Intelligent identification method for illegal opening behavior of box body

Abstract

The invention provides an intelligent identification method for an illegal opening behavior of a box body, and relates to the technical field of data processing, the method comprises the following steps: monitoring physical state data of the box body in real time, the physical state data comprising a vibration signal, an opening and closing state signal and position information; performing feature extraction through the physical state data to obtain an initial behavior feature sequence; based on the behavior characteristic sequence, three specific structure points on the surface of the box body are dynamically selected as a group of reference positions, and the specific structure points comprise a box body lock installation point, a box body hinge connection point and a box body seam monitoring point; according to the selected three specific structure points, constructing a dynamically changing area range; performing partition analysis on the dynamically changed region range to obtain a plurality of sub-regions; and obtaining a behavior characteristic correction coefficient according to the dynamic change characteristic of each sub-region. The illegal opening behavior recognition precision is improved, and the false alarm rate and the missing report rate are reduced.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent method for identifying illegal opening of a cabinet. Background Technology

[0002] In fields such as industrial production, public safety, and energy supply, enclosure-type equipment plays a vital role in protecting core equipment, storing critical materials, and ensuring stable operation. Its security is directly related to the social order and public interests. As society becomes increasingly reliant on infrastructure, the application scenarios of enclosure-type equipment are constantly expanding, from indoor computer rooms to unattended areas such as outdoor roadsides and remote base stations, which increases the risk of unauthorized opening of enclosures.

[0003] Traditional enclosure security mainly relies on two methods: physical protection and manual inspection. In terms of physical protection, ordinary mechanical locks and reinforced enclosure materials are often used. However, these methods are easily damaged by professional tools and cannot detect opening behavior in real time. In terms of manual inspection, maintenance personnel need to go to the site for inspection at fixed intervals. This not only has the problems of long inspection intervals and slow response, but also requires a lot of manpower. Especially in remote areas or scenarios with a large number of enclosures, the inspection efficiency is extremely low.

[0004] With the development of sensor technology and artificial intelligence, some enclosures have begun to adopt simple status monitoring solutions, such as using vibration sensors to detect abnormal vibrations and magnetic sensors to monitor opening and closing status. However, existing technologies generally have the following shortcomings: First, the data dimension is limited, relying on only one or two physical signals, making it difficult to distinguish between normal operation and illegal opening, resulting in high false alarm and false negative rates; second, they lack dynamic adaptability, failing to adjust monitoring parameters according to the enclosure's usage environment and structural deformation, leading to significant differences in recognition accuracy in different scenarios; and third, they lack a closed-loop response mechanism, making it difficult to quickly generate alarm information and retain key evidence even if an anomaly is detected. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent method for recognizing illegal opening behavior of a box, which improves the accuracy of illegal opening behavior recognition and reduces the false alarm rate and false alarm rate.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] Firstly, a method for intelligently recognizing unauthorized opening of a cabinet, the method comprising:

[0008] Real-time monitoring of the enclosure's physical state data, including vibration signals, opening and closing status signals, and position information; feature extraction is performed on the physical state data to obtain an initial behavioral feature sequence;

[0009] Based on the behavioral feature sequence, three specific structural points on the surface of the enclosure are dynamically selected as a set of reference positions. These specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. A dynamically changing area is constructed based on the three selected specific structural points. The dynamically changing area is divided into several sub-regions. A behavioral feature correction coefficient is obtained based on the dynamic change characteristics of each sub-region.

[0010] Based on the behavioral feature correction coefficient, the initial behavioral feature sequence is calibrated to obtain the optimized behavioral feature sequence.

[0011] From the obtained optimized behavior feature sequence, the discrete vibration points and position offset points of the box are extracted; the convex hull boundary covering the discrete vibration points and position offset points is constructed and the area and perimeter parameters are calculated; the area and perimeter parameters are added to the optimized behavior feature sequence to form an adjusted behavior feature sequence, which is then input into a preset behavior classification model to determine whether the opening behavior is illegal.

[0012] When the current opening behavior of the enclosure is determined to be illegal, an alarm signal is generated and relevant information about the illegal opening behavior is recorded.

[0013] Furthermore, the physical state data of the enclosure is monitored in real time, including vibration signals, opening and closing status signals, and position information. Feature extraction is performed on the physical state data to obtain an initial behavioral feature sequence, including:

[0014] Vibration sensors, opening / closing status sensors, and positioning devices deployed on the enclosure are used to collect vibration signals, opening / closing status signals, and position information in real time.

