Subway external intrusion event identification method and device, electronic equipment and storage medium

Abstract

The application discloses a subway external intrusion event identification method and device, electronic equipment and a storage medium. The method converts grating array vibration data into frequency spectrum image data through short-time Fourier transform, improves the processability of the data, and facilitates the improvement of the data analysis effect. Further, the frequency spectrum image data is subjected to enhancement processing, normalization processing, data expansion processing, filter correction and standardization processing, which can greatly improve the feature characteristics of the obtained target frequency spectrum image data, and further, in the process of identifying external intrusion events through a target frequency spectrum image recognition network model, the reliability of the identification result can be better ensured.

Description

Technical Field

[0001] This invention relates to the field of fiber optic monitoring technology, and in particular to a method, device, electronic device, and storage medium for identifying external intrusion events in subways. Background Technology

[0002] As subway lines continue to expand, surface construction is becoming increasingly integrated with subway systems. Drilling operations above subway lines generate intense vibrations that propagate through the ground, reducing soil resistance and increasing structural stress. This can have a destructive impact on underground subway tunnels and track structures. Improperly controlled drilling locations could potentially penetrate the subway tunnel directly, posing a serious threat to the operating route. These potential safety hazards are difficult to assess and predict in advance, and the consequences of an accident could be disastrous. Therefore, it is crucial to strengthen monitoring and early warning systems and implement scientific construction practices to eliminate the possibility of safety accidents.

[0003] Currently, the main safety monitoring solutions for drilling operations above subway stations include: setting up settlement monitoring points around subway tunnels or stations to determine if the subway structure is affected by the drilling; identifying anomalies around the subway through video comparison and sending real-time alarms; and monitoring construction above the subway using manual line patrols to issue warnings for any construction activities that could harm the subway. These methods often rely on a single monitoring indicator, have a low density and coverage of monitoring points, cannot provide comprehensive and timely effective monitoring, and have difficulty guaranteeing monitoring quality. However, using arrayed fiber optic grating vibration sensors deployed along the subway tunnel walls can detect vibration signals generated by illegal drilling operations on the ground, enabling identification and alarm triggering. However, current methods only analyze vibration signals using traditional methods, identifying the type of construction by recognizing straight and curved signals in the spectrum. This approach is difficult to extract signal features, resulting in poor identification and a high risk of false alarms.

[0004] Therefore, existing technologies for monitoring subway intrusion signals via optical fiber suffer from poor signal feature extraction, making it difficult to accurately identify intrusion events. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device, electronic device and storage medium for identifying subway external intrusion events, in order to address the problem that existing technologies, when monitoring subway intrusion signals through optical fibers, are unable to accurately identify intrusion events due to poor signal feature extraction.

[0006] To address the aforementioned problems, this invention provides a method for identifying external intrusion events in subway systems, comprising:

[0007] Vibration data of the grating array above the subway was acquired, and the vibration data of the grating array was converted into spectral image data through short-time Fourier transform;

[0008] Preprocessing the spectral image data yields the target spectral image data;

[0009] External intrusion event identification is performed on target spectrum image data based on a well-trained target spectrum image recognition network model.

[0010] The preprocessing process includes enhancement processing, normalization processing, data expansion processing, filtering correction, and standardization processing.

[0011] Furthermore, the spectral image data is enhanced to obtain the target spectral image data, including:

[0012] Signal enhancement is performed on the spectral image data using logarithmic transformation to obtain enhanced spectral image data:

[0013] The function corresponding to the logarithmic transformation is:

[0014] y = 20 * log 10 x

[0015] x represents the pixel value of the spectral image data, and y represents the pixel value of the enhanced spectral image data.

[0016] Furthermore, the enhanced spectral image data is normalized to obtain the target spectral image data, including:

[0017] The enhanced spectral image data is normalized according to the normalization formula to obtain normalized spectral image data.

[0018] The normalization formula is as follows:

[0019] Normalization (y) =(y-min(y)) / (max(y)-min(y))

[0020] Normalization (y) y represents the pixel value of the normalized spectral image data, min(y) represents the minimum pixel value of the enhanced spectral image data, and max(y) represents the maximum pixel value of the enhanced spectral image data.

