A fire hydrant abnormal state detection method and system based on multi-source data fusion

Abstract

The application discloses a fire hydrant abnormal state detection method and system based on multi-source data fusion. The method constructs a baseline by collecting the position, attitude angle and preset vibration data of the fire hydrant in a static state and sets a dynamic alarm threshold; real-time monitoring compares real-time data with the threshold, intelligently triggers image capture, and obtains visual evidence of position and vibration abnormalities; within a preset time window, vibration and displacement events are analyzed in association, and are classified into two categories: physically associated and independent; finally, multi-modal features are extracted, a multi-level fusion decision tree model is used for intelligent analysis, and various abnormal event types such as traffic accident impact, human damage, construction relocation, natural loosening and the like are accurately identified, and corresponding alarms are automatically generated. The application realizes full-process automatic monitoring from multi-source perception, intelligent triggering to fusion diagnosis, effectively improves detection accuracy and reduces false alarm rate, and provides reliable support for intelligent operation and maintenance of fire hydrants.

Description

Technical Field

[0001] This invention relates to the field of fire hydrant anomaly detection, and more specifically, to a method and system for detecting abnormal states of fire hydrants based on multi-source data fusion. Background Technology

[0002] With the continuous deepening of urban public safety and fire emergency response system construction, highly reliable and timely real-time detection and intelligent diagnosis of the operational status of fire hydrants throughout urban areas has become crucial for ensuring fire rescue efficiency, preventing damage to public facilities, and optimizing urban water management. The traditional static operation and maintenance model, relying on regular manual inspections and passive reporting, struggles to capture the immediate occurrence and status changes of abnormal events such as impacts, water theft, illegal encroachment, or natural damage to fire hydrants. This model has significant shortcomings in rapid event response, accurate responsibility identification, and refined management throughout the facility's entire lifecycle. Therefore, developing a new fire hydrant abnormal status detection technology that integrates multi-source sensing, intelligent triggering, image verification, and possesses event correlation analysis and intelligent decision-making capabilities is of significant application value and management urgency for building a smart fire protection IoT network and improving the level of proactive safety protection and precise operation and maintenance management of public facilities.

[0003] Existing technologies primarily rely on single sensors (such as pressure or flow meters) for detection, making it difficult to distinguish the causes of anomalies. They are also susceptible to environmental interference, leading to high false alarm rates, and cannot effectively detect non-water pressure anomalies (such as impacts, tilting, or obstructions). Furthermore, traditional methods lack multi-source data collaborative analysis and image verification mechanisms, resulting in incomplete alarm information, an inability to accurately pinpoint the event type, and hindering subsequent rapid response and handling. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for detecting abnormal states of fire hydrants based on multi-source data fusion, so as to at least solve the problems of single detection methods and inaccurate detection results in the prior art, thereby improving the operation and maintenance efficiency of fire hydrants.

[0005] To achieve the above objectives, a method and system for detecting abnormal states of fire hydrants based on multi-source data fusion are provided.

[0006] In a first aspect, the present invention provides a method for detecting abnormal states of fire hydrants based on multi-source data fusion, the detection method comprising:

[0007] Acquire preset position data, preset attitude angle data and preset vibration data of fire hydrant in static state, construct position baseline, attitude angle baseline and vibration baseline of fire hydrant based on preset position data, preset attitude angle data and preset vibration data, and set position alarm threshold, attitude angle alarm threshold and vibration alarm threshold based on position baseline, attitude angle baseline and vibration baseline.

[0008] The system acquires real-time location data and real-time acceleration data of fire hydrants. Based on the real-time acceleration data, it calculates the real-time vertical direction vector and real-time vibration data of the fire hydrants. It presets image capture conditions and compares the real-time location data, real-time vertical direction vector, and real-time vibration data with the location alarm threshold, attitude angle alarm threshold, and vibration alarm threshold to obtain comparison results. Based on the comparison results and image capture conditions, it triggers image capture events to obtain captured images. Image capture events include location capture events and vibration capture events.

[0009] A preset analysis time window is set. When a capture event at one location is triggered, capture events at other locations are searched within the analysis time window to obtain search results. Preset event classification conditions are set. Based on the search results, the image capture events are classified into first image capture events and second image capture events according to the event classification conditions.

[0010] The first data features of the first image capture event and the second data features of the second image capture event are extracted respectively. A multi-level fusion decision tree model is constructed. The first data features are analyzed using the multi-level fusion decision tree model to obtain multiple event types. The second data features are directly analyzed to obtain the analysis results. Event alarm conditions are preset. Based on multiple event types, analysis results and event alarm conditions, corresponding event alarms are generated.

[0011] Specifically, based on preset location data, preset attitude angle data, and preset vibration data, the location baseline, attitude angle baseline, and vibration baseline of the fire hydrant are constructed, including:

[0012] Calculate the arithmetic mean of the preset location data to obtain the location baseline;

[0013] Calculate the average value of the preset attitude angle data to obtain the attitude angle baseline;

[0014] Calculate the root mean square of the preset vibration data to obtain the vibration baseline.

[0015] Specifically, position alarm thresholds, attitude angle alarm thresholds, and vibration alarm thresholds are set based on position baselines, attitude angle baselines, and vibration baselines, including:

[0016] Calculate the standard deviations of the preset position data and preset attitude angle data to obtain the position standard deviation and attitude angle standard deviation;

[0017] Preset position safety factor, attitude angle safety factor and vibration safety factor, and calculate position alarm threshold using position safety factor and position standard deviation;

[0018] The attitude angle alarm threshold is calculated using the attitude angle safety factor and the attitude angle standard deviation.

[0019] The vibration alarm threshold is calculated using the vibration safety factor and the vibration baseline.

[0020] Specifically, preset image capture conditions are used to compare real-time position data, real-time vertical vector with position alarm thresholds, and attitude angle alarm thresholds to obtain comparison results. Based on the comparison results and image capture conditions, an image capture event is triggered to obtain a captured image, including:

[0021] The real-time position offset is calculated using real-time position data, and the real-time attitude change is calculated using the real-time vertical direction vector and the attitude angle baseline.

[0022] When the real-time position offset exceeds the position alarm threshold or the real-time attitude change exceeds the attitude angle alarm threshold, a position capture event is triggered, and a captured image of the fire hydrant is obtained.

