Intrinsic security network resilience assessment and decision method, system, device and storage medium

Abstract

The present invention relates to the field of network security technology, and provides an endogenous security network resilience assessment and decision method, system, device and storage medium. The method comprises: step 1: obtaining the environmental perception results of multiple pre-deployed heterogeneous perception decision test components for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment; step 2: pre-processing the multiple environmental perception results; step 3: performing feature extraction on the pre-processed multiple environmental perception results based on the self-attention mechanism, and obtaining the perception features corresponding to each heterogeneous perception decision test component; step 4: performing cross-attention calculation on the multiple perception features based on the cross-attention mechanism, and obtaining the attention scores between different perception features; step 5: calculating the matching degree of different environmental perception results according to the attention scores between different perception features, and obtaining the decision result according to the matching degree.

Description

Technical Field

[0001] The present invention relates to the field of network security technology, and in particular to an intrinsic security network resilience assessment and decision method, system, device and storage medium. Background Art

[0002] With the development of the automotive intelligence industry, intelligent driving systems have become a key component of vehicle systems. The safety issues of intelligent driving have also received more and more attention. However, the existing single product cannot cope with the "unknown unknown" security threats. Therefore, using a variety of heterogeneous test component products to conduct network resilience evaluation has become an important means. Among them, the adjudication module is responsible for unified processing and judgment of the output results of different products in the evaluation, and is the core module in the evaluation.

[0003] At present, there have been studies on endogenous safety adjudication methods for intelligent driving vehicles. Existing methods focus on the design of the system framework level, covering multiple heterogeneous perception decision units and adjudication modules. By dividing the perception area, the correlation between the perception information in different perception decision units is calculated, and then it is determined whether there is an output anomaly. The adjudication module in the existing method is designed in a rule-based manner, which has the following problems: 1) The rule-based method is usually strongly related to the scenario. When the scenario changes, the rules need to be redesigned, which lacks generalization; 2) It involves multiple hyperparameters. Due to the different capabilities of different perception decision units, it is difficult to objectively give parameter values. Summary of the invention

[0004] In order to solve the shortcomings of the existing decision modules and give full play to the advantages of data-driven, the present invention proposes an intrinsic security network resilience assessment and decision method, system, device and storage medium.

[0005] In a first aspect, the present invention provides a method for evaluating and determining the resilience of an endogenous security network, comprising:

[0006] Step 1: Obtaining environmental perception results of a plurality of pre-deployed heterogeneous perception decision test components for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment;

[0007] Step 2: preprocessing the plurality of environmental perception results;

[0008] Step 3: extracting features from the plurality of preprocessed environmental perception results based on a self-attention mechanism to obtain perception features corresponding to each of the heterogeneous perception decision test components;

[0009] Step 4: performing cross-attention calculation on the plurality of perceptual features based on the cross-attention mechanism to obtain attention scores between different perceptual features;

[0010] Step 5: Calculate the matching degree of different environmental perception results based on the attention scores between different perception features, and obtain the judgment result based on the matching degree.

[0011] Furthermore, the preprocessing includes data parsing and interface unification, timestamp unification and coordinate normalization; wherein, data parsing and interface unification refers to independently parsing the environmental perception results of each heterogeneous perception decision test component, and then closing the parsed results into a unified format; timestamp unification refers to aligning the timestamps of the environmental perception results of multiple heterogeneous perception decision test components and unifying them to the same timestamp; coordinate normalization refers to unifying the environmental perception results of multiple heterogeneous perception decision test components to the same coordinate system.

[0012] Furthermore, step 3 specifically includes:

[0013] The feature extraction process is performed on the obstacle information perceived at the current moment in each environmental perception result, specifically including: extracting the initial obstacle features of each frame of data, the initial obstacle features including the obstacle position, size, direction, speed and type; sorting all the initial obstacle features according to the distance between the obstacle and the vehicle to obtain an obstacle information matrix; based on the obstacle information matrix, using the self-attention mechanism to calculate the attention scores between the obstacles, and based on the attention scores, weighting the initial obstacle features to obtain the final obstacle features;

[0014] The feature extraction process is performed on the lane line information perceived at the current moment in each environmental perception result, specifically including: dividing the lane line into multiple segments at fixed intervals, and each segment is represented by the position of the starting point and the end point; sorting all the lane line segments according to the distance between themselves and the vehicle to obtain a lane line information matrix; based on the lane line information matrix, using the self-attention mechanism to calculate the attention scores between different lane line segments, and based on the attention scores, all the lane line segments are weighted to obtain lane line features.

