim 5 is characterized in that: The pixel error calculation expression is: Among them, MES is the pixel error value, n is the total number of pixels, i represents the i-th pixel, I pred (i) represents the value of the optical flow frame of the reconstructed rescuer classification actual combat action at pixel i, I true (i) represents the value of the optical flow frame of the benchmark actual combat action in the training scene at pixel point i.

8. An emergency medical rescue AI training and drill evaluation method system using the emergency medical rescue AI training and drill evaluation method of claim 1, comprising an information collection module, an information preprocessing module, an action classification module, an anomaly detection module, and an action evaluation module, characterized in that: include: The information collection module is used to obtain the AI ​​drill scene and the actual combat action optical flow frames of the drill scene benchmark, and collect the actual combat performance video sequence of the drill scene through the camera queue according to the AI ​​drill scene; The information preprocessing module is used to obtain the optical flow frame of the actual combat performance video of the drill scene through the preprocessing model according to the actual combat performance video sequence of the drill scene; The action classification module is used to obtain the rescuer classification actual combat action optical flow frame through the action recognition classification model according to the actual combat performance video optical flow frame of the drill scene; The anomaly detection module is used to obtain the abnormal performance of the rescuer in actual combat through the actual combat action evaluation model according to the rescuer's classified actual combat action optical flow frame; The action evaluation module is used to obtain an evaluation of the rescuer's actual combat performance through evaluation based on the rescuer's abnormal actual combat performance.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the emergency medical rescue AI training and drill evaluation method as described in any one of claims 1-8 is implemented.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the emergency medical rescue AI training drill evaluation method as described in any one of claims 1-8.

Images