[0021] Furthermore, the normalized spectral image data undergoes data expansion processing to obtain the target spectral image data, including:

[0022] The normalized spectral image data is extended by performing data expansion processing using bilinear interpolation to obtain extended spectral image data.

[0023] Furthermore, the spread spectrum image data is filtered and corrected to obtain the target spectrum image data, including:

[0024] Based on the image transformation formula, the spread spectrum image data is filtered and corrected to obtain the corrected spectrum image data;

[0025] The image transformation formula is as follows:

[0026] z = y γ

[0027] z represents the pixel value of the corrected spectral image data, and the gamma value γ is a constant.

[0028] Furthermore, the corrected spectral image data is standardized to obtain the target spectral image data, including:

[0029] The corrected spectrum image data is standardized according to the standardization formula to obtain the target spectrum image data.

[0030] The standardized formula is as follows:

[0031] Q = (z - mean(z) / std(z))

[0032] Q represents the target spectral image data, mean(z) represents the average pixel value of the corrected spectral image data, and std(z) represents the standard deviation of the pixel value of the corrected spectral image data.

[0033] Furthermore, the target spectrum image recognition network model includes a multi-layer residual neural network module, a multi-scale Transformer module, a cross-channel attention module, and a classification probability fusion module; based on the fully trained target spectrum image recognition network model, external intrusion event identification is performed on target spectrum image data, including:

[0034] Global visual information is obtained by performing global visual extraction on the target spectral image data using a multi-layer residual neural network module.

[0035] Based on the multi-scale Transformer module, the correlation relationship of global visual information is learned to obtain the multi-scale location spatial attention relationship of the target spectral image data;

[0036] Based on the cross-channel attention module, the multi-scale location spatial attention relationship is cross-fused to obtain the local feature information of the target spectral image data;

[0037] The classification probability fusion module combines global visual information and local feature information to generate classification decision features, and uses these features to identify external intrusion events.

[0038] To address the aforementioned problems, the present invention also provides a subway external intrusion event identification device, comprising:

[0039] The spectrum image data acquisition module is used to acquire vibration data of the grating array above the subway and convert the vibration data of the grating array into spectrum image data through short-time Fourier transform.

[0040] The target spectrum image data acquisition module is used to preprocess the spectrum image data to obtain the target spectrum image data;

[0041] The external intrusion event identification module is used to identify external intrusion events from target spectrum image data based on a fully trained target spectrum image recognition network model.

[0042] The preprocessing process includes enhancement processing, normalization processing, data expansion processing, filtering correction, and standardization processing.

[0043] To address the aforementioned problems, the present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it implements the subway external intrusion event identification method described above.

[0044] To address the aforementioned problems, the present invention also provides a storage medium storing computer program instructions. When the computer program instructions are executed by a computer, the computer performs the subway external intrusion event identification method described above.

[0045] The beneficial effects of adopting the above technical solution are as follows: This invention provides a method, device, electronic device, and storage medium for identifying external intrusion events in subways. The method first converts the vibration data of the grating array into spectral image data through short-time Fourier transform, which improves the data processability and facilitates data analysis. Furthermore, the spectral image data is enhanced, normalized, expanded, filtered, corrected, and standardized, which greatly improves the feature characteristics of the obtained target spectral image data. Thus, in the process of identifying external intrusion events through the target spectral image recognition network model, the reliability of the identification results can be better guaranteed. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an embodiment of the subway external intrusion event identification method provided by the present invention;

[0047] Figure 2 A schematic diagram showing the result of an embodiment of the time spectrum signal under normal operating conditions provided by the present invention;

[0048] Figure 3 A schematic diagram showing the result of an embodiment of the time spectrum signal of the demolition machine under operating conditions provided by the present invention;

[0049] Figure 4A schematic diagram showing the results of an embodiment of the time-spectrum signal of drilling conditions provided by the present invention;

[0050] Figure 5 This is a flowchart illustrating an embodiment of the present invention for identifying external intrusion events;

[0051] Figure 6 A schematic diagram of the structure of an embodiment of the subway external intrusion event identification device provided by the present invention;

[0052] Figure 7 A structural block diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0053] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0054] As subway lines continue to expand, surface construction is becoming increasingly integrated with subway systems. Drilling operations above subway lines generate intense vibrations that propagate through the ground, reducing soil resistance and increasing structural stress. This can have a destructive impact on underground subway tunnels and track structures. Improperly controlled drilling locations could potentially penetrate the subway tunnel directly, posing a serious threat to the operating route. These potential safety hazards are difficult to assess and predict in advance, and the consequences of an accident could be disastrous. Therefore, it is crucial to strengthen monitoring and early warning systems and implement scientific construction practices to eliminate the possibility of safety accidents.