[0015] Time-domain and frequency-domain analyses are performed on vibration signals to extract vibration characteristics including amplitude, frequency components, and duration.

[0016] Perform state transition detection on the opening and closing state signal, and extract the timestamps and state change sequences of the opening and closing events;

[0017] Differential calculations are performed on the location information to extract the position offset and motion trajectory features;

[0018] The extracted vibration features, opening and closing event features, and positional motion features are fused and sequenced to obtain the initial behavioral feature sequence.

[0019] Furthermore, based on the behavioral feature sequence, three specific structural points on the surface of the enclosure are dynamically selected as a set of reference positions. These specific structural points include the enclosure lock mounting point, the enclosure hinge connection point, and the enclosure seam monitoring point. A dynamically changing region is constructed based on these three selected specific structural points. This dynamically changing region is then divided into several sub-regions. Based on the dynamic characteristics of each sub-region, a behavioral feature correction coefficient is obtained, including:

[0020] Based on the initial behavioral characteristic sequence, the intensity distribution and positional shift trend of the vibration signal are analyzed to dynamically determine the specific spatial coordinates of the enclosure lock installation point, enclosure hinge connection point, and enclosure seam monitoring point within the current monitoring cycle.

[0021] Based on the spatial coordinates of three specific structural points, the minimum enclosing region formed by the three specific structural points is calculated, and the dynamically changing region range is constructed by expanding the preset tolerance threshold.

[0022] The dynamically changing area is divided into several sub-regions by spatial grid, and the frequency of vibration events and the magnitude of positional shifts within each sub-region are analyzed based on behavioral feature sequences.

[0023] Based on the dynamic variation characteristics of the vibration event frequency and position offset amplitude in each sub-region, a dynamic weighting coefficient is assigned to each sub-region.

[0024] By combining the dynamic weight coefficients and spatial distribution relationships of all sub-regions, the behavioral feature correction coefficients are calculated.

[0025] Furthermore, based on the behavioral feature correction coefficients, the obtained initial behavioral feature sequence is calibrated to obtain an optimized behavioral feature sequence, including:

[0026] Based on the obtained behavioral feature correction coefficients, the initial behavioral feature sequence is adaptively weighted and calibrated to obtain the weighted and adjusted behavioral feature sequence.

[0027] Based on the above weighted and adjusted behavioral feature sequence, a time window-based moving average filtering method is used for smoothing to suppress random noise introduced during data acquisition and generate a smoothed feature sequence.

[0028] Based on the smoothed feature sequence described above, normalization calculations are performed to unify the dimensions and numerical ranges of each feature dimension, thereby generating the optimized behavioral feature sequence.

[0029] Furthermore, discrete vibration points and position offset points of the box are extracted from the obtained optimized behavioral feature sequence; by constructing a convex hull boundary covering the discrete vibration points and position offset points and calculating the area and perimeter parameters, including:

[0030] Based on the optimized behavioral feature sequence, data points exceeding the preset vibration amplitude threshold and position offset threshold are selected and used as discrete vibration points and position offset points, respectively.

[0031] Based on the spatial coordinates of all extracted discrete vibration points and position offset points, the outer points are sequentially connected after polar angle sorting to obtain the minimum convex polygon boundary that completely contains all points.

[0032] Based on the coordinate sequence of each vertex of the convex polygon boundary, the area parameter is obtained by the standard mathematical method for calculating the polygon area, and the perimeter parameter is obtained by accumulating the lengths of each boundary line segment.

[0033] Furthermore, the area and perimeter parameters are added to the optimized behavior feature sequence to form an adjusted behavior feature sequence, which is then input into a preset behavior classification model to determine whether the opening behavior is illegal, including:

[0034] The area parameter and perimeter parameter are added as new feature dimensions and fused into the optimized behavioral feature sequence to obtain an adjusted behavioral feature sequence that includes spatial distribution features.

[0035] Based on the adjusted behavioral feature sequence, a pre-defined machine learning-based behavioral classification model is input for pattern recognition analysis to obtain the output results of the behavioral classification model.

[0036] Based on the output of the behavior classification model, the probability assessment value of the current opening behavior being an illegal opening behavior is obtained;

[0037] Based on the comparison between the probability assessment value and the preset threshold parameter, it is determined whether the current opening behavior is an illegal opening behavior.