[0023] Specifically, the method includes setting preset image capture conditions, comparing real-time vibration data with a vibration alarm threshold to obtain a comparison result, triggering an image capture event based on the comparison result and the image capture conditions to obtain a captured image, and also includes:

[0024] A vibration time window and a vibration frequency threshold are preset. When the number of consecutive times the real-time vibration data exceeds the vibration alarm threshold within a vibration time window is greater than or equal to the vibration frequency threshold, a vibration capture event is triggered, and a capture image of the fire hydrant is obtained at the same time.

[0025] Specifically, search results include both successful and unsuccessful searches.

[0026] Specifically, preset event classification conditions are used to classify image capture events into first image capture events and second image capture events based on the search results matching the event classification conditions. These include:

[0027] A preset physical delay threshold is set, and the time difference between the vibration peak and the displacement trigger is calculated when the search result indicates a successful search.

[0028] When the search result is successful and the time difference is less than or equal to the physical delay threshold, the image capture event is classified as the first image capture event.

[0029] When the search result is "search failed", the image capture event is assigned to the second image capture event.

[0030] Specifically, the first data features of the first image capture event and the second data features of the second image capture event are extracted respectively, including:

[0031] Extract the vibration intensity, vibration time, horizontal displacement angle, vibration-displacement delay, vehicle appearance value, and tool appearance value of the first image capture event;

[0032] Extract the occurrence values ​​of construction machinery, construction fence, displacement speed, and image occlusion values ​​of the second image capture event.

[0033] Specifically, a multi-level fusion decision tree model is used to analyze the first data features to obtain various event types, and the second data features are directly analyzed to obtain the analysis results, including:

[0034] Preset vibration intensity threshold, vibration time threshold, and vibration-displacement delay threshold;

[0035] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-displacement delay is less than the vibration-displacement delay threshold, the horizontal displacement angle is perpendicular to the road direction, and the vehicle occurrence value is 1, it is marked as a traffic accident collision event.

[0036] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, and the tool occurrence value is 1, it is marked as a human-caused damage event.

[0037] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, and the vehicle occurrence value is 0, and the tool occurrence value is 0, it is marked as an impact of unknown cause;

[0038] A preset displacement growth rate threshold is set. If the value of construction equipment is 1 or the value of construction fence is 1, it is marked as a construction relocation event.

[0039] If the displacement growth rate is greater than the displacement growth rate threshold, and the value of the construction equipment is 0, and the value of the construction fence is 0, it is marked as a natural loosening event.

[0040] If the image occlusion value is 1, it is marked as a special event.

[0041] Secondly, this invention provides a fire hydrant abnormal state detection system based on multi-source data fusion. The detection system is applied to the detection method of the first aspect, and the detection system includes:

[0042] The baseline establishment unit is used to acquire preset position data, preset attitude angle data and preset vibration data of the fire hydrant in a static state, construct the position baseline, attitude angle baseline and vibration baseline of the fire hydrant, and set the position alarm threshold, attitude angle alarm threshold and vibration alarm threshold based on the position baseline, attitude angle baseline and vibration baseline.

[0043] The event triggering unit is connected to the baseline establishment unit. The event triggering unit is used to acquire the real-time position data, real-time acceleration data, real-time vertical direction vector and real-time vibration data of the fire hydrant, preset image capture conditions, and compare the real-time position data, real-time vertical direction vector, real-time vibration data with the position alarm threshold, attitude angle alarm threshold and vibration alarm threshold respectively to obtain the comparison results. Based on the comparison results and the image capture conditions, the image capture event is triggered to obtain the captured image.

[0044] The event association unit is connected to the event triggering unit. The event association unit is used to preset the analysis time window. When a location capture event is triggered, it searches for other location capture events within the analysis time window to obtain the search results. It presets the event classification conditions and classifies the image capture events into the first image capture event and the second image capture event according to the event classification conditions matched with the search results.

[0045] The anomaly diagnosis unit, connected to the event association unit, is used to extract first data features and second data features, construct a multi-level fusion decision tree model, analyze the first data features using the multi-level fusion decision tree model to obtain various event types, directly analyze the second data features to obtain analysis results, preset event alarm conditions, and generate corresponding event alarms based on various event types, analysis results, and event alarm conditions.

[0046] This application provides a method and system for detecting abnormal states of fire hydrants based on multi-source data fusion. The method acquires multi-source data of the fire hydrant in a static state, establishes baselines for position, attitude angle, and vibration, and sets dynamic alarm thresholds. It then triggers image capture events based on the comparison of real-time data with these thresholds. Furthermore, within a preset time window, it correlates and analyzes vibration and displacement events, classifying them into two categories: physically related events (Category I) and independent events (Category II). Finally, using a constructed multi-level fusion decision tree model, combined with rich features extracted from the captured images, it intelligently analyzes and diagnoses the two types of events, accurately identifying various abnormal types such as traffic accident collisions, vandalism, and construction relocation, and automatically generating corresponding alarms. This achieves fully automated monitoring of abnormal states of fire hydrants, from perception and triggering to correlation and intelligent diagnosis. Attached Figure Description

[0047] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 A flowchart illustrating a method for detecting abnormal states of fire hydrants based on multi-source data fusion, provided in this application;

[0049] Figure 2This application provides a connection diagram for a fire hydrant abnormal state detection system based on multi-source data fusion. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0052] In this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0053] This application provides a method and system for detecting abnormal states of fire hydrants based on multi-source data fusion. The method first analyzes historical data of the fire hydrant in its static state to establish a baseline model of its position, orientation, and vibration, and then sets a dynamic alarm threshold accordingly. During real-time monitoring, image capture is intelligently triggered by comparing real-time sensor data with the threshold, capturing visual evidence of the abnormal moment. Subsequently, the system correlates vibration and displacement events over time, classifying abnormal events into physically related and independently occurring types. Finally, based on a multi-level fusion decision tree model and combined with semantic features extracted from image recognition, the system achieves accurate diagnosis and classification alarms for various abnormal types, such as traffic accident collisions, human-caused damage, construction relocation, and natural loosening, thus constructing a fully automated monitoring system from multi-source sensing and intelligent triggering to fusion diagnosis.