[0055] Currently, the main safety monitoring solutions for drilling operations above subway stations include: setting up settlement monitoring points around subway tunnels or stations to determine if the subway structure is affected by the drilling; identifying anomalies around the subway through video comparison and sending real-time alarms; and monitoring construction above the subway using manual line patrols to issue warnings for any construction activities that could harm the subway. These methods often rely on a single monitoring indicator, have a low density and coverage of monitoring points, cannot provide comprehensive and timely effective monitoring, and have difficulty guaranteeing monitoring quality. However, using arrayed fiber optic grating vibration sensors deployed along the subway tunnel walls can detect vibration signals generated by illegal drilling operations on the ground, enabling identification and alarm triggering. However, current methods only analyze vibration signals using traditional methods, identifying the type of construction by recognizing straight and curved signals in the spectrum. This approach is difficult to extract signal features, resulting in poor identification and a high risk of false alarms.

[0056] Therefore, existing technologies for monitoring subway intrusion signals via optical fiber suffer from poor signal feature extraction, making it difficult to accurately identify intrusion events.

[0057] To address the aforementioned problems, this invention provides a method, apparatus, electronic device, and storage medium for identifying external intrusion events in subway systems, which will be described in detail below.

[0058] like Figure 1 As shown, Figure 1 A flowchart illustrating an embodiment of the subway external intrusion event identification method provided by the present invention includes:

[0059] Step S101: Obtain vibration data of the grating array above the subway, and convert the vibration data of the grating array into spectral image data through short-time Fourier transform;

[0060] Step S102: Preprocess the spectrum image data to obtain the target spectrum image data;

[0061] Step S103: Based on the fully trained target spectrum image recognition network model, identify external intrusion events in the target spectrum image data;

[0062] The preprocessing process includes enhancement processing, normalization processing, data expansion processing, filtering correction, and standardization processing.

[0063] In this embodiment, firstly, vibration data of the grating array above the subway is acquired, and then the vibration data of the grating array is converted into spectral image data through short-time Fourier transform; then, the spectral image data is preprocessed to obtain target spectral image data; finally, external intrusion event identification is performed on the target spectral image data based on a well-trained target spectral image recognition network model.

[0064] In this embodiment, the vibration data of the grating array is first converted into spectral image data through short-time Fourier transform, which improves the data processability and facilitates data analysis. Furthermore, the spectral image data is enhanced, normalized, expanded, filtered, corrected, and standardized, which greatly improves the feature characteristics of the obtained target spectral image data. As a result, the reliability of the identification results can be better guaranteed during the identification of external intrusion events through the target spectral image recognition network model.

[0065] In one specific embodiment, in step S101, in order to obtain spectral image data, firstly, the vibration data of the grating array above the subway is acquired, then the data is filtered by a 1Hz high-pass filter, and then the original signal data is converted into time spectral image data using short-time Fourier transform.

[0066] Furthermore, during the data model training process, specific spectral image data requires targeted classification. To determine whether the spectral image data corresponds to an external intrusion event or a non-intrusion event, firstly, the spectral data is divided into several independent data sets according to a certain duration and sliding window size, retaining the frequency ranges with drilling operation characteristics. Then, the dataset is labeled. Specifically, firstly, the spectral signal characteristics of normal operating conditions corresponding to non-intrusion events are: signals without obvious vibrations in the frequency dimension, or signals with regular vibrations, such as... Figure 2 As shown, Figure 2 This is a schematic diagram of the result of an embodiment of the time spectrum signal under normal operating conditions provided by the present invention.

[0067] Second, the characteristics of the breaching machine's operating conditions corresponding to external intrusion events are: multiple continuous features with small amplitudes, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the result of an embodiment of the time spectrum signal of the demolition machine under operating conditions provided by the present invention.