[0038] Furthermore, when the current opening behavior of the enclosure is determined to be illegal, an alarm signal is generated, and relevant information about the illegal opening behavior is recorded, including:

[0039] Based on the illegal opening determination result output by the behavior classification model, a corresponding alarm signal is generated and an emergency response protocol is activated.

[0040] Based on the optimized behavioral feature sequence and convex hull geometric parameter data, the timestamp of the illegal opening behavior and the current position information of the box are extracted;

[0041] Based on the timestamp of the occurrence, the current location information of the enclosure, and the geometric parameters of the convex hull, a complete record of the illegal opening behavior is generated and stored in the security event database.

[0042] Based on the alarm signal, a security alarm notification containing detailed information about the unauthorized opening behavior is sent to the monitoring center and relevant responsible personnel through a preset communication interface;

[0043] Based on the record of unauthorized opening, update the security status assessment of the enclosure and trigger the corresponding equipment security protection mechanism.

[0044] Secondly, an intelligent recognition system for illegal opening of a cabinet includes:

[0045] The acquisition module is used to monitor the physical state data of the enclosure in real time. The physical state data includes vibration signals, opening and closing status signals, and position information. Feature extraction is performed on the physical state data to obtain an initial behavioral feature sequence.

[0046] The analysis module is used to dynamically select three specific structural points on the surface of the enclosure as a set of reference positions based on the behavioral feature sequence. The specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. Based on the selected three specific structural points, a dynamically changing area is constructed. The dynamically changing area is divided into several sub-regions. Based on the dynamic change characteristics of each sub-region, a behavioral feature correction coefficient is obtained.

[0047] The calculation module is used to calibrate the initial behavior feature sequence based on the behavior feature correction coefficient to obtain an optimized behavior feature sequence; extract the discrete vibration points and position offset points of the box from the optimized behavior feature sequence; construct a convex hull boundary covering the discrete vibration points and position offset points and calculate the area and perimeter parameters; supplement the area and perimeter parameters into the optimized behavior feature sequence to form an adjusted behavior feature sequence, and input it into a preset behavior classification model to determine whether the opening behavior is illegal opening;

[0048] The processing module is used to generate an alarm signal and record relevant information about the illegal opening behavior when it is determined that the current opening behavior of the cabinet is illegal.

[0049] Thirdly, a computing device includes:

[0050] One or more processors;

[0051] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0052] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0053] The above-described solution of the present invention has at least the following beneficial effects:

[0054] This method employs multi-sensor fusion to collect multi-dimensional physical state data on vibration, opening / closing, and position, constructing an initial behavioral feature sequence. It then dynamically selects specific structural points, such as the lock installation point, to construct a region and calculates behavioral feature correction coefficients to calibrate the initial feature sequence. Simultaneously, it extracts discrete vibration points and position offset points to construct a convex hull to supplement spatial features, which are then input into a machine learning classification model for judgment. This forms a closed-loop mechanism for alarm, recording, and emergency response. Therefore, it overcomes the technical problems of existing technologies, such as the difficulty in distinguishing between normal operation and illegal opening due to a single data dimension, the lack of dynamic adaptability leading to large differences in recognition accuracy across different scenarios, and the inability to quickly alarm and retain evidence due to the lack of a closed-loop response mechanism. Ultimately, it achieves the technical effects of improving the accuracy of illegal opening behavior recognition, reducing false alarm and false negative rates, enhancing the method's adaptability to different cabinet types and usage scenarios, enabling real-time monitoring and rapid response to illegal behavior, and ensuring event traceability. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating an intelligent identification method for illegal opening of a cabinet, provided by an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of an intelligent recognition system for illegal opening of a cabinet, provided by an embodiment of the present invention. Detailed Implementation

[0057] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0058] like Figure 1 As shown, an embodiment of the present invention proposes an intelligent identification method for illegal opening of a cabinet, the method comprising the following steps:

[0059] Step 1: Monitor the physical state data of the enclosure in real time. The physical state data includes vibration signals, opening and closing status signals, and position information. Extract features from the physical state data to obtain an initial behavioral feature sequence.

[0060] Step 2: Based on the behavioral feature sequence, dynamically select three specific structural points on the surface of the enclosure as a set of reference positions. The specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. Based on the selected three specific structural points, construct a dynamically changing area. Perform a partition analysis on the dynamically changing area to obtain several sub-regions. Based on the dynamic change characteristics of each sub-region, obtain a behavioral feature correction coefficient.