[0015] Furthermore, step 4 specifically includes:

[0016] For any two heterogeneous perception decision test components, the key matrix and the value matrix are calculated according to the obstacle features or lane features corresponding to one of the heterogeneous perception decision test components, and the query matrix is ​​calculated according to the obstacle features or lane features corresponding to the other heterogeneous perception decision test component; the calculated key matrix, value matrix and query matrix are substituted into the cross-attention calculation formula to obtain the attention scores of the obstacle features or lane features; the attention scores of the obstacle features and the lane features are concatenated to obtain the attention score between the perception features of any two heterogeneous perception decision test components.

[0017] Furthermore, in step 5, a cosine similarity calculation function is used to calculate the matching degree of different environment perception results.

[0018] In a second aspect, the present invention provides an endogenous security network resilience assessment and decision system, comprising:

[0019] A perception information acquisition module, used to acquire the environmental perception results of a plurality of pre-deployed heterogeneous perception decision test components for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment;

[0020] A preprocessing module, used for preprocessing the plurality of environmental perception results;

[0021] A feature extraction module, used to perform feature extraction on the plurality of preprocessed environmental perception results based on a self-attention mechanism to obtain perception features corresponding to each of the heterogeneous perception decision test components; and to perform cross-attention calculation on the plurality of perception features based on a cross-attention mechanism to obtain attention scores between different perception features;

[0022] The decision module is used to calculate the matching degree of different environmental perception results according to the attention scores between different perception features, and obtain the decision result according to the matching degree.

[0023] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0024] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0025] Beneficial effects of the present invention:

[0026] (1) The attention mechanism-based method of the present invention can learn the baseline capabilities of different perception and decision-making components in different scenarios by collecting a large amount of training data, without the need to manually set matching rules, thereby solving the problem of difficulty in generalization.

[0027] (2) The present invention utilizes a method based on the attention mechanism to automatically learn the correlation between the output results of different perception decision components without manually specifying information such as the area to be matched, thereby solving the problem of manually setting hyper-parameters in existing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1One of the flow charts of an intrinsic security network resilience assessment and decision method provided in an embodiment of the present invention;

[0029] Figure 2 A second flow chart of an intrinsic security network resilience assessment and decision method provided by an embodiment of the present invention;

[0030] Figure 3 A schematic diagram of the structure of an intrinsic security network resilience assessment and decision system provided by an embodiment of the present invention;

[0031] Figure 4 A structural block diagram of an electronic device provided by an embodiment of the present invention.

[0032] In the figure: processor 401 , communication interface 402 , memory 403 , and communication bus 404 . DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The embodiment of the present invention proposes an endogenous security network resilience assessment and decision method. For multiple heterogeneous perception decision test components (hereinafter referred to as heterogeneous test components or products), the method of the present invention is used to process the output results of the heterogeneous test components to obtain a final decision result, that is, whether there is an anomaly in the output result and the specific product with the anomaly, such as Figure 1 As shown, the specific process includes:

[0035] S101: Obtaining environmental perception results of a plurality of pre-deployed heterogeneous perception decision test components for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment;

[0036] S102: Preprocessing the plurality of environment perception results;

[0037] Specifically, the preprocessing includes three processing steps: 1) Data parsing and interface unification. Since the CAN protocols of different products are different, it is necessary to perform independent protocol parsing for each product and close the parsing results into a unified format. 2) Timestamp alignment. Due to differences in the software and hardware of different products, there will be certain differences in the time it takes to send each frame of data out after processing. In order to solve the problem of mismatched results caused by inconsistent timestamps, the timestamps need to be aligned. Using the kinematics and historical frame data of the obstacles, the output results of different products are timestamped and the data is unified to the same timestamp. 3) Coordinate normalization. Since the coordinate system definitions of different products may not be unified, in order to solve the problem of differences in the same perception results caused by this problem, the data needs to be normalized to the same coordinate system.

[0038] S103: performing feature extraction on the plurality of preprocessed environmental perception results based on a self-attention mechanism to obtain perception features corresponding to each of the heterogeneous perception decision test components;

[0039] S104: performing cross-attention calculation on the plurality of perception features based on a cross-attention mechanism to obtain attention scores between different perception features;

[0040] S105: Calculate the matching degree of different environmental perception results according to the attention scores between different perception features, and obtain a decision result according to the matching degree.