[0068] Third, the drilling conditions corresponding to external intrusion events are characterized by alternating linear, oscillating, and linear patterns in the frequency dimension, such as... Figure 4 As shown, Figure 4 This is a schematic diagram of the result of an embodiment of the time spectrum signal of drilling conditions provided by the present invention.

[0069] As a preferred embodiment, in step S102, in order to explain in detail the specific process of enhancing the spectrum image data to obtain the target spectrum image data, specifically, the spectrum image data is enhanced by logarithmic transformation to obtain enhanced spectrum image data;

[0070] The function corresponding to the logarithmic transformation is:

[0071] y = 20 * log 10 x

[0072] x represents the pixel value of the spectral image data, and y represents the pixel value of the enhanced spectral image data.

[0073] Secondly, after obtaining the enhanced spectrum image data, normalization processing is required. Specifically, the enhanced spectrum image data is normalized according to the normalization formula to obtain normalized spectrum image data.

[0074] The normalization formula is as follows:

[0075] Normalization (y) =(y-min(y)) / (max(y)-min(y))

[0076] Normalization (y) y represents the pixel value of the normalized spectral image data, min(y) represents the minimum pixel value of the enhanced spectral image data, and max(y) represents the maximum pixel value of the enhanced spectral image data.

[0077] Next, the normalized spectral image data needs to be extended. Specifically, bilinear interpolation is used to extend the normalized spectral image data to obtain extended spectral image data.

[0078] Furthermore, the spread spectrum image data after the expansion processing needs to be filtered and corrected. Specifically, the spread spectrum image data is filtered and corrected according to the image transformation formula to obtain the corrected spectrum image data.

[0079] The image transformation formula is as follows:

[0080] z = y γ

[0081] z represents the pixel value of the corrected spectral image data, and the gamma value γ is a constant.

[0082] Finally, in order to improve the uniformity of the data, it is also necessary to standardize the calibrated spectrum image data to obtain the target spectrum image data. Specifically, the calibrated spectrum image data is standardized according to the standardization formula to obtain the target spectrum image data.

[0083] The standardized formula is as follows:

[0084] Q = (z - mean(z)) / std(z)

[0085] Q represents the target spectral image data, mean(z) represents the average pixel value of the corrected spectral image data, and std(z) represents the standard deviation of the pixel value of the corrected spectral image data.

[0086] In one specific embodiment, in order to better identify the straight and curved features in the subway external intrusion spectrum, it is necessary to preprocess the data to remove the influence of noise and enhance useful information.

[0087] Specifically, firstly, the original signal is enhanced by using logarithmic transformation, which allows signals under different operating conditions to exhibit the characteristics of straight lines and curves on the image, which is beneficial for identification.

[0088] Secondly, normalizing the data to a range of 0 to 1 makes the importance of each feature clearer and the influence of each feature more uniform, which helps to improve the model's convergence speed, generalization ability, and computational stability.

[0089] Then, bilinear interpolation was used on the data to expand it from the original 144×310 to 256×256, which is beneficial for convolution operations in the network, reduces the amount of computation, and accelerates model training.

[0090] Next, gamma filtering can reduce noise in spectral image data and enhance the frequency signals in the data. It works by adjusting the gamma value γ of the image to change its brightness and contrast: specifically, when the gamma value is 1, the image is linear; when the gamma value is less than 1, the image becomes darker; and when the gamma value is greater than 1, the image becomes brighter.

[0091] Finally, a minimum value is assigned to the zero value in the data, and the data is standardized to convert it into a standard Gaussian distribution with values ​​between -1 and 1, a mean of 0, and a standard deviation of 1. This allows the data to respond better to the activation function, improves the data's expressiveness, reduces gradient explosion and gradient vanishing, and facilitates model training.

[0092] In this embodiment, by performing targeted feature extraction on the spectrum image data, it is possible to improve the analysis perspective of effective data while ignoring interference information, thereby better ensuring the reliability of the final target spectrum image data.