[0061] Step 3: Based on the behavioral feature correction coefficient, calibrate the obtained initial behavioral feature sequence to obtain the optimized behavioral feature sequence;

[0062] Step 4: Extract the discrete vibration points and position offset points of the box from the obtained optimized behavior feature sequence; construct a convex hull boundary covering the discrete vibration points and position offset points and calculate the area and perimeter parameters; supplement the area and perimeter parameters into the optimized behavior feature sequence to form an adjusted behavior feature sequence, and input it into the preset behavior classification model to determine whether the opening behavior belongs to illegal opening;

[0063] Step 5: When the current opening behavior of the cabinet is determined to be illegal opening, an alarm signal is generated and the relevant information of the illegal opening behavior is recorded.

[0064] In this embodiment of the invention, the technical means of real-time monitoring of multi-dimensional physical state data and extraction of initial behavioral feature sequences, dynamic construction of regional range based on specific structural points such as cabinet lock installation points and calculation of behavioral feature correction coefficients to calibrate initial features, extraction of discrete vibration points and position offset points to construct convex hulls to supplement spatial features and inputting them into a preset behavioral classification model for judgment, and generation of alarm signals and recording of relevant information when illegal opening occurs, overcome the technical problems of existing cabinet security technologies, such as the difficulty in distinguishing normal operation from illegal opening due to single data dimensions, lack of dynamic adaptability leading to large differences in recognition accuracy in different scenarios, and the inability to quickly alarm and retain evidence due to the lack of a closed-loop response mechanism. Thus, it achieves the technical effect of distinguishing between normal and illegal opening of the cabinet to reduce false alarm and false alarm rates, improving the adaptability of the method to different cabinet structures and usage environments, realizing real-time monitoring, rapid response and traceable management of illegal opening behavior, and effectively ensuring the safety of cabinet equipment.

[0065] In a preferred embodiment of the present invention, step 1 above may include:

[0066] Step 1.1 involves real-time acquisition of vibration signals, opening / closing status signals, and location information using vibration sensors, opening / closing status sensors, and positioning devices deployed on the enclosure. Specifically, this includes determining the deployment locations of the sensors and positioning devices based on the enclosure type and usage scenario. For example, vibration sensors are installed near the locks of the power distribution cabinet and on the sides of the cabinet doors; opening / closing status sensors are installed at the joint between the cabinet door and the enclosure; and positioning devices are installed on the top or sides of the enclosure. The vibration sensors selected are models capable of acquiring signals in both the time and frequency domains. The opening / closing status sensors are types that can accurately detect changes in the door's opening / closing status. The positioning devices are selected from... The equipment supports real-time positioning and is adaptable to complex outdoor environments. After deployment, the data acquisition frequency of each device is set. Typically, the acquisition frequency of vibration sensors and opening / closing status sensors is set to 1 to 10 times per second. The acquisition frequency of positioning devices is adjusted according to whether the cabinet is easy to move. The positioning acquisition frequency of unattended outdoor cabinets can be set to 1 to 5 times per minute. Then, all devices are started. Vibration sensors collect vibration signals generated by external force impacts, lock prying, etc., in real time. Opening / closing status sensors collect status signals of cabinet doors from closing to opening or from opening to closing in real time. Positioning devices collect the geographical location information of the cabinet in real time.

[0067] Step 1.2 involves performing time-domain and frequency-domain analysis on the vibration signal to extract vibration features including amplitude, frequency components, and duration. Specifically, after the data processing terminal receives the vibration signal transmitted by the vibration sensor, it first preprocesses the original vibration signal to remove invalid data caused by sensor noise or environmental interference. Then, it performs time-domain analysis to determine the start and end times of each vibration event and calculate the duration of the vibration event by analyzing the changes in the vibration signal over time. Simultaneously, it extracts key indicators such as the maximum amplitude and average amplitude of the vibration signal in the time domain, which reflect the intensity of the vibration. Next, it performs frequency-domain analysis by using signal processing methods to convert the time-domain vibration signal into a frequency-domain signal, analyzing the proportion of different frequency components in the signal, and identifying characteristic frequencies. For example, vibrations caused by lock picking usually have specific high-frequency components, while vibrations caused by wind are mostly low-frequency components. By distinguishing different frequency components, the duration, amplitude, and frequency components are finally integrated to form vibration features.