[0054] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0055] Figure 1 A flowchart illustrating a method for detecting abnormal states of fire hydrants based on multi-source data fusion, as provided in this application, is shown below. Figure 1 As shown in this embodiment, a method for detecting abnormal states of fire hydrants based on multi-source data fusion is provided. The detection method includes:

[0056] Acquire preset position data, preset attitude angle data and preset vibration data of fire hydrant in static state, construct position baseline, attitude angle baseline and vibration baseline of fire hydrant based on preset position data, preset attitude angle data and preset vibration data, and set position alarm threshold, attitude angle alarm threshold and vibration alarm threshold based on position baseline, attitude angle baseline and vibration baseline.

[0057] The system acquires real-time location data and real-time acceleration data of fire hydrants. Based on the real-time acceleration data, it calculates the real-time vertical direction vector and real-time vibration data of the fire hydrants. It presets image capture conditions and compares the real-time location data, real-time vertical direction vector, and real-time vibration data with the location alarm threshold, attitude angle alarm threshold, and vibration alarm threshold to obtain comparison results. Based on the comparison results and image capture conditions, it triggers image capture events to obtain captured images. Image capture events include location capture events and vibration capture events.

[0058] A preset analysis time window is set. When a capture event at one location is triggered, capture events at other locations are searched within the analysis time window to obtain search results. Preset event classification conditions are set. Based on the search results, the image capture events are classified into first image capture events and second image capture events according to the event classification conditions.

[0059] The first data features of the first image capture event and the second data features of the second image capture event are extracted respectively. A multi-level fusion decision tree model is constructed. The first data features are analyzed using the multi-level fusion decision tree model to obtain multiple event types. The second data features are directly analyzed to obtain the analysis results. Event alarm conditions are preset. Based on multiple event types, analysis results and event alarm conditions, corresponding event alarms are generated.

[0060] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method first establishes baselines for position, attitude, and vibration by analyzing historical data of the fire hydrant in a static state and sets dynamic alarm thresholds. Second, during real-time monitoring, image capture events are intelligently triggered to capture on-site footage by comparing real-time sensor data (position, vertical vector, vibration) with preset thresholds. Subsequently, the system analyzes the sequentially triggered capture events within a preset time window, classifying them into either physically related first-class events or independent second-class events. Finally, data features of both types of events are extracted, and a multi-level fusion decision tree model is used for intelligent analysis to accurately identify abnormal types such as impacts, damage, and construction, and automatically generate corresponding alarms, thereby achieving fully automated monitoring from multi-source sensing and intelligent triggering to fusion diagnosis.

[0061] This method overcomes the limitations of single-sensor monitoring by fusing multi-source sensor data such as position, attitude, and vibration with image information, significantly improving the comprehensiveness and reliability of anomaly detection. It innovatively introduces event correlation analysis and classification mechanisms, effectively distinguishing between instantaneous impacts (such as collisions) and gradual anomalies (such as loosening), achieving preliminary identification of the root cause of anomalies. Employing a multi-level fusion decision tree model for intelligent diagnosis, combined with semantic features of image recognition, it not only significantly improves alarm accuracy and reduces false alarms but also outputs specific and actionable anomaly event types, providing crucial decision support for rapid response and precise handling by maintenance personnel. Overall, this method constructs an efficient, intelligent, and reliable fire hydrant anomaly status monitoring system.

[0062] Specifically, based on preset location data, preset attitude angle data, and preset vibration data, the location baseline, attitude angle baseline, and vibration baseline of the fire hydrant are constructed, including:

[0063] Calculate the arithmetic mean of the preset location data to obtain the location baseline;

[0064] Calculate the average value of the preset attitude angle data to obtain the attitude angle baseline;

[0065] Calculate the root mean square of the preset vibration data to obtain the vibration baseline.

[0066] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. This method establishes a quantitative benchmark for the normal state of each fire hydrant through statistical analysis: its standard position (position baseline) is determined using the arithmetic mean of preset position data; its standard attitude (attitude angle baseline) is determined using the average of preset attitude angle data; and its background vibration level is quantified by calculating the root mean square of preset vibration data, thereby determining its inherent vibration characteristics (vibration baseline). This process provides a stable and personalized reference benchmark for subsequent abnormal detection.

[0067] The formula for calculating the location baseline is:

[0068]

[0069] in, Indicates the location baseline, This represents the average position of the fire hydrant across all coordinate axes. This represents the preset position data of each coordinate axis of the fire hydrant. Indicates the number of sampling points. This indicates the fire hydrant's number index.

[0070] The formula for calculating the attitude angle baseline is:

[0071]

[0072] in, Indicates the attitude angle baseline. Indicates the roll angle baseline. Indicates the pitch baseline. Indicates the sampling time. Indicates the first The roll angle at any given moment, Indicates the first The pitch angle at any given moment. Indicates the number of sampling points.

[0073] The formula for calculating the vibration baseline is:

[0074]

[0075] in, The vibration baseline is used to characterize the intensity of environmental vibration. Indicates the sampling time. , and They represent the first The accelerations along the three axes at each moment. Indicates the number of sampling points.

[0076] This method first establishes a baseline by calculating statistical measures such as the average and root mean square. This simple and reliable method effectively filters out random measurement noise and extracts the essential characteristics of fire hydrants in their normal, static state. Secondly, this baseline establishment method tailors a unique "normal state fingerprint" for each fire hydrant, fully considering the influence of different installation locations, surrounding environments, and individual differences. This makes subsequent alarm threshold settings more personalized and accurate, laying a solid foundation for high-precision anomaly detection.

[0077] Specifically, position alarm thresholds, attitude angle alarm thresholds, and vibration alarm thresholds are set based on position baselines, attitude angle baselines, and vibration baselines, including:

[0078] Calculate the standard deviations of the preset position data and preset attitude angle data to obtain the position standard deviation and attitude angle standard deviation;

[0079] Preset position safety factor, attitude angle safety factor and vibration safety factor, and calculate position alarm threshold using position safety factor and position standard deviation;

[0080] The attitude angle alarm threshold is calculated using the attitude angle safety factor and the attitude angle standard deviation.

[0081] The vibration alarm threshold is calculated using the vibration safety factor and the vibration baseline.