[0093] In a preferred embodiment, in step S103, the target spectrum image recognition network model specifically includes a multi-layer residual neural network module, a multi-scale Transformer module, a cross-channel attention module, and a classification probability fusion module; and in order to identify external intrusion events from target spectrum image data based on the fully trained target spectrum image recognition network model, such as Figure 5 As shown, Figure 5 A flowchart illustrating an embodiment of external intrusion event identification provided by the present invention includes:

[0094] Step S131: Perform global visual extraction on the target spectral image data using the multi-layer residual neural network module to obtain global visual information;

[0095] Step S132: Learn the correlation between global visual information based on the multi-scale Transformer module to obtain the multi-scale location spatial attention relationship of the target spectral image data;

[0096] Step S133: Cross-fuse the multi-scale location spatial attention relationship according to the cross-channel attention module to obtain the local feature information of the target spectral image data;

[0097] Step S134: Combine global visual information and local feature information according to the classification probability fusion module to generate classification decision features, and identify external intrusion events based on the classification decision features.

[0098] In this embodiment, firstly, global visual information is obtained by performing global visual extraction on the target spectrum image data using a multi-layer residual neural network module; secondly, multi-scale spatial attention relationships are obtained by learning the correlation between the global visual information and the target spectrum image data using a multi-scale Transformer module; then, the multi-scale spatial attention relationships are cross-fused using a cross-channel attention module to obtain local feature information of the target spectrum image data; finally, the global visual information and local feature information are combined using a classification probability fusion module to generate classification decision features, and external intrusion event identification is performed based on the classification decision features.

[0099] In this embodiment, by extracting the distribution features of the target spectral image data, not only is global visual information extracted through the multi-layer residual neural network module, but also local feature extraction is performed on the global visual information through the multi-scale Transformer module and the cross-channel attention module to obtain local feature information. Finally, the classification probability fusion module combines the global visual information and local feature information to generate classification decision features, so as to obtain the classification decision result by combining global features and local features.

[0100] Specifically, extracting convolutional features from different levels of the spectrogram using a multi-layer residual neural network module helps the model extract multi-dimensional feature information from the spectrogram. These convolutional features include low-level features such as color and texture, as well as higher-level semantic features such as frequency distribution. The multi-scale Transformer module can learn the attention relationships between different positions and subspaces of the spectrogram, enhancing the model's ability to learn the whole picture. These attention relationships include the correlation between different frequency bands. The cross-channel attention module cross-fuses the features extracted by different multi-layer residual neural network modules and the multi-scale Transformer module to form a composite feature representation that includes contextual relationships, which helps extract the overall structural information of the spectrogram. The classification probability fusion module combines global visual features and local features to generate the final classification decision features, which is beneficial for considering both the overall and local details of the spectrogram and improving recognition accuracy.

[0101] It should be noted that, in order to ensure the reliability of the target spectrum image recognition network model, the initial spectrum image recognition network model needs to be trained and the training results verified. Specifically, firstly, the labeled subway external intrusion data is split. After the data is randomly shuffled, the dataset is split into training set, validation set and test set in a ratio of 8:1:1. Then, the original data and the label corresponding to each data are packaged according to the above split.

[0102] It should be noted that the original signal data samples are extremely unevenly distributed, with the number of normal operating conditions far exceeding the number of alarm conditions such as demolition machines and drilling rigs. This imbalance in sample size significantly impacts the training performance of the network model. Therefore, it is necessary to reselect the training samples to form a relatively balanced subset for training, ensuring a more balanced amount of data across different operating conditions and thus mitigating the impact of imbalanced data on the training results.

[0103] Then, during training, the training sample dataset is input into the network model for iterative training. Each training iteration includes two steps: forward propagation and back propagation. a) Forward propagation: A batch of spectrum samples is input into the network model, and the output result is obtained through forward computation. b) Back propagation: The classification cross-entropy loss function is calculated, and the gradient of the weight parameters is calculated based on the loss. c) Parameter update: The Adam optimization algorithm is used to update the parameters based on the gradient of the loss function, realizing mini-batch stochastic gradient descent training.