[0068] Step 1.3 involves detecting state transitions in the opening / closing status signal and extracting the timestamps and state change sequences of the opening / closing events. Specifically, after receiving the opening / closing status signal from the sensor, a state monitoring mechanism is established to continuously track signal changes. When the opening / closing status signal changes from a value representing closure to a value representing open, or vice versa, an opening / closing event is determined, and the specific time of each state transition is recorded, forming a timestamp for the event. Simultaneously, the state changes of all opening / closing events within a given time period are organized chronologically. For example, within one hour, the cabinet door sequentially undergoes a process of closing, opening, closing, opening, and closing again. This series of state changes is sorted by occurrence time to form a state change sequence of the opening / closing events.

[0069] Step 1.4 involves differential calculation of the location information to extract the position offset and motion trajectory features. Specifically, after receiving the location information transmitted by the positioning device, the validity of each collected location information is verified to exclude invalid location data with large positioning deviations caused by signal obstruction or interference. Then, valid location information from two adjacent acquisition cycles is selected for differential calculation. The position coordinates of the previous acquisition cycle are subtracted from the position coordinates of the later acquisition cycle to obtain the position offset of the box within the two cycles. The position offset can be used to determine whether the box has moved and the distance it has moved. At the same time, the coordinates corresponding to all valid location information within a certain period are connected sequentially according to time to form the motion trajectory features of the box. If the box has not moved, the motion trajectory is a fixed point or a very small range; if the box has been moved, the motion trajectory will show a line. The motion trajectory features can be used to further understand the movement path of the box and help determine whether the box has been illegally moved.

[0070] Step 1.5 involves multimodal fusion and serialization alignment of the extracted vibration features, opening / closing event features, and positional motion features to obtain an initial behavioral feature sequence. Specifically, this includes: first, determining a unified timeline, using the timestamps of data collected by each sensor and positioning device as a reference, aligning the vibration features, opening / closing event features, and positional motion features in chronological order to ensure that features corresponding to the same time point can match each other. For example, associating the vibration amplitude at a certain moment, the opening / closing state near that moment, and the positional offset at that moment. Then, using a multimodal fusion method, the aligned three types of features are integrated into the same data framework. During the fusion process, key information of each type of feature is retained, while duplicate or redundant data is removed. Finally, the fused feature data is sorted in chronological order to form the initial behavioral feature sequence.

[0071] In this embodiment of the invention, by deploying vibration sensors, opening / closing state sensors, and positioning devices in the enclosure to collect multiple types of physical signals in real time, performing time-domain and frequency-domain analysis on the vibration signals to extract multi-dimensional vibration features, performing jump detection on the opening / closing state signals to extract event timestamps and state sequences, and performing differential calculation on the position information to extract offset and trajectory features, and performing multimodal fusion and serialization alignment of the three types of features, the invention overcomes the difficulty in completely depicting the state of the enclosure due to the single data acquisition dimension and fragmented feature extraction in the prior art. Thus, it achieves the comprehensive capture of the three key state information of the enclosure: vibration, opening / closing, and position, forming an initial behavioral feature sequence with consistent time dimension and complete feature dimension.

[0072] In a preferred embodiment of the present invention, step 2 above may include:

[0073] Step 2.1: Based on the initial behavioral feature sequence, analyze the intensity distribution and positional shift trend of the vibration signal to dynamically determine the specific spatial coordinates of the lock mounting point, hinge connection point, and seam monitoring point within the current monitoring cycle. Specifically, this includes: extracting vibration signal-related feature data from the obtained initial behavioral feature sequence; analyzing the distribution of vibration signal intensity on the box surface; typically, areas with relatively concentrated vibration intensity appear near the lock mounting point due to frequent operation, thus initially locating the lock mounting point's range; simultaneously extracting positional shift-related features; analyzing the positional shift trend of various parts of the box; the hinge connection point exhibits a regular shift direction and amplitude during box opening; and the seam monitoring point shows specific shift changes with box opening and closing. Combining these two features further narrows down the range of the three specific structural points. Within each monitoring cycle, based on the latest vibration intensity distribution and positional shift trend, accurately determine the specific spatial coordinates of the lock mounting point, hinge connection point, and seam monitoring point on the box surface to ensure that the coordinates reflect the actual state of the current box structure.

