An AI analysis-based optical fiber perimeter intrusion detection method and system

Abstract

The application is suitable for the technical field of intelligent perimeter protection, and provides a fiber perimeter intrusion detection method and system based on AI analysis, which comprises the following steps: determining a target monitoring area corresponding to a current disturbance event, acquiring historical monitoring records of a jurisdiction to which the target monitoring area belongs and an initial similarity score corresponding to the current disturbance event; and identifying a reference monitoring area with the lowest misjudgment rate based on the historical monitoring records. The application introduces a first correction factor based on the difference of physical attachment structure and a second correction factor reflecting the evolution trend of the structure, and constructs a double-factor score correction mechanism for the fiber perimeter intrusion detection scene. The mechanism can effectively perceive the difference in the propagation characteristics of the target area and the reference area in the physical attachment structure, and further combine the stability trend of the structure state changing over time to dynamically calibrate the similarity score output by the AI model.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent perimeter protection, and particularly relates to a fiber perimeter intrusion detection method and system based on AI analysis. BACKGROUND

[0002] The existing fiber perimeter intrusion detection technology mainly relies on a distributed optical fiber sensing system to collect and analyze external disturbance signals, combines signal strength, frequency characteristics and time domain patterns and the like to determine whether there is an intrusion behavior. In recent years, artificial intelligence algorithms have been widely introduced into such systems, and by constructing a pattern recognition model and training a disturbance behavior feature library, the recognition accuracy and real-time response capability of the disturbance signals are improved. Some systems introduce a similarity scoring mechanism on this basis to quantify the matching degree between the current disturbance event and historical intrusion samples to assist in decision output. However, the above scoring mostly relies on static feature extraction and fixed model output, and lacks sensitive adaptive capability to changes in the deployment environment.

[0003] Especially in large-scale deployment application scenarios, the physical structure types to which the optical fibers are attached are various, such as fences, walls, buried structures and pipelines, and there are essential differences in the response mechanism and propagation path of the disturbance signals of different attachment structures. The existing AI model is usually difficult to cover all structure types in the training process, so once the target area is inconsistent with the structure type of the training sample, it is easy to cause scoring deviation or even misjudgment. At the same time, the attachment structure of the optical fiber will also appear loosening, fatigue or aging phenomenon with the use time, which further affects the signal coupling effect, and the existing system generally lacks dynamic recognition and adaptive compensation capability for the evolution process of such structure, and it is difficult to guarantee the stability and long-term usability of the scoring result. SUMMARY

[0004] The purpose of the present application is to provide a fiber perimeter intrusion detection method and system based on AI analysis, which aims to solve the problems proposed in the background art.

[0005] The present application is implemented as follows: a fiber perimeter intrusion detection method based on AI analysis, the method comprising:

[0006] determining a target monitoring area corresponding to a current disturbance event, obtaining historical monitoring records of a jurisdiction to which the target monitoring area belongs, and an initial similarity score corresponding to the current disturbance event;

[0007] based on the historical monitoring records, identifying a reference monitoring area with the lowest misjudgment rate, and respectively extracting a preset number of disturbance samples of the target monitoring area and disturbance samples of the reference monitoring area from the historical monitoring records;

[0008] The frequency domain feature analysis is respectively performed on the target monitoring area disturbance sample and the reference monitoring area disturbance sample, high-frequency component attenuation parameters corresponding to each disturbance sample are extracted, the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples are respectively calculated, and a first correction factor is generated according to the difference amplitude between the two average values;

[0009] Based on the historical monitoring records of the target monitoring area, the attachment structure state indicators corresponding to each target monitoring area disturbance sample are extracted, a structure state evolution time sequence is constructed, and a second correction factor is generated based on the trend characteristics of the time sequence;

[0010] The initial similarity score is corrected based on the first correction factor and the second correction factor.

[0011] As a further limitation of the technical scheme of the embodiment of the application, the target monitoring area disturbance sample and the reference monitoring area disturbance sample have the following consistency features:

[0012] Both correspond to the same type of disturbance event, the disturbance categories of the disturbance events are consistent, the environmental parameters of the acquisition time period are within the same range, and the corresponding signal acquisition quality meets a preset frequency domain comparability threshold.

[0013] As a further limitation of the technical scheme of the embodiment of the application, based on the historical monitoring records, the reference monitoring area with the lowest misjudgment rate is identified, and the steps of extracting a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records include:

[0014] Based on the historical monitoring records, the historical disturbance events corresponding to each monitoring area in the preset range are extracted, and candidate monitoring areas with the number of historical disturbance events exceeding a specified threshold are screened out;

[0015] The probability that each candidate monitoring area is confirmed as a real disturbance event is calculated, and the area with the lowest misjudgment rate is identified as the reference monitoring area based on the probability;

[0016] The attachment structure types corresponding to the target monitoring area and the reference monitoring area are respectively determined, and it is judged whether they are consistent, if they are consistent, the initial similarity score is retained, and the sample comparison and correction process is not performed;

[0017] If they are not consistent, a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples are respectively extracted from the historical monitoring records.

[0018] As a further limitation of the technical scheme of the embodiment of the present application, the steps of generating the first correction factor based on the difference between the first average value and the second average value include:

[0019] Performing Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain corresponding frequency response curves;

[0020] Based on the frequency response curves, extracting the attenuation slope of the amplitude with the increase of the frequency within the preset high-frequency band as the high-frequency component attenuation parameter of each disturbance sample;

[0021] Statistically analyzing all the high-frequency component attenuation parameters of the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, and calculating the first average value corresponding to the target monitoring area disturbance samples and the second average value corresponding to the reference monitoring area disturbance samples;

[0022] Generating the first correction factor based on the difference between the first average value and the second average value.