[0041] It can be understood that it is necessary to use training data to train the self-attention mechanism and cross-attention mechanism in the above steps, and learn the matching of the perception results of different products in normal or abnormal scenarios based on the positive samples (perception data that is not adversely affected) and negative samples (data with some abnormal perception results caused by adverse effects) in the training data, and set the normal matching range of the positive samples. In this way, in the test phase, the actual matching degree can be compared with the set normal matching degree range of the positive sample to obtain the judgment result, for example, including: judging the product with the lowest matching degree with the other two products and a matching degree less than the normal matching degree range of the positive sample as an abnormal product, and outputting the judgment result and the corresponding product serial number.

[0042] The endogenous security network resilience assessment and decision method provided in the embodiment of the present invention does not require manual setting of matching rules. It can use a large amount of sample data and an attention mechanism to learn the matching relationship between different product perception results, thereby improving the reliability of the matching results, and then automatically obtain the decision results based on the matching results. It has strong automatic matching learning capabilities and high generalization.

[0043] Based on the above embodiments, Figure 2As shown in the figure, taking three different advanced driving assistance system (ADAS) products (which can be expanded to more types in practice) as examples, the corresponding intrinsic safety network resilience assessment and decision method includes the following steps:

[0044] S201: Obtain the current environmental perception results sent by three different ADAS products (i.e., three heterogeneous assisted driving system executives) via the CAN protocol. It should be noted that the environmental information processed by each ADAS product at the same time is consistent.

[0045] S202: Preprocessing the environmental perception results of the above three ADAS products.

[0046] Specifically, the preprocessing includes three processing processes: 1) Data parsing and interface unification. Since the CAN protocols of different products are different, it is necessary to perform independent protocol parsing for each product and close the parsing results to a unified format. The parsed data includes the timestamp of each frame of data, obstacle information, and lane line information. Specifically, the obstacle information in this embodiment includes the position, size, direction, speed, and type of the obstacle, and the lane line information includes the curve type of the lane line and the curve representation parameters of the lane line. 2) Timestamp alignment. Specifically, in this embodiment, the data is unified to the timestamp of the nearest ADAS product. Assuming that the timestamps of the currently received ADAS1, ADAS2, and ADAS3 data are t1, t2, and t3, respectively, where t1 is greater than t2 and t3, then according to the above scheme, the ADAS2 and ADAS3 data are time extrapolated and unified to the t1 timestamp. 3) Coordinate normalization. Specifically, this embodiment utilizes the coordinate system definition method in the ADAS product protocol, and through translation and rotation transformation, unifies the coordinates of the perception data into a coordinate system with the center of the front of the car as the coordinate origin, the right side parallel to the front of the car as the positive direction of the x-axis, and the direction along the front of the car as the positive direction of the y-axis.

[0047] S203: performing feature extraction on the plurality of preprocessed environmental perception results based on a self-attention mechanism to obtain perception features corresponding to each of the heterogeneous perception decision test components;

[0048] Specifically, feature extraction is performed on obstacle data and lane line data obtained through preprocessing. Obstacle feature extraction: Extract the initial obstacle features of each frame of data, including obstacle position, size, orientation, speed and type; Sort all initial obstacle features according to the distance between the obstacle and the vehicle to obtain an obstacle information matrix; Based on the obstacle information matrix, the self-attention mechanism is used to calculate the attention scores between obstacles, and the initial obstacle features are weighted based on the attention scores to obtain the final obstacle features.

[0049] Lane line feature extraction: The lane line is divided into multiple segments at fixed intervals, and each segment is represented by the position of the start and end points; all lane line segments are sorted according to the distance between themselves and the vehicle to obtain a lane line information matrix; based on the lane line information matrix, the self-attention mechanism is used to calculate the attention scores between different lane line segments, and all lane line segments are weighted based on the attention scores to obtain lane line features.

[0050] Each ADAS product is processed using the above-mentioned obstacle feature extraction and lane line feature extraction processes to obtain the obstacle features and lane line features corresponding to the three products.

[0051] S204: Perform cross-attention calculation on the multiple perception features based on the cross-attention mechanism to obtain attention scores between different perception features.

[0052] Specifically, in order to subsequently calculate the matching degree between the results of different ADAS products, a cross-attention calculation is performed on the perceptual features of different ADAS products to obtain the attention scores between different perceptual features. For any two heterogeneous perception decision test components, such as product 1 and product 2, the key matrix K and value matrix V required for the cross-attention model are calculated based on the obstacle characteristics of product 1; the query matrix Q required for the cross-attention model is calculated based on the obstacle characteristics of product 2. Then use the cross-attention calculation formula to obtain its attention score. For lane line features, a similar process as above is performed. The processed obstacle and lane line data are spliced ​​to obtain the final feature results. It can be understood that in the cross-attention calculation, the perceptual features of multiple ADAS products need to be divided into two-by-two combinations for calculation.