[0074] Step 2.2: Based on the spatial coordinates of the three specific structural points, calculate the minimum enclosing area formed by the three specific structural points, and construct a dynamically changing area range by expanding a preset tolerance threshold. Specifically, this includes: using geometric calculation methods to find the minimum area that can completely contain the three specific structural points based on their spatial coordinates. The area is usually triangular. A preset tolerance threshold is set, taking into account the size of the enclosure, the usage environment, and the range affected by illegal opening behavior in historical data. This threshold will be adjusted according to whether the enclosure is outdoors or whether it is prone to deformation. Based on the minimum enclosing area, the preset tolerance threshold is uniformly expanded in all directions outside the area to form a dynamically changing area range.

[0075] Step 2.3 involves dividing the dynamically changing area into several sub-regions using a spatial grid. Based on the behavioral feature sequence, the frequency and magnitude of vibration events within each sub-region are analyzed. Specifically, this includes dividing the dynamically changing area into multiple spatial grids of equal size at fixed spatial intervals, with each grid representing a sub-region. Based on the initial behavioral feature sequence, the number of vibration events occurring in each sub-region within a set time period is counted, and the frequency of vibration events in each sub-region is calculated. Simultaneously, the changes in position coordinates within each sub-region are analyzed, and the maximum and average magnitude of the positional shift are calculated.

[0076] Step 2.4: Based on the dynamic change characteristics of the vibration event frequency and position offset amplitude of each sub-region, assign a dynamic weight coefficient to each sub-region. Specifically, this includes: analyzing the dynamic change characteristics of each sub-region based on the obtained vibration event frequency and position offset amplitude. For sub-regions with high vibration event frequency and large position offset amplitude, it indicates that their changes are significant during the opening of the enclosure and that they are strongly correlated with illegal opening behavior, so a higher dynamic weight coefficient is assigned to them. For sub-regions with low vibration event frequency and small position offset amplitude, their changes are not obvious and their role in judging illegal opening behavior is weak, so a lower dynamic weight coefficient is assigned to them. At the same time, the distance between the sub-region and the three specific structural points is considered. Sub-regions that are closer to the point of contact usually have a higher weight coefficient. This completes the assignment of dynamic weight coefficients for all sub-regions.

[0077] Step 2.5: Based on the dynamic weight coefficients and spatial distribution relationships of all sub-regions, calculate the behavior feature correction coefficient. This includes: collecting the dynamic weight coefficients of all sub-regions; analyzing the spatial distribution relationships of each sub-region within the dynamically changing region, including the relative positions between sub-regions and the weight differences between adjacent sub-regions; adjusting the dynamic weight coefficients of each sub-region appropriately based on the spatial distribution characteristics of the sub-regions. For example, for multiple adjacent sub-regions with relatively high weight coefficients, appropriately increasing the overall weight ratio by comprehensively considering their overall impact; and finally, by comprehensively calculating the adjusted dynamic weight coefficients of all sub-regions, obtaining a behavior feature correction coefficient that reflects the importance of different regions in identifying illegal opening behavior.

[0078] In this embodiment of the invention, the spatial coordinates of three specific structural points, such as the lock installation point of the enclosure, are dynamically determined based on the analysis of vibration intensity distribution and positional offset trends of the initial behavioral feature sequence. A minimum enclosing region is constructed based on these three points, and a preset tolerance threshold is expanded to form a dynamically changing region. This region is then divided into sub-regions according to a spatial grid, and the frequency of vibration events and the amplitude of positional offset within each sub-region are analyzed. Dynamic weight coefficients are assigned based on the dynamic change characteristics of the sub-regions, and behavioral feature correction coefficients are calculated by comprehensively considering the weight coefficients and spatial distribution relationships. This technique overcomes the problems of existing technologies lacking dynamic adaptability, being unable to adjust monitoring parameters according to the enclosure's usage environment and structural deformation, and exhibiting large differences in recognition accuracy across different scenarios. It achieves adaptive adjustment of the monitoring area and feature weights, improving the recognition accuracy of illegal enclosure opening behavior in different scenarios.