[0023] As a further limitation of the technical scheme of the embodiment of the present application, based on the historical monitoring records of the target monitoring area, the steps of extracting the attachment structure state indicators corresponding to each target monitoring area disturbance sample, constructing a structure state evolution time series, and generating a second correction factor based on the trend characteristics of the time series include:

[0024] For each target monitoring area disturbance sample, extract its corresponding attachment structure state indicators, including one or more of the weighted combination of the fiber coupling strength, the physical connection stability, the signal response gain or the structure stiffness estimation value;

[0025] Arrange the attachment structure state indicators in the time sequence of the target monitoring area disturbance samples to construct a structure state evolution time series of the target monitoring area;

[0026] Perform fitting analysis on the time series to calculate the average slope value of the overall evolution trend, and set the average slope value as the second correction factor.

[0027] As a further limitation of the technical scheme of the embodiment of the present application, when correcting the initial similarity score, a preset correction model is used, and the correction model is:

[0028] Wherein is the corrected similarity score, is the initial similarity score, refers to the first average value, refers to the second average value, refers to the first correction factor, that is, the difference between the first average value and the second average value, refers to the adjustment weight corresponding to the first correction factor, refers to the second correction factor, that is, the average slope value of the structure state evolution time sequence, refers to the adjustment weight corresponding to the second correction factor.

[0029] An AI analysis-based optical fiber perimeter intrusion detection system, the system comprising: a data acquisition module, a sample extraction module, a first correction factor determination module, a second correction factor determination module, and a score correction module, wherein:

[0030] The data acquisition module is configured to determine a target monitoring area corresponding to a current disturbance event, acquire historical monitoring records of a jurisdiction to which the target monitoring area belongs, and an initial similarity score corresponding to the current disturbance event;

[0031] The sample extraction module is configured to identify a reference monitoring area with the lowest misjudgment rate based on the historical monitoring records, and extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records; the target monitoring area disturbance samples and the reference monitoring area disturbance samples have the following consistency features: they all correspond to the same type of disturbance event, and the disturbance events have consistent disturbance categories, the environmental parameters in the collection time period are within the same range, and the corresponding signal collection quality meets a preset frequency domain comparability threshold;

[0032] The first correction factor determination module is configured to perform frequency domain feature analysis on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, extract a high-frequency component attenuation parameter corresponding to each disturbance sample, calculate the average value of the high-frequency component attenuation parameters of the two groups of disturbance samples respectively, and generate a first correction factor according to the difference between the two average values;

[0033] The second correction factor determination module is configured to extract an attached structure state index corresponding to each target monitoring area disturbance sample based on the historical monitoring records of the target monitoring area, construct a structure state evolution time sequence, and generate a second correction factor based on the trend feature of the time sequence;

[0034] The score correction module is configured to correct the initial similarity score based on the first correction factor and the second correction factor.

[0035] As a further limitation of the technical scheme of the embodiment of the present application, the sample extraction module specifically comprises:

[0036] The data screening unit is configured to extract historical disturbance events corresponding to the monitoring area in each preset range based on the historical monitoring records, and screen out candidate monitoring areas whose number of historical disturbance events exceeds a specified threshold;

[0037] The reference monitoring area determination unit is configured to calculate a probability that a historical disturbance event in each candidate monitoring area is confirmed as a real disturbance event, and identify a monitoring area with the lowest misjudgment rate as the reference monitoring area based on the probability;

[0038] The first type consistency judgment unit is configured to determine the attachment structure types corresponding to the target monitoring area and the reference monitoring area respectively, and judge whether the two are consistent, if so, the initial similarity score is retained, and the sample comparison and correction process is not performed.

[0039] The second type consistency judgment unit is configured to extract a preset number of disturbance samples of the target monitoring area and disturbance samples of the reference monitoring area from the historical monitoring records if the two are not consistent.

[0040] As a further limitation of the technical scheme of the embodiment of the present application, the first correction factor determination module specifically comprises:

[0041] The curve acquisition unit is configured to perform Fourier transform on the disturbance samples of the target monitoring area and the disturbance samples of the reference monitoring area respectively within a preset sampling time window to obtain corresponding frequency response curves.

[0042] The attenuation parameter calculation unit is configured to extract an attenuation slope of an amplitude with increasing frequency in a preset high-frequency frequency band as a high-frequency component attenuation parameter of each disturbance sample based on the frequency response curve.

[0043] The average value calculation unit is configured to respectively count all high-frequency component attenuation parameters of the disturbance samples of the target monitoring area and the disturbance samples of the reference monitoring area, and calculate a first average value corresponding to the disturbance samples of the target monitoring area and a second average value corresponding to the disturbance samples of the reference monitoring area.

[0044] The difference amplitude quantification unit is configured to generate the first correction factor based on the difference amplitude between the first average value and the second average value.

[0045] As a further limitation of the technical scheme of the embodiment of the present application, the second correction factor determination module specifically comprises:

[0046] The state index extraction unit is configured to extract an attachment structure state index corresponding to each target monitoring area disturbance sample, and the state index comprises a weighted combination of one or more of fiber coupling strength, physical connection stability, signal response gain, or structure stiffness estimation value.

[0047] a time sequence construction unit configured to arrange the structure state indicators of the disturbance samples in the target monitoring area in time sequence according to the target monitoring area, so as to construct a structure state evolution time sequence of the target monitoring area;

[0048] a slope value calculation unit configured to perform fitting analysis on the time sequence, calculate an average slope value of the overall evolution trend, and set the average slope value as a second correction factor.