[0082] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. This method, based on an established static baseline, further sets dynamic alarm thresholds through statistical analysis and safety factors: First, it calculates the standard deviation of historical position and attitude angle data to quantify their inherent fluctuation range; second, it introduces preset position safety factors and attitude angle safety factors, multiplying them by their corresponding standard deviations to calculate the dynamic alarm thresholds for position and attitude angles. Simultaneously, it uses a preset vibration safety factor multiplied by the calculated vibration baseline value to determine the vibration alarm threshold. In this way, quantified abnormality judgment boundaries are set for three key monitoring parameters.

[0083] The formula for calculating the standard deviation of a location is:

[0084] ;

[0085] ;

[0086] ;

[0087] in, , and These represent the standard deviations of the positions of the three axes, respectively.

[0088] The formula for calculating the standard deviation of attitude angles is:

[0089]

[0090] in, Indicates the standard deviation of attitude angles. Indicates the standard deviation of the roll angle. This represents the standard deviation of the pitch angle.

[0091] The formula for calculating the location alarm threshold is:

[0092]

[0093] in, Indicates the location alarm threshold. This indicates the location safety factor, which is preset to 3-5.

[0094] The formula for calculating the attitude angle alarm threshold is:

[0095]

[0096] in, Indicates the attitude angle alarm threshold. This represents the attitude angle safety factor, preset to 5-10. This represents the minimum absolute threshold, which is preset to 2°-5°.

[0097] The formula for calculating the vibration alarm threshold is:

[0098]

[0099] in, Indicates the vibration alarm threshold. This indicates the vibration safety factor, which is preset to 5-10.

[0100] This method combines the statistical fluctuation characteristics (standard deviation) of data with the engineering safety margin (safety factor). The set alarm threshold reflects the natural fluctuation range of fire hydrants under normal conditions and provides a reasonable buffer for occasional noise or minor disturbances, effectively reducing the false alarm rate and enhancing the system's robustness. This method achieves dynamic and personalized alarm thresholds; the threshold is not a fixed value but is calculated based on the historical behavior data of each fire hydrant, allowing it to adapt to different installation environments and individual differences, improving the targeting and accuracy of anomaly detection. This method uses the safety factor as an adjustable parameter, providing maintenance personnel with an interface to flexibly adjust it according to actual management requirements and risk tolerance, giving the system good configurability and adaptability.

[0101] Specifically, preset image capture conditions are used to compare real-time position data, real-time vertical vector with position alarm thresholds, and attitude angle alarm thresholds to obtain comparison results. Based on the comparison results and image capture conditions, an image capture event is triggered to obtain a captured image, including:

[0102] The real-time position offset is calculated using real-time position data, and the real-time attitude change is calculated using the real-time vertical direction vector and the attitude angle baseline.

[0103] When the real-time position offset exceeds the position alarm threshold or the real-time attitude change exceeds the attitude angle alarm threshold, a position capture event is triggered, and a captured image of the fire hydrant is obtained.

[0104] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method defines an image capture triggering mechanism based on changes in physical state. Specifically, the system continuously calculates the real-time position offset (relative to the position baseline) and attitude change (the change in the real-time vertical vector relative to the attitude angle baseline) of the fire hydrant, and compares these two values ​​with their respective dynamic alarm thresholds in real time. Once the real-time position offset exceeds the position alarm threshold, or the real-time attitude change exceeds the attitude angle alarm threshold, the system determines that a significant position or attitude abnormality has occurred and immediately triggers a "position capture event," synchronously controlling the image acquisition device to capture an image of the current fire hydrant.

[0105] The formula for calculating the real-time position offset is:

[0106]

[0107] in, This indicates the real-time position offset. Indicates time The measured three-dimensional coordinates , , They represent The measured values ​​of the x-axis, y-axis, and z-axis of the three-dimensional coordinates.

[0108] The formula for calculating real-time attitude changes is:

[0109]

[0110] in, This represents the real-time attitude change. Indicates time The current vertical direction vector is estimated using the accelerometer.

[0111] like or If the tilt angle threshold is reached (e.g., 5°), a displacement event is triggered, and the displacement event object... Includes: trigger time Displacement vector Tilt angle change .

[0112] when When being recorded, the main control module immediately sends a trigger command to the camera via the wireless network; after receiving the command, the camera... Capture a high-resolution image at a specific moment (usually less than 1 second). The system retrieves the fire hydrant from storage. The most recent normal state image taken by a timed task before a certain time (e.g., ΔT = 1 hour) is used as the reference background image. .

[0113] This method enables automatic and real-time correlation between "numerical anomalies" and "image evidence." When sensor data detects a physical anomaly, the system can capture visual images of the scene immediately, providing intuitive and tamper-proof first-hand evidence for subsequent event diagnosis and liability determination, greatly improving the credibility and traceability of monitoring results. The method's triggering logic is clear and efficient, transforming complex physical state monitoring (position and attitude) into simple threshold comparisons. It has a fast response speed and low computational resource consumption, meeting the real-time requirements of large-scale urban fire hydrant networks. By capturing images, this method not only verifies the authenticity of sensor alarms and reduces false alarms caused by sensor false alarms, but also provides rich data sources for subsequent intelligent analysis (such as identifying surrounding vehicles, personnel, and tools), achieving effective collaboration and value enhancement of multimodal data.

[0114] Specifically, the method includes setting preset image capture conditions, comparing real-time vibration data with a vibration alarm threshold to obtain a comparison result, triggering an image capture event based on the comparison result and the image capture conditions to obtain a captured image, and also includes:

[0115] A vibration time window and a vibration frequency threshold are preset. When the number of consecutive times the real-time vibration data exceeds the vibration alarm threshold within a vibration time window is greater than or equal to the vibration frequency threshold, a vibration capture event is triggered, and a capture image of the fire hydrant is obtained at the same time.

[0116] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method defines an image capture triggering mechanism based on vibration patterns. Specifically, the system not only monitors whether real-time vibration data exceeds a preset vibration alarm threshold, but also introduces two key parameters: a "vibration time window" and a "vibration frequency threshold." The system continuously monitors within the specified time window. Only when the number of consecutive times the real-time vibration data exceeds the alarm threshold reaches or exceeds the set frequency threshold is it determined to be a valid continuous abnormal vibration, triggering a "vibration capture event" and simultaneously capturing on-site images of the fire hydrant.

[0117] The formula for calculating real-time vibration energy is as follows:

[0118]

[0119] in, Represents real-time vibrational energy. , and Representing time respectively The square of the raw reading of the triaxial accelerometer.