[0079] In a preferred embodiment of the present invention, step 3 above may include:

[0080] Step 3.1: Based on the obtained behavior feature correction coefficients, adaptive weighted calibration is performed on the initial behavior feature sequence to obtain the weighted behavior feature sequence. Specifically, this includes: first, obtaining the calculated behavior feature correction coefficients and clarifying the feature types corresponding to different weights in the behavior feature correction coefficients. For example, features related to key areas such as lock mounting points and hinge connection points correspond to higher weights, while features related to non-key areas such as the edge of the box correspond to lower weights. Next, each feature data in the initial behavior feature sequence is matched with its corresponding weight in the behavior feature correction coefficients. For features related to key areas, such as vibration amplitude near the lock and positional offset at the hinge, their values ​​are amplified according to higher weights, making these features, which are more valuable for identifying illegal opening behavior, more prominent in the sequence. For features related to non-key areas, such as slight vibration data on the side of the box, their values ​​are reduced according to lower weights to reduce the interference of irrelevant features on the identification. Through adaptive adjustment, the weighted calibration of the initial behavior feature sequence is completed, and the weighted behavior feature sequence is finally obtained.

[0081] Step 3.2: Based on the weighted and adjusted behavioral feature sequence described above, a time window-based moving average filtering method is used for smoothing to suppress random noise introduced during data acquisition and generate a smoothed feature sequence. Specifically, this involves determining the length of the time window based on the environment in which the enclosure is used. If the enclosure is located outdoors in a windy environment with frequent vehicle traffic and high environmental noise, the time window length can be set to a longer period, such as 10 to 15 seconds. If the enclosure is located indoors in a quiet environment with less noise, the time window length can be set to 5 to 8 seconds. Then, using the set time window as a unit, the time window is slid sequentially along the time axis. After each slide, the average value of all data in the weighted and adjusted behavioral feature sequence within the current time window is calculated, and the average value is used to replace the original feature data at the corresponding moment within the time window. In this way, the entire weighted and adjusted behavioral feature sequence is gradually smoothed, eliminating interference from random noise caused by sensor instantaneous errors, sudden outdoor wind vibrations, and conducted vibrations caused by vehicles passing by during data acquisition. This makes the trend of the feature sequence more stable, ultimately generating a smoothed feature sequence.

[0082] Step 3.3: Based on the smoothed feature sequence described above, normalization calculations are performed to unify the dimensions and numerical ranges of each feature dimension, generating the optimized behavioral feature sequence. Specifically, this includes: first, performing statistical analysis on each feature dimension in the smoothed feature sequence; for features like vibration amplitude, calculating its maximum and minimum values ​​throughout the entire sequence; for features like position offset, similarly calculating their corresponding maximum and minimum values; if some feature data are scattered, the mean and standard deviation of that feature dimension can also be calculated. Then, based on the statistically obtained values, the mean and standard deviation of each feature dimension are calculated. Data is converted to a uniform numerical range, commonly 0 to 1. For vibration amplitude features, the minimum value of the feature is subtracted from each data point under that feature, and then divided by the difference between the maximum and minimum values ​​to obtain the converted value. For position offset features, the same calculation logic is used for conversion. Through the above normalization calculation, the differences caused by different units such as vibration amplitude in millimeters and position offset in meters are eliminated. At the same time, the numerical range of different feature dimensions is kept consistent to avoid affecting the recognition and judgment due to a feature value that is too large or too small. Finally, an optimized behavioral feature sequence is generated.

[0083] In this embodiment of the invention, the initial behavioral feature sequence is adaptively weighted and calibrated based on the behavioral feature correction coefficient. The weighted sequence is then smoothed using a time window moving average filtering method to suppress random noise. Furthermore, the dimensions and numerical ranges of each feature dimension are unified through normalization calculation. This overcomes the technical problems in the prior art, such as the feature sequence not distinguishing the importance of key areas, being greatly affected by environmental noise, and the inconsistent dimensions and ranges of each feature dimension leading to low recognition accuracy. As a result, the invention strengthens the weight of key features, reduces the impact of noise, and unifies feature standards, making the optimized behavioral feature sequence more accurately reflect the true state of the box.