[0049] Compared with the prior art, the present application has the following beneficial effects:

[0050] The present application introduces a first correction factor based on the difference of physical attachment structure and a second correction factor reflecting the structure evolution trend, and constructs a double-factor score correction mechanism for the fiber perimeter intrusion detection scene. The mechanism can effectively perceive the difference in propagation characteristics of the target area and the reference area on the physical attachment structure, and further combine the stability trend of the structure state changing with time to dynamically calibrate the similarity score output by the AI model.

[0051] Compared with the traditional recognition method relying on fixed threshold or static feature matching, the present application has stronger structure adaptability and score credibility control ability, and is particularly suitable for the optical fiber perimeter system with strong structure heterogeneity in complex deployment environment, which improves the recognition accuracy and enhances the robustness of the system to the change of physical environment. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 a flowchart of the method provided by the embodiment of the present application;

[0053] Figure 2 a flowchart of acquiring the disturbance sample in the method provided by the embodiment of the present application;

[0054] Figure 3 a flowchart of determining the first correction factor in the method provided by the embodiment of the present application;

[0055] Figure 4 a flowchart of determining the second correction factor in the method provided by the embodiment of the present application;

[0056] Figure 5 an application architecture diagram of the system provided by the embodiment of the present application;

[0057] Figure 6 a structural block diagram of the sample extraction module in the system provided by the embodiment of the present application;

[0058] Figure 7 a structural block diagram of the first correction factor determination module in the system provided by the embodiment of the present application;

[0059] Figure 8A structural block diagram of a second correction factor determination module in a system provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0061] Figure 1 A flowchart of a method provided by an embodiment of the present application is shown.

[0062] Specifically, an optical fiber perimeter intrusion detection method based on AI analysis, which specifically comprises the following steps:

[0063] Step S100, determining a target monitoring area corresponding to a current disturbance event, obtaining historical monitoring records of a jurisdiction to which the target monitoring area belongs, and an initial similarity score corresponding to the current disturbance event.

[0064] In an embodiment of the present application, the system is deployed with distributed optical fiber sensing nodes for the target optical fiber perimeter area, which can collect vibration signals along the line in real time and transmit them to the backend analysis module. The current disturbance event refers to an abnormal disturbance signal captured by the optical fiber sensor at a certain time point. After the signal is triggered, the system automatically determines the target monitoring area corresponding to the disturbance, i.e., the physical division unit to which the optical fiber belongs, through geographical mapping rules and signal source identification. The system pre-divides the range covered by the entire optical fiber line into a plurality of spatially continuous and uniquely numbered monitoring areas according to factors such as deployment topography, installation density, signal recognition boundary, etc. These areas form a complete jurisdiction division structure for organizing and classifying the attribution relationship of different disturbance events.

[0065] The historical monitoring records refer to a set of disturbance data logs generated by each monitoring area within the system operation period, which specifically include but are not limited to the following contents: the timestamp of each disturbance event, the region number to which the event belongs, the corresponding original signal sequence (time domain waveform), frequency domain characteristic parameters (such as main frequency, amplitude spectrum, energy density distribution, etc.), disturbance type label (annotated by AI model or manually), historical false alarm or confirmation mark, environmental parameter snapshot (such as temperature and humidity, wind speed level, etc.), physical deployment state (such as whether to maintain, whether to construct), etc. These data are archived in a structured format by the central data storage module of the system after receiving the data uploaded by the sensing end, forming a historical monitoring database that can be queried and compared.

[0066] The initial similarity score refers to the score value generated by the system based on the matching results between the signal features of the current disturbance event and the historical intrusion event feature library after the current disturbance event occurs. The score is completed by an AI model (such as a convolutional neural network, a time sequence matching network, etc.) or a vector similarity calculation module (such as cosine similarity, DTW dynamic time warping), which is used to evaluate the similarity between the current disturbance and the known high-risk event. The score is usually a continuous value or interval level score in the interval [0-1], which is used as a basis for subsequent judgment of whether the event constitutes a potential intrusion behavior. The score generation method has a relatively mature application basis in the prior art, such as the AI model based on optical fiber perimeter intrusion recognition, which usually has the ability to output behavior similarity, which is the output result of the existing model judgment module and is used as an input parameter for the subsequent correction mechanism in the present application.

[0067] Further, the optical fiber perimeter intrusion detection method based on AI analysis further comprises the following steps:

[0068] Step S200, based on historical monitoring records, identifying the reference monitoring area with the lowest misjudgment rate, and extracting a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records.

[0069] The target monitoring area disturbance samples and the reference monitoring area disturbance samples have the following consistency features: they correspond to the same type of disturbance event, the disturbance categories of the disturbance events are consistent, the environmental parameters of the collection time period are within the same range, and the corresponding signal collection quality meets the preset frequency domain comparability threshold.

[0070] Specifically, Figure 2 A flowchart for obtaining disturbance samples is shown.

[0071] Among them, based on historical monitoring records, identifying the reference monitoring area with the lowest misjudgment rate, and extracting a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records specifically includes the following steps:

[0072] Step S201, based on historical monitoring records, extracting historical disturbance events corresponding to each monitoring area in a preset range, and screening out candidate monitoring areas with a number of historical disturbance events exceeding a specified threshold;

[0073] Step S202, calculating the probability that each candidate monitoring area is confirmed as a real disturbance event, and identifying the area with the lowest misjudgment rate as the reference monitoring area based on the probability;

[0074] Step S203, respectively determining the types of the attached structures corresponding to the target monitoring area and the reference monitoring area, and judging whether they are consistent, if they are consistent, retaining the initial similarity score, and not performing the sample comparison and correction process;

[0075] Step S204, if inconsistent, then extracting a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records respectively.