[0053] The cross attention calculation formula is as follows: is the number of columns of the key matrix K.

[0054]

[0055] S205: Calculate the matching degree of different environmental perception results according to the attention scores between different perception features, and obtain a decision result according to the matching degree.

[0056] Specifically, this embodiment uses the cosine similarity calculation function to calculate the matching degree of perception results of different environments. Based on the positive samples (perception data that is not adversely affected) and negative samples (data with some abnormal perception results caused by adverse effects) in the training data, the above model can be used to learn the matching of perception results of different products. The product with the lowest matching degree with the other two products and a matching degree less than the normal matching degree range of the positive sample is determined as an abnormal product, and the judgment result and the corresponding product serial number are output.

[0057] It can be understood that the method of the present invention is also applicable to the testing of more than three ADAS products. For more than three products, it is necessary to perform cross-attention calculations on multiple products in pairs in the cross-attention calculation of step S204, and calculate the pairwise similarities in step S205. The other processes are consistent with the above steps.

[0058] like Figure 3 As shown, an embodiment of the present invention provides an endogenous security network resilience assessment and decision system, including a perception information acquisition module, a preprocessing module, a feature extraction module and a decision module.

[0059] Specifically, the perception information acquisition module is used to obtain the environmental perception results of multiple heterogeneous perception decision test components pre-deployed for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment; the preprocessing module is used to preprocess the multiple environmental perception results; the feature extraction module is used to perform feature extraction on the multiple environmental perception results after preprocessing based on the self-attention mechanism, and obtain the perception features corresponding to each of the heterogeneous perception decision test components; and is used to perform cross-attention calculation on the multiple perception features based on the cross-attention mechanism, and obtain the attention scores between different perception features; the judgment module is used to calculate the matching degree of different environmental perception results according to the attention scores between different perception features, and obtain the judgment result according to the matching degree.

[0060] It should be noted that the network resilience assessment and decision system provided in the embodiment of the present invention is for implementing the above method. The specific functions thereof can be referred to the above method embodiments, which will not be described in detail here.

[0061] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4As shown, the electronic device may include: a processor (processor) 401, a communication interface (Communications Interface) 402, a memory (memory) 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 complete mutual communication through the communication bus 404. The processor 401 can call the logic instructions in the memory 403 to execute the network resilience evaluation and decision method, which includes: step 1: obtaining the environmental perception results of multiple heterogeneous perception decision test components pre-deployed for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment; step 2: preprocessing the multiple environmental perception results; step 3: extracting features of the preprocessed multiple environmental perception results based on the self-attention mechanism, and obtaining the perception features corresponding to each heterogeneous perception decision test component; step 4: performing cross-attention calculation on the multiple perception features based on the cross-attention mechanism to obtain the attention scores between different perception features; step 5: calculating the matching degree of different environmental perception results according to the attention scores between different perception features, and obtaining the decision result according to the matching degree.

[0062] In addition, when the logic instructions in the above-mentioned memory 403 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0063] An embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: Step 1: Obtain environmental perception results of multiple heterogeneous perception decision test components pre-deployed for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment; Step 2: Preprocess the multiple environmental perception results; Step 3: Based on the self-attention mechanism, feature extraction is performed on the multiple environmental perception results after preprocessing to obtain the perception features corresponding to each of the heterogeneous perception decision test components; Step 4: Based on the cross-attention mechanism, cross-attention calculation is performed on the multiple perception features to obtain attention scores between different perception features; Step 5: Calculate the matching degree of different environmental perception results based on the attention scores between different perception features, and obtain a judgment result based on the matching degree.

[0064] An embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the methods provided by the above-mentioned method embodiments are implemented, for example, including: Step 1: Obtain environmental perception results of multiple heterogeneous perception decision test components pre-deployed for perceiving environmental information; the environmental perception results include obstacle information and lane line information perceived at the current moment; Step 2: Preprocess the multiple environmental perception results; Step 3: Perform feature extraction on the multiple environmental perception results after preprocessing based on the self-attention mechanism, and obtain the perception features corresponding to each of the heterogeneous perception decision test components; Step 4: Perform cross-attention calculation on the multiple perception features based on the cross-attention mechanism, and obtain attention scores between different perception features; Step 5: Calculate the matching degree of different environmental perception results based on the attention scores between different perception features, and obtain a judgment result based on the matching degree.