[0120] Set a time window (e.g., 0.1 seconds corresponds to 20 sampling points). If in Inside, there are continuous (e.g., 15) sampling points Exceed If the vibration event occurs, it is determined that a valid vibration event has occurred.

[0121] Generate vibration event objects Includes: the time of the first threshold exceeding the limit ; Time to reach maximum value Time of the last threshold exceeding the limit Peak energy .

[0122] when When being recorded, the main control module immediately sends a trigger command to the camera via the wireless network; after receiving the command, the camera... Capture a high-resolution image at a specific moment (usually less than 1 second). The system retrieves the fire hydrant from storage. The most recent normal state image taken by a timed task before a certain time (e.g., ΔT = 1 hour) is used as the reference background image. .

[0123] This method effectively filters out transient interference and significantly reduces false alarms. By requiring multiple consecutive exceedances of a threshold, it avoids transient vibration interference such as vehicles briefly passing by or single impacts, ensuring that the capture is triggered by truly noteworthy, persistent abnormal events (such as continuous damage or mechanical construction). This improves the accuracy and evidentiary value of event capture. This "multiple confirmation" mechanism makes triggered capture events more likely to correspond to on-site situations with analytical value, and the captured images are more likely to contain key processes or related objects of the event, providing more targeted material for subsequent analysis. The parameters (time window and number of thresholds) can be flexibly configured, allowing the system to adapt to the sensitivity requirements of different scenarios, balancing false alarm and false negative rates, and enhancing the system's adaptability and practicality in complex real-world environments.

[0124] Specifically, search results include both successful and unsuccessful searches.

[0125] Specifically, preset event classification conditions are used to classify image capture events into first image capture events and second image capture events based on the search results matching the event classification conditions. These include:

[0126] A preset physical delay threshold is set, and the time difference between the vibration peak and the displacement trigger is calculated when the search result indicates a successful search.

[0127] When the search result is successful and the time difference is less than or equal to the physical delay threshold, the image capture event is classified as the first image capture event.

[0128] When the search result is "search failed", the image capture event is assigned to the second image capture event.

[0129] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method defines a classification logic based on the spatiotemporal correlation of events. Specifically, when a location capture event (triggered by displacement / attitude anomalies) occurs, the system will backtrack within a preset analysis time window to search for whether a vibration capture event has occurred before. If the search is successful, the "time difference" between the peak time of the vibration event and the trigger time of the current displacement event is calculated. The system presets a "physical delay threshold." If this time difference is less than or equal to the threshold, the vibration and displacement are considered to have a strong physical causal relationship, and the associated capture event is classified as a "first image capture event." Conversely, if no associated vibration event is found within the time window (i.e., the search fails), the current location capture event is separately classified as a "second image capture event."

[0130] Set a fixed time window length (e.g., 30 seconds), when a new In time When this occurs, the system defines the associated time window as follows: .

[0131] Search the database to see if it exists. To meet its peak time Located within the associated time window, and .

[0132] Key judgment criterion: Calculate the time difference between the peak vibration and the displacement trigger. .

[0133] Physical rationality judgment: Set a maximum physical delay (e.g., 5 seconds). If Then it is considered that and There is a high probability of causal relationship between them and the corresponding... , Bind to a first image capture event .

[0134] If no matching criteria are found Then Marked as the second image capture event, it is only bound to the image.

[0135] This method achieves preliminary intelligent differentiation of the root causes of abnormal events. By analyzing the tight temporal coupling between vibration and displacement events, it can effectively distinguish between "abrupt state changes caused by instantaneous impact" (such as collisions) and "gradual state changes without obvious external impact" (such as natural loosening), laying the foundation for subsequent adoption of different diagnostic strategies. The threshold judgment based on physical laws is logically clear and highly interpretable. The setting of the "physical delay threshold" conforms to the causal time range of displacement caused by most physical impacts, making the classification results have a reliable physical basis, rather than simple data correlation. This classification mechanism greatly simplifies and optimizes the subsequent analysis process. The system can perform in-depth feature fusion analysis on the first type of events with strong physical correlations, while adopting a more direct analysis path for isolated second type events, thereby improving overall processing efficiency and diagnostic targeting.

[0136] Specifically, the first data features of the first image capture event and the second data features of the second image capture event are extracted respectively, including:

[0137] Extract the vibration intensity, vibration time, horizontal displacement angle, vibration-displacement delay, vehicle appearance value, and tool appearance value of the first image capture event;

[0138] Extract the occurrence values ​​of construction machinery, construction fence, displacement speed, and image occlusion values ​​of the second image capture event.

[0139] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method designs differentiated feature extraction strategies according to the core attributes of different types of events. For "first image capture events" classified as having a strong correlation between vibration and displacement, the system mainly extracts their dynamic features (such as vibration intensity, time, and vibration-displacement delay) and the semantic features of immediate objects in the image (such as vehicles and tools). For "second image capture events" without obvious impact correlation, the system focuses on extracting their state evolution features (such as displacement acceleration), long-term and scene-specific semantic features in the image (such as construction machinery and fencing), and image quality information (such as image occlusion).

[0140] Vibration intensity Vibration time ; Displacement horizontal direction angle ; Vibration-shift delay Displacement growth rate .

[0141] in, Indicates vibration intensity. This represents the maximum value of real-time vibration energy reached during a single vibration event. Indicates the vibration time. This indicates the moment when the vibration energy at the first sampling point exceeds the vibration alarm threshold in a single vibration event. This indicates the moment when the last sampling point in the event exceeds the threshold. Indicates the horizontal direction angle of displacement. This represents the components of the displacement vector in the horizontal plane (X-axis, Y-axis). Indicates vibration-shift delay, This indicates that the vibrational energy reaches its peak during a single vibrational event. This indicates the trigger time of the displacement event (position capture event) associated with the vibration event. Indicates the rate of increase in displacement. This represents the magnitude of the displacement vector, i.e., the total offset of the real-time position relative to the position baseline. Indicates the trigger time of the displacement event (time counted from zero or reference time).