[0076] In the embodiments of the present application, the reason why the target monitoring area disturbance sample and the reference monitoring area disturbance sample are required to have consistent characteristics is mainly to ensure that the subsequent frequency domain feature analysis and high frequency component comparison have statistical significance and physical comparability. The consistency of disturbance event types can ensure that the signal generation mechanisms are basically the same, avoiding confusing behaviors such as “treading” and “cutting” for comparison; the consistency of disturbance categories (such as human interference vs. animal disturbance) can further limit the background environment of the interference behavior; the environmental parameters in the collection time period are within the same range in order to eliminate the interference effects of environmental conditions such as temperature, humidity, and wind speed on signal propagation; and the signal collection quality meets the frequency domain comparability threshold, which ensures the consistency of the sample in key parameters such as sampling length, amplitude dynamic range, and signal-to-noise ratio, thereby improving the accuracy of high frequency component attenuation analysis and avoiding error accumulation to distort the correction factor.

[0077] In the present application, the number of historical disturbance events exceeding a specified threshold is set as a screening condition for candidate monitoring areas, in order to ensure that the samples used to evaluate the false rejection rate have sufficient statistical basis. If the number of historical disturbance events in a certain area is too small, the false rejection rate evaluation will not be representative and is easy to be misled by individual events. Therefore, the setting of the specified threshold can ensure the stability of the calculation while improving the reliability of the selection of the reference monitoring area.

[0078] Identifying the area with the lowest false rejection rate as the reference monitoring area is to provide a “high confidence reference sample source” for feature comparison with the target monitoring area. A low false rejection rate indicates that the area has higher AI model discrimination accuracy in the historical disturbance identification process, and its disturbance features are more representative and stable, which can be regarded as a benchmark template of “true disturbance signal features”, thereby providing a reliable reference for the difference correction of the target monitoring area sample.

[0079] The attachment structure type refers to the physical attachment or installation carrier structure of the optical fiber in a specific area, including but not limited to: fences, buried soil layers, concrete floors, bridge columns, metal plates, etc. Since different structure materials, installation densities, and mechanical coupling methods can significantly affect the propagation characteristics of disturbance signals, especially in the high frequency part, the difference is particularly obvious. Therefore, the core of the present application is to establish a mechanism for correcting the differences in disturbance characteristics between different attachment structure types, to ensure that the AI discrimination output has cross-structure consistency, thereby enhancing the practicality and generalization ability of the system in complex deployment scenarios.

[0080] When it is judged that the attachment structure type of the target monitoring area and the reference monitoring area is inconsistent, the system will extract the disturbance samples of the two areas from the historical monitoring records for subsequent difference analysis and correction factor calculation. The extraction process includes: preferentially screening historical disturbance samples that meet the type consistency, environmental condition matching and signal quality standard; selecting a preset number of representative samples based on time sequence or feature distribution coverage; and performing frequency domain preprocessing on each group of samples to ensure comparability.

[0081] Further, the optical fiber perimeter intrusion detection method based on AI analysis further includes the following steps:

[0082] Step S300, frequency domain feature analysis is performed on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, the high-frequency component attenuation parameters corresponding to each disturbance sample are extracted, the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples are calculated respectively, and the first correction factor is generated according to the difference amplitude between the two average values.

[0083] Specifically, Figure 3 A flowchart for determining the first correction factor is shown.

[0084] Among them, the frequency domain feature analysis is performed on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, the high-frequency component attenuation parameters corresponding to each disturbance sample are extracted, the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples are calculated respectively, and the first correction factor is generated according to the difference amplitude between the two average values. Specifically, the following steps are included:

[0085] Step S301, Fourier transform is performed on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain corresponding frequency response curves;

[0086] Step S302, based on the frequency response curve, the attenuation slope of the amplitude with the increase of the frequency in the preset high-frequency frequency band is extracted as the high-frequency component attenuation parameter of each disturbance sample;

[0087] Step S303, the total high-frequency component attenuation parameters of the target monitoring area disturbance samples and the reference monitoring area disturbance samples are respectively counted, and the first average value corresponding to the target monitoring area disturbance samples and the second average value corresponding to the reference monitoring area disturbance samples are calculated;

[0088] Step S304, the first correction factor is generated based on the difference amplitude between the first average value and the second average value.

[0089] In the embodiment of the present application, the specific implementation process of step S301 is: obtaining the original time domain vibration signal of each disturbance sample, and performing fast Fourier transform (FFT) on the signal sequence within the system preset sampling time window (for example, 2 seconds or 5 seconds) to convert the time domain signal into a frequency domain signal, thereby obtaining the corresponding frequency response curve. The frequency response curve represents the energy distribution of the disturbance signal at different frequencies, and is a basic parameter reflecting the characteristics of the physical signal propagation path. The Fourier transform operation can be realized by a conventional signal processing library, which is a mature digital signal processing method in the prior art.

[0090] In step S302, the system selects a specific high-frequency frequency range from the obtained frequency response curve, for example, from the frequency range above 1 kHz, and extracts the amplitude variation trend in the frequency rising process. By fitting the amplitude decay curve in this section, the slope value is calculated as the high-frequency component attenuation parameter of the sample. This parameter reflects the attenuation degree of the disturbance signal when passing through the physical path, especially the transmission ability of the high-frequency component. The slope extraction can be completed by using linear regression, local polynomial fitting and other methods, and can also be realized by combining existing numerical calculation means such as least square fitting, which has a complete implementation basis in technology.

[0091] In step S303, all the extracted high-frequency component attenuation parameters of the disturbance samples in the target monitoring area and the reference monitoring area are statistically processed, and the average values of the two groups of samples are calculated respectively. The first average value represents the overall performance of the target monitoring area in the high-frequency propagation characteristics, and the second average value represents the characteristic value of the same structure of the reference monitoring area.