[0065] 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.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A method for evaluating and determining the resilience of an endogenous security network, characterized in that: include: Step 1: Obtain environmental perception results of multiple heterogeneous perception decision test components pre-deployed for perceiving environmental information; The environmental perception result includes obstacle information and lane line information perceived at the current moment; Step 2: preprocessing the plurality of environmental perception results; Step 3: extracting features from the plurality of preprocessed environmental perception results based on a self-attention mechanism to obtain perception features corresponding to each of the heterogeneous perception decision test components; Step 4: performing cross-attention calculation on the plurality of perceptual features based on the cross-attention mechanism to obtain attention scores between different perceptual features; Step 5: Calculate the matching degree of different environmental perception results according to the attention scores between different perception features, compare the calculated matching degree with the preset normal matching degree range, and judge the heterogeneous perception decision test component with the lowest matching degree with other heterogeneous perception decision test components and below the normal matching degree range as abnormal to obtain the judgment result.

2. According to claim 1, a method for evaluating and determining the resilience of an endogenous security network is characterized in that: The preprocessing includes data parsing and interface unification, timestamp unification and coordinate normalization; wherein, data parsing and interface unification refers to independently parsing the environmental perception results of each heterogeneous perception decision test component and then converting the parsed results into a unified format; timestamp unification refers to aligning the timestamps of the environmental perception results of multiple heterogeneous perception decision test components and unifying them to the same timestamp; coordinate normalization refers to unifying the environmental perception results of multiple heterogeneous perception decision test components into the same coordinate system.

3. According to claim 1, a method for evaluating and determining the resilience of an endogenous security network is characterized in that: Step 3 specifically includes: The feature extraction process is performed on the obstacle information perceived at the current moment in each environmental perception result, specifically including: extracting the initial obstacle features of each frame of data, the initial obstacle features including the obstacle position, size, direction, speed and type; sorting all the initial obstacle features according to the distance between the obstacle and the vehicle to obtain an obstacle information matrix; based on the obstacle information matrix, using the self-attention mechanism to calculate the attention scores between the obstacles, and based on the attention scores, weighting the initial obstacle features to obtain the final obstacle features; The feature extraction process is performed on the lane line information perceived at the current moment in each environmental perception result, specifically including: dividing the lane line into multiple segments at fixed intervals, and each segment is represented by the position of the starting point and the end point; sorting all the lane line segments according to the distance between themselves and the vehicle to obtain a lane line information matrix; based on the lane line information matrix, using the self-attention mechanism to calculate the attention scores between different lane line segments, and based on the attention scores, all the lane line segments are weighted to obtain lane line features.

4. According to claim 3, a method for evaluating and determining the resilience of an endogenous security network is characterized in that: Step 4 specifically includes: For any two heterogeneous perception decision test components, the key matrix and the value matrix are calculated according to the obstacle features or lane features corresponding to one of the heterogeneous perception decision test components, and the query matrix is ​​calculated according to the obstacle features or lane features corresponding to the other heterogeneous perception decision test component; the calculated key matrix, value matrix and query matrix are substituted into the cross-attention calculation formula to obtain the attention scores of the obstacle features or lane features; the attention scores of the obstacle features and the lane features are concatenated to obtain the attention score between the perception features of any two heterogeneous perception decision test components.

5. According to claim 1, a method for evaluating and determining the resilience of an endogenous security network is characterized in that: In step 5, a cosine similarity calculation function is used to calculate the matching degree of different environment perception results.

6. An endogenous security network resilience assessment and decision system, characterized in that: include: A perception information acquisition module, used to acquire environmental perception results of a plurality of heterogeneous perception decision test components pre-deployed for perceiving environmental information; The environmental perception result includes obstacle information and lane line information perceived at the current moment; A preprocessing module, used for preprocessing the plurality of environmental perception results; A feature extraction module, used to extract features from the plurality of preprocessed environmental perception results based on a self-attention mechanism, to obtain perception features corresponding to each of the heterogeneous perception decision test components; and for performing cross-attention calculation on the plurality of perceptual features based on a cross-attention mechanism to obtain attention scores between different perceptual features; The judgment module is used to calculate the matching degree of different environmental perception results according to the attention scores between different perception features, compare the calculated matching degree with the preset normal matching degree range, and judge the heterogeneous perception decision test component with the lowest matching degree with other heterogeneous perception decision test components and below the normal matching degree range as abnormal to obtain the judgment result.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images