[0104] The following adjustment strategies were adopted during training: a) The number of iterations was increased from 100 to 300 to ensure model convergence in recognizing spectral features. b) The learning rate was adjusted using a gradual decreasing or increasing strategy to find the optimal learning efficiency for the network in learning spectral data features. c) The batch size was optimized to balance the computational efficiency of spectral samples and convergence performance. d) A dropout layer was added to improve the model's generalization ability. Through continuous iteration and parameter updates, a network model with high classification accuracy was finally obtained.

[0105] Finally, the model training results were tested. In one specific embodiment, the optimal network model obtained from the training was applied to a test set of spectrum data for evaluation. The overall classification accuracy of the test results was 97.69%, the false alarm rate was 1.85%, and the false negative rate was 1.06%, indicating that the model can effectively identify most alarm events and has a high recognition accuracy.

[0106] In one specific embodiment, to avoid false alarms, the traditional monitoring method is combined with the deep learning-based subway external intrusion event identification method in this application to jointly monitor and alarm subway external intrusion events. Specifically, the event is considered real only when both the traditional monitoring method and the subway external intrusion event identification method determine that a subway external intrusion event has occurred, and then an alarm operation is performed.

[0107] In this embodiment, in addition to identifying external intrusion events based on the target spectrum image recognition network model, traditional monitoring methods are also incorporated. Since traditional monitoring methods often mistakenly identify interference events as external intrusion events, incorporating traditional monitoring methods will not affect the accuracy of the final external intrusion event identification. On the contrary, it will reduce the false alarm rate of the target spectrum image recognition network model from the perspective of current monitoring.

[0108] The above method first converts the vibration data of the grating array into spectral image data through short-time Fourier transform, improving the data's processability and thus enhancing the data analysis effect. Furthermore, by performing enhancement, normalization, data expansion, filtering correction, and standardization on the spectral image data, the characteristic features of the obtained target spectral image data can be significantly improved. This ensures the reliability of the identification results during the identification of external intrusion events using the target spectral image recognition network model.

[0109] To address the aforementioned problems, the present invention also provides a subway external intrusion event identification device, such as... Figure 6 As shown, Figure 6 This is a schematic diagram of an embodiment of the subway external intrusion event identification device provided by the present invention. The subway external intrusion event identification device 600 includes:

[0110] The spectrum image data acquisition module 601 is used to acquire vibration data of the grating array above the subway and convert the vibration data of the grating array into spectrum image data through short-time Fourier transform.

[0111] The target spectrum image data acquisition module 602 is used to preprocess the spectrum image data to obtain the target spectrum image data;

[0112] The external intrusion event identification module 603 is used to identify external intrusion events from target spectrum image data based on a fully trained target spectrum image identification network model.

[0113] The preprocessing process includes enhancement processing, normalization processing, data expansion processing, filtering correction, and standardization processing.

[0114] The present invention also provides an electronic device, such as... Figure 7 As shown, Figure 7 This is a structural block diagram of an embodiment of the electronic device provided by the present invention. The electronic device 700 can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The electronic device 700 includes a processor 701, a memory 702, and a display 703, wherein the memory 702 stores a subway external intrusion event identification program.

[0115] In some embodiments, memory 702 may be an internal storage unit of a computer device, such as a hard disk or memory. In other embodiments, memory 702 may be an external storage device of a computer device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, memory 702 may include both internal and external storage units of the computer device. Memory 702 is used to store application software and various types of data installed on the computer device, such as program code for installing the computer device. Memory 702 can also be used to temporarily store data that has been output or will be output. In one embodiment, a subway external intrusion event identification program can be executed by processor 701 to implement the subway external intrusion event identification method of the various embodiments of the present invention.

[0116] In some embodiments, processor 701 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 702 or process data, such as executing a subway external intrusion event identification program.

[0117] In some embodiments, display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 703 is used to display information from electronic device 700 and to display a visual user interface. Components 701-703 of electronic device 700 communicate with each other via a system bus.

[0118] This embodiment also provides a computer-readable storage medium storing a subway external intrusion event identification program. When the program is executed by a processor, it implements the subway external intrusion event identification method as described in any of the above technical solutions.