[0092] In step S304, the system takes the difference between the first average value and the second average value as the first correction factor, which is used to correct the initial similarity score of the current disturbance event. The core of this processing method is to reflect the difference in signal propagation characteristics caused by the inconsistency of the physical attachment structure type between the target monitoring area and the reference monitoring area. Because different structural materials (such as soil layer, fence, cement plate, etc.) have significant differences in impedance, absorption characteristics and coupling efficiency in the high-frequency propagation path, even if the disturbance behavior is the same, there is a quantifiable deviation in the high-frequency attenuation trend presented in the frequency domain. Therefore, taking the inter-group average difference of the high-frequency component attenuation parameters as the correction factor is a physical modeling path based on the structure propagation mechanism, which can objectively reflect the similarity distortion caused by the structural difference and provide a correction mechanism based on physics for the AI score result, ensuring the judgment accuracy and generalization stability of the system in the structural heterogeneous deployment environment.

[0093] Further, the optical fiber perimeter intrusion detection method based on AI analysis further comprises the following steps:

[0094] Step S400, based on the historical monitoring records of the target monitoring area, extract the attachment structure state indicators corresponding to each target monitoring area disturbance sample, construct the structure state evolution time series, and generate the second correction factor based on the trend characteristics of the time series.

[0095] The historical monitoring records also contain basic and derived data for characterizing the structure state, such as: fiber laying method and structure attachment type mark (fence, wall, buried, etc.), long-term response amplitude variation trend, unit disturbance response energy coefficient, continuous disturbance response consistency parameter, structure correlation of adjacent disturbance signals between events, system regular inspection results (such as manual inspection records, connection point maintenance log, tension adjustment record) and the like.

[0096] Specifically, Figure 4 A flowchart for determining the second correction factor is shown.

[0097] Among them, based on the historical monitoring records of the target monitoring area, extract the attachment structure state indicators corresponding to each target monitoring area disturbance sample, construct the structure state evolution time series, and generate the second correction factor based on the trend characteristics of the time series, which specifically includes the following steps:

[0098] Step S401, for each target monitoring area disturbance sample, extract its corresponding attachment structure state indicators, state indicators include one or more of the weighted combination of fiber coupling strength, physical connection stability, signal response gain or structure stiffness estimate value;

[0099] Step S402, arrange the attachment structure state indicators in the time sequence of the target monitoring area disturbance samples to construct the structure state evolution time series of the target monitoring area;

[0100] Step S403, fitting analysis is performed on the time series, the average slope value of the overall evolution trend is calculated, and the average slope value is set as the second correction factor.

[0101] In the embodiment of the present application, the extraction of the attached structure state indicator is based on dynamic feature perception of the physical attachment environment corresponding to each disturbance sample of the target monitoring area. The indicator is constructed from the following quantifiable dimensions: (1) fiber coupling strength, which can be inferred from the stability and amplitude-frequency consistency of the vibration response to the contact tightness between the fiber and its attached structure; (2) physical connection stability, which can be evaluated in combination with maintenance records, on-site manual verification data or long-term vibration response consistency level to assess the connection status; (3) signal response gain, which refers to the energy amplitude of the output signal under unit disturbance excitation, reflecting the structure energy transfer efficiency; (4) structure stiffness estimate, which can be calculated by the main frequency drift rate or low-frequency response delay characteristics in a specific frequency band. The system supports weighted combination of one or more parameters for comprehensive characterization of the current attached structure, which can be realized by known feature fusion methods such as principal component weighting, multi-feature linear regression model or expert experience weighting method, which belongs to mature data fusion technology that can be implemented.

[0102] In step S402, the system sorts the disturbance samples according to their collection time stamps, arranges the attached structure state indicators at the corresponding time points in sequence, and forms a structure state evolution time series with time dimension. This process can be realized by constructing a structured data table in the form of "time-value" key-value pairs through database query results, and generating sequence nodes based on equal interval time. This process is very common in existing time series modeling, and can be realized based on standard tools such as Pandas, Numpy or MATLAB time series functions of Python.

[0103] In step S403, the system uses fitting analysis to model the trend of the above structure state evolution time series. Common methods include linear regression (least squares line fitting), sliding window moving average slope estimation, or exponential weighted moving average trend extraction, etc. By modeling the numerical trend of the sequence, the average slope value of the overall sequence is calculated as a quantitative expression of the evolution of the structure state over time. The sign and absolute value of the slope value represent the improvement and deterioration direction and speed of the structure state, respectively. For example, a negative slope indicates that the structure is in a gradual loosening or coupling decline trend, and a slope close to zero indicates that the structure state is stable over a long period of time.

[0104] The purpose of taking the average slope value as the second correction factor is to introduce the stability change trend of the physical attachment structure corresponding to the target region from the time dimension to dynamically adjust the similarity score of the current disturbance event. The long-term coupling state of the physical attachment structure directly affects the response mechanism of the optical fiber to the disturbance signal. If the structure is in a state of continuous stability or gradual optimization, its high-frequency propagation characteristics are closer to the stable conditions of the known reference region, the current disturbance signal can be regarded as a more true physical response, and the comparison result is more reliable, and the score should be enhanced. On the contrary, if the physical attachment structure is in a loose, fatigue or degradation trend, its signal coupling path will gradually deviate from the structure of the reference region, which may introduce non-structural noise or abnormal attenuation characteristics, resulting in an initial score that is too high, and the score should be adjusted downward. The second correction factor thus constitutes a quantitative feedback mechanism for the evolution state of the physical attachment structure.

[0105] Further, the optical fiber perimeter intrusion detection method based on AI analysis further comprises the following steps:

[0106] Step S500, based on the first correction factor and the second correction factor, correcting the initial similarity score.