[0142] Using the MobileNet-SSD network (existing technology), the presence values ​​of vehicles in the images of the first image capture event and the second image capture event are identified as follows: vehicle presence value is 1 when occlusion occurs, and 0 otherwise; tool presence value is 1 when occlusion occurs, and 0 otherwise; construction equipment presence value is 1 when occlusion occurs, and 0 otherwise; construction fence presence value is 1 when occlusion occurs, and 0 otherwise; and image occlusion value is 1 when occlusion occurs, and 0 otherwise.

[0143] This method achieves highly targeted feature extraction. It extracts the most relevant and discriminative feature combinations based on the physical nature of different types of events (instantaneous impact vs. gradual change) and the semantic focus of images (immediate participants vs. scene state), avoiding feature redundancy and providing an optimal data foundation for subsequent accurate diagnosis. It integrates deep information from multimodal data. This method not only extracts quantitative parameters from sensor data but also extracts key semantic values ​​(such as vehicles, tools, and construction equipment) from captured images through intelligent analysis, achieving deep fusion of numerical signals and visual semantics, greatly improving the dimensionality and accuracy of event understanding. By extracting quality features such as "image occlusion values," it enhances the system's fault tolerance and self-reflection capabilities, enabling the identification of abnormal data caused by non-fire hydrant-related issues such as camera obstruction, thereby avoiding misjudgments and improving the overall reliability of the system.

[0144] Specifically, a multi-level fusion decision tree model is used to analyze the first data features to obtain various event types, and the second data features are directly analyzed to obtain the analysis results, including:

[0145] Preset vibration intensity threshold, vibration time threshold, and vibration-displacement delay threshold;

[0146] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-displacement delay is less than the vibration-displacement delay threshold, the horizontal displacement angle is perpendicular to the road direction, and the vehicle occurrence value is 1, it is marked as a traffic accident collision event.

[0147] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, and the tool occurrence value is 1, it is marked as a human-caused damage event.

[0148] If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, and the vehicle occurrence value is 0, and the tool occurrence value is 0, it is marked as an impact of unknown cause;

[0149] A preset displacement growth rate threshold is set. If the value of construction equipment is 1 or the value of construction fence is 1, it is marked as a construction relocation event.

[0150] If the displacement growth rate is greater than the displacement growth rate threshold, and the value of the construction equipment is 0, and the value of the construction fence is 0, it is marked as a natural loosening event.

[0151] If the image occlusion value is 1, it is marked as a special event.

[0152] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. The method employs a rule-based decision tree model to intelligently analyze extracted features and classify events. For the first type of event (strong impact correlation), the model determines whether vibration characteristics (intensity, duration, delay) exceed a preset threshold, and combines this with semantic information such as the presence of vehicles or tools in the image to accurately distinguish between "traffic accident collision," "human sabotage," and "impact of unknown cause." For the second type of event (no strong impact correlation), the model determines whether the displacement change rate is abnormal, and combines this with the presence of construction equipment or barriers in the image to distinguish between "construction relocation" and "natural loosening," while also handling special cases such as "image occlusion" separately. Finally, each eligible event is labeled with a specific event type.

[0153] This method employs an "if...then..." rule format, with each judgment having clear physical meaning or factual basis (such as the presence of tools), making the final event classification results easily understandable and verifiable by operations and maintenance personnel, avoiding the trust issues associated with "black box" models. It boasts high classification accuracy and strong practicality. By fusing and judging multi-source features (vibration parameters, displacement rates, and various image semantics) in a multi-level and orderly manner, it can effectively distinguish events that may appear similar but have different underlying causes (such as sabotage and traffic accidents). The specific event type output provides a direct basis for developing differentiated response plans. Model construction and maintenance are relatively simple and efficient. Compared to complex machine learning models that require large amounts of labeled data for training, this rule-based method relies on domain knowledge, has a short development cycle, and the rules can be flexibly added, deleted, or modified based on actual operational experience and new anomaly patterns, exhibiting good scalability and maintainability.

[0154] Figure 2 A connection diagram of a fire hydrant abnormal state detection system based on multi-source data fusion provided in this application is shown below. Figure 2 As shown in this embodiment, a fire hydrant abnormal state detection system based on multi-source data fusion is provided. The detection system includes:

[0155] The baseline establishment unit is used to acquire preset position data, preset attitude angle data and preset vibration data of the fire hydrant in a static state, construct the position baseline, attitude angle baseline and vibration baseline of the fire hydrant, and set the position alarm threshold, attitude angle alarm threshold and vibration alarm threshold based on the position baseline, attitude angle baseline and vibration baseline.

[0156] The event triggering unit is connected to the baseline establishment unit. The event triggering unit is used to acquire the real-time position data, real-time acceleration data, real-time vertical direction vector and real-time vibration data of the fire hydrant, preset image capture conditions, and compare the real-time position data, real-time vertical direction vector, real-time vibration data with the position alarm threshold, attitude angle alarm threshold and vibration alarm threshold respectively to obtain the comparison results. Based on the comparison results and the image capture conditions, the image capture event is triggered to obtain the captured image.

[0157] The event association unit is connected to the event triggering unit. The event association unit is used to preset the analysis time window. When a location capture event is triggered, it searches for other location capture events within the analysis time window to obtain the search results. It presets the event classification conditions and classifies the image capture events into the first image capture event and the second image capture event according to the event classification conditions matched with the search results.

[0158] The anomaly diagnosis unit, connected to the event association unit, is used to extract first data features and second data features, construct a multi-level fusion decision tree model, analyze the first data features using the multi-level fusion decision tree model to obtain various event types, directly analyze the second data features to obtain analysis results, preset event alarm conditions, and generate corresponding event alarms based on various event types, analysis results, and event alarm conditions.

[0159] This application provides a fire hydrant abnormal state detection system based on multi-source data fusion. First, a baseline establishment unit analyzes historical static data of the fire hydrant to establish personalized benchmarks and dynamic alarm thresholds for its position, orientation, and vibration. Second, an event triggering unit intelligently triggers image capture based on the comparison of real-time data and thresholds, capturing visual evidence of the abnormal moment. Subsequently, an event correlation unit correlates and analyzes the captured events triggered sequentially in the time dimension, classifying them into physically related and independently occurring types. Finally, an anomaly diagnosis unit extracts multimodal features of various events, uses a multi-level fusion decision tree model for intelligent analysis, accurately identifies specific anomaly types, and automatically generates alarms. The entire system forms a closed-loop processing flow from "data perception - threshold judgment - event triggering - correlation classification - intelligent diagnosis".