[0119] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other storage media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0120] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying external intrusion events in a subway system, characterized in that, include: Vibration data of a grating array above a subway station is acquired, and the vibration data of the grating array is converted into spectral image data through short-time Fourier transform; The spectral image data is preprocessed to obtain the target spectral image data; External intrusion event identification is performed on the target spectrum image data based on a fully trained target spectrum image recognition network model. The preprocessing includes enhancement processing, normalization processing, data expansion processing, filtering correction and standardization processing in sequence. The target spectrum image recognition network model includes a multi-layer residual neural network module, a multi-scale Transformer module, a cross-channel attention module, and a classification probability fusion module; the external intrusion event recognition based on the fully trained target spectrum image recognition network model includes: Global visual information is obtained by performing global visual extraction on the target spectral image data based on the multi-layer residual neural network module. Based on the multi-scale Transformer module, the global visual information is learned to obtain the multi-scale location spatial attention relationship of the target spectral image data; The multi-scale location spatial attention relationship is cross-fused according to the cross-channel attention module to obtain the local feature information of the target spectral image data; The classification probability fusion module combines the global visual information and the local feature information to generate classification decision features, and uses these features to identify external intrusion events.

2. The subway external intrusion event identification method according to claim 1, characterized in that, The spectral image data is enhanced to obtain target spectral image data, including: The spectrum image data is enhanced by performing a logarithmic transformation to obtain enhanced spectrum image data; The function corresponding to the logarithmic transformation is: The pixel values ​​of the spectral image data. The pixel value of the enhanced spectral image data.

3. The subway external intrusion event identification method according to claim 2, characterized in that, The enhanced spectral image data is normalized to obtain the target spectral image data, including: The enhanced spectral image data is normalized according to the normalization formula to obtain normalized spectral image data. The normalization formula is as follows: The pixel values ​​of the normalized spectral image data. To enhance the minimum pixel value of spectral image data, To enhance the maximum pixel value of the spectral image data.

4. The subway external intrusion event identification method according to claim 3, characterized in that, The normalized spectral image data is subjected to data expansion processing to obtain target spectral image data, including: The normalized spectral image data is extended by bilinear interpolation to obtain extended spectral image data.

5. The subway external intrusion event identification method according to claim 4, characterized in that, The spread spectrum image data is filtered and corrected to obtain the target spectrum image data, including: According to the image transformation formula, the spread spectrum image data is filtered and corrected to obtain corrected spectrum image data; The image transformation formula is as follows: The pixel value of the corrected spectral image data is γ, which is a constant.

6. The subway external intrusion event identification method according to claim 5, characterized in that, The corrected spectral image data is standardized to obtain the target spectral image data, including: The corrected spectral image data is standardized according to the standardization formula to obtain the target spectral image data. The standardized formula is as follows: Q represents the target spectral image data. The average value of the pixel values ​​in the corrected spectral image data. The standard deviation of the pixel values ​​in the corrected spectral image data is given.

7. A subway external intrusion event identification device, characterized in that, include: The spectrum image data acquisition module is used to acquire vibration data of the grating array above the subway, and convert the vibration data of the grating array into spectrum image data through short-time Fourier transform; The target spectrum image data acquisition module is used to preprocess the spectrum image data to obtain the target spectrum image data; An external intrusion event identification module is used to identify external intrusion events in the target spectrum image data based on a fully trained target spectrum image recognition network model. The preprocessing includes enhancement processing, normalization processing, data expansion processing, filtering correction and standardization processing in sequence. The target spectrum image recognition network model includes a multi-layer residual neural network module, a multi-scale Transformer module, a cross-channel attention module, and a classification probability fusion module; the external intrusion event recognition based on the fully trained target spectrum image recognition network model includes: Global visual information is obtained by performing global visual extraction on the target spectral image data based on the multi-layer residual neural network module. Based on the multi-scale Transformer module, the global visual information is learned to obtain the multi-scale location spatial attention relationship of the target spectral image data; The multi-scale location spatial attention relationship is cross-fused according to the cross-channel attention module to obtain the local feature information of the target spectral image data; The classification probability fusion module combines the global visual information and the local feature information to generate classification decision features, and uses these features to identify external intrusion events.

8. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the subway external intrusion event identification method as described in any one of claims 1-6.

9. A storage medium, characterized in that, The storage medium stores computer program instructions, which, when executed by a computer, cause the computer to perform the subway external intrusion event identification method according to any one of claims 1 to 6.

Images