[0107] When correcting the initial similarity score, a preset correction model is used, and the correction model is:

[0108] , wherein denotes the similarity score after correction, denotes the initial similarity score, denotes the first average value, denotes the second average value, denotes the first correction factor, that is, the difference amplitude between the first average value and the second average value, denotes the adjustment weight corresponding to the first correction factor, denotes the second correction factor, that is, the average slope value of the structure state evolution time sequence, denotes the adjustment weight corresponding to the second correction factor.

[0109] In the embodiment of the present application, the initial similarity score is jointly corrected in combination with the first correction factor and the second correction factor, aiming to realize the coordinated optimization of the score result from the two dimensions of "lateral structural difference" and "longitudinal structural evolution". The first correction factor reflects the response deviation of the disturbance signal caused by the difference in the physical attachment structure between the target monitoring region and the reference monitoring region, which belongs to the difference compensation in the spatial structure level. The second correction factor describes the trend information of the stability change of the physical structure of the target region in the time dimension, which is used to dynamically correct the similarity score deviation caused by the long-term evolution of structure degradation, coupling loosening, etc.

[0110] The two types of correction factors can be used alone, but when they are used together, not only can the scoring calibration of the "structurally different" area be performed, but the continuous deviation trend caused by the "structurally changing" area can also be further captured, so that the scoring correction effect is more scene-adaptive and prediction-accurate. The system has the ability of longitudinal self-evolution feedback while being horizontally aligned, can significantly enhance the robustness and generalization ability of the scoring model under actual deployment, and forms a collaborative correction strategy with a feedback complementary mechanism.

[0111] The scoring correction model adopts a linear combination method which is relatively intuitive and easy to deploy, and a correction coefficient based on "deviation amount" is formed by multiplying the difference amplitude and the structure trend by their respective adjustment weights and then superimposing them. The model has good interpretability and implementation efficiency. However, it should be noted that the present application does not limit the correction model to this form, and various calculation methods such as exponential decay function, segmented interval correction strategy, neural network weight learning mechanism, or other dynamic correction models based on historical scoring deviation can be used to implement the correction process to adapt to scoring correction tasks in different complex structural environments.

[0112] Further, Figure 5 The application architecture diagram of the system provided by the embodiment of the present application is shown.

[0113] In another preferred embodiment provided by the present application, an optical fiber perimeter intrusion detection system based on AI analysis comprises:

[0114] The data acquisition module 100 is configured to determine a target monitoring area corresponding to a current disturbance event, acquire historical monitoring records of a jurisdiction to which the target monitoring area belongs, and an initial similarity score corresponding to the current disturbance event.

[0115] Further, the optical fiber perimeter intrusion detection system based on AI analysis further comprises:

[0116] The sample extraction module 200 is configured to identify a reference monitoring area with the lowest misjudgment rate based on the historical monitoring records, and extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records.

[0117] The target monitoring area disturbance samples and the reference monitoring area disturbance samples have the following consistency features: they correspond to the same type of disturbance event, the disturbance categories of the disturbance events are consistent, the environmental parameters of the collection time period are within the same range, and the corresponding signal collection quality meets a preset frequency domain comparability threshold.

[0118] Specifically, Figure 6A structural block diagram of the sample extraction module 200 in the system provided by the embodiment of the present application is shown.

[0119] In the preferred embodiment provided by the present application, the sample extraction module 200 specifically comprises:

[0120] The data screening unit 201 is configured to extract historical disturbance events corresponding to the monitoring area in each preset range based on the historical monitoring records, and screen out candidate monitoring areas whose number of historical disturbance events exceeds a specified threshold value;

[0121] The reference monitoring area determination unit 202 is configured to calculate a probability that the historical disturbance events in each candidate monitoring area are confirmed as real disturbance events, and identify the area with the lowest misjudgment rate as the reference monitoring area based on the probability;

[0122] The first type consistency judgment unit 203 is configured to determine the attachment structure types corresponding to the target monitoring area and the reference monitoring area respectively, and judge whether the two are consistent, if so, the initial similarity score is retained, and the sample comparison and correction process is not executed;

[0123] The second type consistency judgment unit 204 is configured to extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records respectively if the two are not consistent.

[0124] Further, the AI analysis-based optical fiber perimeter intrusion detection system further comprises:

[0125] The first correction factor determination module 300 is configured to perform frequency domain feature analysis on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, extract the high-frequency component attenuation parameters corresponding to each disturbance sample, calculate the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples respectively, and generate a first correction factor according to the difference between the two average values.

[0126] Specifically, Figure 7 A structural block diagram of the first correction factor determination module 300 in the system provided by the embodiment of the present application is shown.

[0127] In the preferred embodiment provided by the present application, the first correction factor determination module 300 specifically comprises:

[0128] The curve acquisition unit 301 is configured to perform Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain corresponding frequency response curves;

[0129] The attenuation parameter calculation unit 302 is configured to extract an attenuation slope of the amplitude with the increase of the frequency within a preset high-frequency frequency band as the high-frequency component attenuation parameter of each disturbance sample based on the frequency response curve.

[0130] The average value calculation unit 303 is configured to respectively calculate all high-frequency component attenuation parameters of the target monitoring area disturbance samples and the reference monitoring area disturbance samples, and calculate a first average value corresponding to the target monitoring area disturbance samples and a second average value corresponding to the reference monitoring area disturbance samples.

[0131] The difference amplitude quantification unit is configured to generate a first correction factor based on the difference amplitude between the first average value and the second average value.