[0160] High-precision positioning module: The GNSS / BeiDou positioning antenna and receiver supporting RTK / PPP are installed at the center of the top of the fire hydrant (such as the hydrant cover or the top of a special column) to obtain millimeter-level three-dimensional coordinates in real time.

[0161] Attitude and vibration sensors: An IMU (Inertial Measurement Unit) integrating a three-axis high-frequency accelerometer and a three-axis gyroscope, or a dedicated tilt / vibration sensor, is installed inside the fire hydrant body or the inner wall of the valve box, rigidly connected to the fire hydrant. It is used to measure the tilt angle in static conditions and to collect high-frequency vibration acceleration in dynamic conditions.

[0162] Image acquisition equipment: Low-power wide-area network smart camera (supporting infrared night vision and event triggering), installed on light poles, dedicated poles, or the side of buildings near fire hydrants, with a field of view covering the fire hydrant body and surrounding road areas. Used to capture high-definition images of the scene after receiving a trigger command.

[0163] Edge control unit: A low-power microcontroller (MCU) or embedded processing module (such as ARM Cortex-M series), installed inside the fire hydrant valve box (waterproof and dustproof enclosure). It is used to synchronously receive and preprocess all sensor data; perform baseline calculations, threshold comparisons, and event triggering logic; control camera capture; and manage wireless communication.

[0164] Communication module: NB-IoT / LTE Cat.1 / LoRaWAN wireless communication module, integrated in the same chassis as the edge control unit. Used to upload event data and captured images to the cloud / edge server and receive configuration commands.

[0165] Power module: Solar panel + lithium battery pack, or municipal power supply; solar panel is installed at a high, unobstructed location; battery pack is housed inside the valve box. Used to provide stable power to all sensors, main control unit, and communication modules.

[0166] Positioning module: Connects to the edge control unit via UART serial port or SPI interface. The positioning antenna is connected to the receiver via coaxial cable. The antenna is mounted on top of the fire hydrant, and the receiver module is placed inside the valve box and connected to the main control unit.

[0167] Attitude and vibration sensors: These typically connect to the edge control unit via an I2C or SPI digital interface. The sensors should be rigidly fixed to the fire hydrant body or the inner wall of the enclosure using screws or high-strength adhesive to ensure their attitude is perfectly synchronized with the fire hydrant.

[0168] Image acquisition device: Connected to the edge control unit via digital I / O port or UART serial port for trigger control. Image data is uploaded directly to the cloud server via independent Wi-Fi, 4G / 5G or wired Ethernet, or aggregated by the control unit before uploading.

[0169] All equipment within the enclosure (main control unit, positioning receiver, IMU, communication module) is powered by a unified power management module. Wiring should be neat and properly waterproofed and rodent-proofed. The communication module antenna should extend outside the enclosure to ensure signal quality.

[0170] High-precision time base: The edge master unit's RTC (real-time clock) is synchronized with high precision through one of the following two methods:

[0171] Method 1: GNSS time synchronization. The UTC timestamp is directly obtained from the high-precision positioning module and used as the time source for the entire system, achieving nanosecond-level accuracy.

[0172] Method 2: Network time synchronization. After the communication module connects to the network, it synchronizes with the time server via NTP or PTP protocol.

[0173] Data stamping: All sensor data (position, acceleration, image trigger) are stamped with high-precision timestamps by the main control unit using a unified time source at the moment of acquisition or triggering, ensuring the accuracy of subsequent spatiotemporal correlation analysis.

[0174] The connection between the IMU and the fire hydrant body must be rigid to avoid vibration signal attenuation or attitude measurement distortion caused by buffering. The GNSS antenna should be installed in an open location as far away as possible from tall buildings to ensure positioning accuracy and reliability. The camera installation must consider the infrared performance at night, avoid direct sunlight on the lens, and ensure that the field of view covers the direction in which the fire hydrant may be impacted (such as the side facing the road) and its overall condition. All external equipment (antenna, camera) and enclosures must meet an IP67 or higher protection rating and be able to withstand outdoor high temperature, low temperature, humidity, salt spray and other environments.

[0175] The system boasts a clear architecture and well-defined responsibilities. Four units collaborate in a pipeline manner, breaking down complex monitoring tasks into four logically distinct stages: baseline establishment, real-time triggering, event correlation, and intelligent diagnosis. This modular approach facilitates system development, deployment, and maintenance. The system automates and intelligently processes data. It autonomously completes the entire process from raw data acquisition to final event alarm, significantly reducing reliance on manual inspections and judgments, and substantially improving the operational efficiency of the city-level fire hydrant network. The system possesses strong comprehensive capabilities and clear value output. Through multi-unit collaboration, it not only achieves high-precision, low-false-alarm anomaly detection but also outputs specific and interpretable anomaly event types (such as traffic accident collisions and construction relocations), providing powerful decision support for management departments' rapid response, accurate dispatching, and responsibility identification. This truly realizes a value leap from "monitoring" to "management."

Claims

1. A method for detecting abnormal states of fire hydrants based on multi-source data fusion, characterized in that, The detection method includes: Acquire preset position data, preset attitude angle data, and preset vibration data of the fire hydrant in a static state; construct the position baseline, attitude angle baseline, and vibration baseline of the fire hydrant based on the preset position data, preset attitude angle data, and preset vibration data; and set the position alarm threshold, attitude angle alarm threshold, and vibration alarm threshold based on the position baseline, attitude angle baseline, and vibration baseline. The system acquires real-time position data and real-time acceleration data of the fire hydrant. Based on the real-time acceleration data, it calculates the real-time vertical direction vector and real-time vibration data of the fire hydrant. It presets image capture conditions and compares the real-time position data, the real-time vertical direction vector, and the real-time vibration data with the position alarm threshold, the attitude angle alarm threshold, and the vibration alarm threshold to obtain comparison results. Based on the comparison results and the image capture conditions, it triggers an image capture event to obtain a captured image. The image capture event includes a position capture event and a vibration capture event. A preset analysis time window is defined. When a location capture event is triggered, other location capture events are searched within the analysis time window to obtain search results. Preset event classification conditions are defined. Based on the search results and the event classification conditions, the image capture events are classified into a first image capture event and a second image capture event. First data features of the first image capture event and second data features of the second image capture event are extracted respectively. A multi-level fusion decision tree model is constructed. The first data features are analyzed using the multi-level fusion decision tree model to obtain multiple event types. The second data features are directly analyzed to obtain analysis results. Event alarm conditions are preset. Based on the multiple event types, the analysis results and the event alarm conditions, corresponding event alarms are generated.

2. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The construction of the fire hydrant's position baseline, attitude angle baseline, and vibration baseline based on the preset position data, the preset attitude angle data, and the preset vibration data includes: The arithmetic mean of the preset location data is calculated to obtain the location baseline; Calculate the average value of the preset attitude angle data to obtain the attitude angle baseline; The root mean square of the preset vibration data is calculated to obtain the vibration baseline.

3. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The setting of position alarm thresholds, attitude angle alarm thresholds, and vibration alarm thresholds based on the position baseline, the attitude angle baseline, and the vibration baseline includes: Calculate the standard deviation of the preset position data and the preset attitude angle data to obtain the position standard deviation and attitude angle standard deviation; The position safety factor, attitude angle safety factor and vibration safety factor are preset, and the position alarm threshold is calculated using the position safety factor and the position standard deviation; The attitude angle alarm threshold is calculated using the attitude angle safety factor and the attitude angle standard deviation; The vibration alarm threshold is calculated using the vibration safety factor and the vibration baseline.

4. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The preset image capture conditions are compared with the real-time position data, the real-time vertical direction vector, the position alarm threshold, and the attitude angle alarm threshold to obtain a comparison result. Based on the comparison result and the image capture conditions, an image capture event is triggered to obtain a captured image, including: The real-time position offset is calculated using the real-time position data, and the real-time attitude change is calculated using the real-time vertical direction vector and the attitude angle baseline. When the real-time position offset is greater than the position alarm threshold or the real-time attitude change is greater than the attitude angle alarm threshold, the position capture event is triggered, and the captured image of the fire hydrant is acquired simultaneously.

5. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The preset image capture conditions, comparing the real-time vibration data with the vibration alarm threshold to obtain a comparison result, and triggering an image capture event based on the comparison result and the image capture conditions to obtain a captured image, further include: A preset vibration time window and a vibration frequency threshold are defined. When the number of consecutive times the real-time vibration data exceeds the vibration alarm threshold within a vibration time window is greater than or equal to the vibration frequency threshold, the vibration capture event is triggered, and the captured image of the fire hydrant is acquired simultaneously.

6. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The search results include both successful and unsuccessful searches.

7. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 6, characterized in that, The preset event classification conditions, based on the search results matching the event classification conditions, divide the image capture events into first image capture events and second image capture events, including: A preset physical delay threshold is set, and when the search result indicates that the search is successful, the time difference between the vibration peak and the displacement trigger is calculated. When the search result indicates that the search was successful and the time difference is less than or equal to the physical delay threshold, the image capture event is attributed to the first image capture event; When the search result indicates that the search has failed, the image capture event is attributed to the second image capture event.

8. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 1, characterized in that, The step of extracting the first data feature of the first image capture event and the second data feature of the second image capture event respectively includes: Extract the vibration intensity, vibration time, horizontal displacement angle, vibration-displacement delay, vehicle appearance value, and tool appearance value of the first image capture event; Extract the occurrence values ​​of construction machinery, construction fence, displacement speed, and image occlusion value of the second image capture event.

9. The method for detecting abnormal states of fire hydrants based on multi-source data fusion according to claim 8, characterized in that, The process of analyzing the first data features using the multi-level fusion decision tree model to obtain multiple event types, and directly analyzing the second data features to obtain analysis results, includes: Preset vibration intensity threshold, vibration time threshold, and vibration-displacement delay threshold; If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-displacement delay is less than the vibration-displacement delay threshold, the horizontal displacement angle is perpendicular to the road direction, and the vehicle occurrence value is 1, it is marked as a traffic accident collision event. If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, and the tool occurrence value is 1, it is marked as a human-caused sabotage event; If the vibration intensity is greater than the vibration intensity threshold, the vibration time is less than the vibration time threshold, the vibration-transfer delay is less than the vibration-transfer delay threshold, the vehicle occurrence value is 0, and the tool occurrence value is 0, it is marked as an impact of unknown cause; A preset displacement growth rate threshold is set. If the value of the construction equipment is 1 or the value of the construction fence is 1, it is marked as a construction relocation event. If the displacement growth rate is greater than the displacement growth rate threshold, and the occurrence value of the construction equipment is 0, and the occurrence value of the construction fence is 0, it is marked as a natural loosening event; If the image occlusion value is 1, it is marked as a special event.

10. A fire hydrant abnormal state detection system based on multi-source data fusion, characterized in that, The detection system is applied to the detection method according to any one of claims 1 to 9, and the detection system comprises: A baseline establishment unit is used to acquire the preset position data, preset attitude angle data and preset vibration data of the fire hydrant in a static state, construct the position baseline, attitude angle baseline and vibration baseline of the fire hydrant, and set the position alarm threshold, attitude angle alarm threshold and vibration alarm threshold based on the position baseline, attitude angle baseline and vibration baseline. An event triggering unit, connected to the baseline establishment unit, is used to acquire the real-time position data, real-time acceleration data, real-time vertical direction vector, and real-time vibration data of the fire hydrant; preset the image capture conditions; compare the real-time position data, real-time vertical direction vector, and real-time vibration data with the position alarm threshold, the attitude angle alarm threshold, and the vibration alarm threshold to obtain the comparison result; and trigger the image capture event based on the comparison result and the image capture conditions to obtain the captured image. An event association unit is connected to the event triggering unit. The event association unit is used to preset the analysis time window. When a location capture event is triggered, other location capture events are searched within the analysis time window to obtain the search results. The event classification conditions are preset. The image capture event is divided into the first image capture event and the second image capture event according to the search results and the event classification conditions. An anomaly diagnosis unit, connected to the event association unit, is used to extract the first data features and the second data features, construct the multi-level fusion decision tree model, analyze the first data features using the multi-level fusion decision tree model to obtain multiple event types, directly analyze the second data features to obtain the analysis results, preset the event alarm conditions, and generate corresponding event alarms based on the multiple event types, the analysis results, and the event alarm conditions.

Images