[0132] Further, the AI analysis-based optical fiber perimeter intrusion detection system further comprises:

[0133] The second correction factor determination module 400 is configured to extract an attached structure state indicator corresponding to each target monitoring area disturbance sample based on historical monitoring records of the target monitoring area, construct a structure state evolution time sequence, and generate a second correction factor based on a trend feature of the time sequence.

[0134] Specifically, Figure 8 The structure block diagram of the second correction factor determination module 400 in the system provided by the embodiment of the present application is shown.

[0135] In the preferred embodiment provided by the present application, the second correction factor determination module 400 specifically comprises:

[0136] The state indicator extraction unit 401 is configured to extract an attached structure state indicator corresponding to each target monitoring area disturbance sample, and the state indicator comprises a weighted combination of one or more of the optical fiber coupling strength, the physical connection stability, the signal response gain, or the structure stiffness estimation value.

[0137] The time sequence construction unit 402 is configured to arrange the attached structure state indicators in a time sequence of the target monitoring area disturbance samples to construct a structure state evolution time sequence of the target monitoring area.

[0138] The slope value calculation unit 403 is configured to perform fitting analysis on the time sequence to calculate an average slope value of the overall evolution trend, and set the average slope value as the second correction factor.

[0139] Further, the AI analysis-based optical fiber perimeter intrusion detection system further comprises:

[0140] The score correction module 500 is configured to correct the initial similarity score based on the first correction factor and the second correction factor.

[0141] It should be understood that, although the steps in the flowcharts of the embodiments of the present application are shown in a certain order according to the arrows, the steps are not necessarily executed in the order of the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in order, and the steps can be executed in other orders. Moreover, at least some of the steps in the embodiments can include a plurality of sub-steps or a plurality of stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the sub-steps or stages is not necessarily sequential, but can be round-robin or alternately executed with at least some of the other steps or sub-steps or stages of the other steps.

[0142] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0143] The technical features of the above-mentioned embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0144] The above embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0145] The above merely describes the preferred embodiments of the present application and should not be used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. An AI analysis-based optical fiber perimeter intrusion detection method, characterized by, The method comprises: determining a target monitoring area corresponding to the current disturbance event, obtaining historical monitoring records of the jurisdiction to which the target monitoring area belongs, and an initial similarity score corresponding to the current disturbance event; based on the historical monitoring records, identifying the reference monitoring area with the lowest misjudgment rate, and extracting a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records; performing frequency domain feature analysis on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, extracting the high-frequency component attenuation parameter corresponding to each disturbance sample, calculating the average value of the high-frequency component attenuation parameters of the two groups of disturbance samples respectively, and generating a first correction factor according to the difference between the two average values; the high-frequency component attenuation parameter refers to: performing Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain the corresponding frequency response curve; based on the frequency response curve, the attenuation slope of the amplitude with the increase of the frequency is extracted in the preset high-frequency frequency band as the high-frequency component attenuation parameter of each disturbance sample; the preset high-frequency frequency band refers to the frequency interval above 1 kHz; based on the historical monitoring records of the target monitoring area, extracting the attachment structure state indicators corresponding to each target monitoring area disturbance sample, constructing a structure state evolution time series, and generating a second correction factor based on the trend characteristics of the time series; the state indicators include a weighted combination of one or more of fiber coupling strength, physical connection stability, signal response gain, or structure stiffness estimate value; based on the first correction factor and the second correction factor, the initial similarity score is corrected; when correcting the initial similarity score, a preset correction model is used, and the correction model is: wherein denotes the modified similarity score, denotes the initial similarity score, denotes the first average value, denotes the second average value, denotes the first correction factor, i.e. the difference magnitude between the first average value and the second average value, denotes the adjustment weight corresponding to the first correction factor, denotes the second correction factor, i.e. the average slope value of the structural state evolution time series, denotes the adjustment weight corresponding to the second correction factor. 2.The AI analytics-based optical fiber perimeter intrusion detection method of claim 1, wherein, the target monitoring area disturbance samples and the reference monitoring area disturbance samples have the following consistency characteristics: both correspond to the same type of disturbance event, and the disturbance categories of the disturbance events are consistent, the environmental parameters of the acquisition time period are within the same range, and the corresponding signal acquisition quality meets the preset frequency domain comparability threshold. 3.The AI analytics-based optical fiber perimeter intrusion detection method of claim 2, wherein, Based on the historical monitoring records, the reference monitoring area with the lowest misjudgment rate is identified, and a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples are extracted from the historical monitoring records, which comprises: based on the historical monitoring records, extract the historical disturbance events corresponding to the monitoring areas in each preset range, and select the candidate monitoring areas whose number of historical disturbance events exceeds the specified threshold; calculate the probability that each candidate monitoring area is confirmed as a real disturbance event, and identify the area with the lowest misjudgment rate as the reference monitoring area based on the probability; determine the attachment structure type corresponding to the target monitoring area and the reference monitoring area respectively, and determine whether they are consistent, if they are consistent, keep the initial similarity score, do not perform the sample comparison and correction process; if not, extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records. 4.The AI analytics-based optical fiber perimeter intrusion detection method of claim 3, wherein, The steps of performing frequency domain feature analysis on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, extracting the high-frequency component attenuation parameters corresponding to each disturbance sample, calculating the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples respectively, and generating the first correction factor according to the difference between the two average values include: Performing Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain the corresponding frequency response curves; Based on the frequency response curves, extracting the attenuation slope of the amplitude with the increase of the frequency within the preset high-frequency frequency band as the high-frequency component attenuation parameter of each disturbance sample; Statistically calculating the high-frequency component attenuation parameters of the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, and calculating the first average value corresponding to the target monitoring area disturbance samples and the second average value corresponding to the reference monitoring area disturbance samples; Generating the first correction factor based on the difference between the first average value and the second average value. 5.The AI analytics-based optical fiber perimeter intrusion detection method of claim 1, wherein, Based on the historical monitoring records of the target monitoring area, extracting the attachment structure state indicators corresponding to each target monitoring area disturbance sample, constructing a structure state evolution time series, and generating a second correction factor based on the trend characteristics of the time series include the following steps: For each target monitoring area disturbance sample, extract its corresponding attachment structure state indicators, including one or more of the weighted combination of fiber coupling strength, physical connection stability, signal response gain or structure stiffness estimate value; Arrange the attachment structure state indicators in chronological order of the target monitoring area disturbance samples to construct a structure state evolution time series of the target monitoring area; Perform fitting analysis on the time series to calculate the average slope value of the overall evolution trend, and set the average slope value as the second correction factor.

6. An AI analysis-based optical fiber perimeter intrusion detection system, characterized by, The system includes a data acquisition module, a sample extraction module, a first correction factor determination module, a second correction factor determination module, and a score correction module, wherein: The data acquisition module is used to determine the target monitoring area corresponding to the current disturbance event, acquire the historical monitoring records of the jurisdiction to which the target monitoring area belongs, and the initial similarity score corresponding to the current disturbance event; The sample extraction module is used to identify the reference monitoring area with the lowest misjudgment rate based on the historical monitoring records, and extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records; the target monitoring area disturbance samples and the reference monitoring area disturbance samples have the following consistency features: they all correspond to the same type of disturbance event, and the disturbance categories of the disturbance events are consistent, the environmental parameters of the acquisition time period are within the same range, and the corresponding signal acquisition quality meets the preset frequency domain comparability threshold value; The first correction factor determination module is configured to perform frequency domain feature analysis on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively, extract high-frequency component attenuation parameters corresponding to each disturbance sample, calculate the average values of the high-frequency component attenuation parameters of the two groups of disturbance samples respectively, and generate a first correction factor according to the difference between the two average values. The high-frequency component attenuation parameter refers to performing Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain corresponding frequency response curves. Based on the frequency response curves, an attenuation slope of an amplitude with an increase in frequency is extracted within a preset high-frequency frequency band as the high-frequency component attenuation parameter of each disturbance sample. The preset high-frequency frequency band refers to a frequency interval above 1 kHz. The second correction factor determination module is configured to extract an attachment structure state index corresponding to each target monitoring area disturbance sample based on historical monitoring records of the target monitoring area, construct a structure state evolution time sequence, and generate a second correction factor based on the trend characteristics of the time sequence. The state index includes a weighted combination of one or more of fiber coupling strength, physical connection stability, signal response gain, or structure stiffness estimation value. The score correction module is configured to correct the initial similarity score based on the first correction factor and the second correction factor. When correcting the initial similarity score, a preset correction model is used. The correction model is as follows: wherein denotes the modified similarity score, denotes the initial similarity score, denotes the first average value, denotes the second average value, denotes the first correction factor, i.e. the difference magnitude between the first average value and the second average value, denotes the adjustment weight corresponding to the first correction factor, denotes the second correction factor, i.e. the average slope value of the structural state evolution time series, denotes the adjustment weight corresponding to the second correction factor. 7.The AI analytics based optical fiber perimeter intrusion detection system of claim 6, wherein, The sample extraction module specifically includes: The data screening unit is configured to extract historical disturbance events corresponding to the monitoring areas in each preset range based on the historical monitoring records, and screen out candidate monitoring areas with a number of historical disturbance events exceeding a specified threshold. The reference monitoring area determination unit is configured to calculate the probability that a historical disturbance event in each candidate monitoring area is confirmed as a real disturbance event, and identify the area with the lowest misjudgment rate as the reference monitoring area based on the probability. The first type consistency judgment unit is configured to determine the attachment structure types corresponding to the target monitoring area and the reference monitoring area respectively, and judge whether they are consistent. If they are consistent, the initial similarity score is retained, and the sample comparison and correction process is not performed. The second type consistency judgment unit is configured to extract a preset number of target monitoring area disturbance samples and reference monitoring area disturbance samples from the historical monitoring records if they are not consistent. 8.The AI analytics based optical fiber perimeter intrusion detection system of claim 7, wherein, The first correction factor determination module specifically includes: The curve acquisition unit is configured to perform Fourier transform on the target monitoring area disturbance samples and the reference monitoring area disturbance samples respectively within a preset sampling time window to obtain corresponding frequency response curves. The attenuation parameter calculation unit is configured to extract an attenuation slope of an amplitude with an increase in frequency within a preset high-frequency frequency band as the high-frequency component attenuation parameter of each disturbance sample based on the frequency response curves. The average value calculation unit is configured to respectively count all high-frequency component attenuation parameters of the target monitoring area disturbance samples and the reference monitoring area disturbance samples, and calculate a first average value corresponding to the target monitoring area disturbance samples and a second average value corresponding to the reference monitoring area disturbance samples. The difference amplitude quantization unit is configured to generate the first correction factor based on the difference amplitude between the first average value and the second average value. 9.The AI analytics based optical fiber perimeter intrusion detection system of claim 8, wherein, The second correction factor determination module specifically comprises: The state index extraction unit is configured to extract, for each target monitoring area disturbance sample, its corresponding attachment structure state index, which includes a weighted combination of one or more of fiber coupling strength, physical connection stability, signal response gain, or structure stiffness estimation value. The time series construction unit is configured to arrange the attachment structure state index in a time sequence of the target monitoring area disturbance sample to construct a structure state evolution time series of the target monitoring area. The slope value calculation unit is configured to perform fitting analysis on the time series, calculate the average slope value of the overall evolution trend, and set the average slope value as the second correction factor.

Images