[0084] In a preferred embodiment of the present invention, step 4 above may include:

[0085] Step 4.1: Based on the optimized behavioral feature sequence, data points exceeding the preset vibration amplitude threshold and position offset threshold are selected and used as discrete vibration points and position offset points, respectively. Specifically, this involves first determining the preset vibration amplitude threshold and position offset threshold by considering the material characteristics of the enclosure, the usage scenario, and historical anomaly data. For example, for outdoor metal power distribution cabinets, which are easily affected by wind and minor impacts, the vibration amplitude threshold can be set to a slightly higher value to avoid misjudgment; for indoor plastic communication equipment boxes, which experience less environmental interference, the vibration amplitude threshold can be set to a lower value. The system sets a threshold value to ensure that minor abnormal vibrations can be captured. The position offset threshold is adjusted according to whether the enclosure is fixedly installed. For fixed enclosures, the threshold is set to a minimum value to prevent missed detection when the enclosure is slightly moved. For movable enclosures, the threshold is set according to the normal movement range. Then, the optimized behavior feature sequence is traversed, and the vibration amplitude of each data point in the sequence is compared with the preset vibration amplitude threshold. Data points with vibration amplitudes exceeding the threshold are marked as discrete vibration points. At the same time, the position offset of each data point is compared with the preset position offset threshold. Data points with position offsets exceeding the threshold are marked as position offset points.

[0086] Step 4.2: Based on the spatial coordinates of all extracted discrete vibration points and position offset points, the boundary of the smallest convex polygon that completely contains all points is obtained by sorting the points by polar angle and then connecting the outer points in sequence. Specifically, this involves: first, collecting the spatial coordinates of all selected discrete vibration points and position offset points. These coordinates must accurately reflect the specific position of each point on the surface of the box. Then, selecting the point with the smallest coordinate value as the reference point, and using this reference point as the origin, calculating the polar angles of all other points relative to the reference point. The calculation of the polar angles needs to consider the horizontal and vertical distances between each point and the reference point. Then, sorting all points in ascending order of polar angles. If points with the same polar angle appear during the sorting process, the points that are farther away from the reference point are retained, and the points that are closer are removed. Finally, starting from the first sorted point, connecting adjacent points in sequence. When connecting to the last point, connecting the last point to the first point forms a polygon. By checking whether this polygon can completely contain all discrete vibration points and position offset points and whether there are any concave parts, it is ensured that the boundary of the smallest convex polygon that completely contains all points is obtained.

[0087] Step 4.3: Based on the coordinate sequence of each vertex of the convex polygon boundary, the area parameter is obtained by the standard mathematical method for calculating the polygon area, and the perimeter parameter is obtained by accumulating the lengths of each boundary line segment. Specifically, this includes: first, organizing the coordinate sequence of each vertex according to the obtained order of the convex polygon boundary vertices, ensuring that the coordinate sequence can completely reflect the outline of the convex polygon in a clockwise or counterclockwise direction; when calculating the area parameter, the standard mathematical method for calculating the polygon area is adopted, taking two adjacent vertices in sequence according to the vertex coordinate sequence, and calculating the area parameter by combining the horizontal and vertical coordinates of the vertices; summing the calculation results of all adjacent vertices, taking the absolute value, and then dividing by two to obtain the area parameter of the convex polygon. The area parameter can reflect the size of the influence range of abnormal vibration and positional displacement on the surface of the box. When calculating the perimeter parameter, the length of the line segment between two adjacent vertices is calculated in sequence according to the vertex coordinate sequence. The calculation of the line segment length needs to be combined with the difference between the horizontal and vertical coordinates of the two vertices, and the lengths of the line segments between all adjacent vertices are accumulated to obtain the perimeter parameter of the convex polygon. The perimeter parameter can reflect the degree of dispersion of abnormal points on the surface of the box.

[0088] In this embodiment of the invention, discrete vibration points and position offset points are selected from the optimized behavioral feature sequence based on preset vibration amplitude thresholds and position offset thresholds. Based on the spatial coordinates of these points, the outer points are connected to obtain the boundary of the smallest convex polygon that completely contains all points after polar angle sorting. Then, the area parameter is calculated using standard mathematical methods based on the vertex coordinates of the convex polygon, and the perimeter parameter is obtained by accumulating the length of the boundary line segments. Therefore, this invention overcomes the technical problem of existing technologies lacking the extraction of spatial distribution features of abnormal box states, making it difficult to accurately characterize illegal opening behavior through spatial attributes, resulting in low differentiation between normal operation and illegal opening behavior, and high false alarm and false alarm rates. This invention thus supplements the spatial dimension features of abnormal box behavior, enabling the behavior classification model to more comprehensively judge the nature of opening behavior based on features.

[0089] In a preferred embodiment of the present invention, step 4 above may include:

[0090] Step 4.4: Integrate the area parameter and perimeter parameter as new feature dimensions into the optimized behavioral feature sequence to obtain an adjusted behavioral feature sequence containing spatial distribution features. Specifically, this involves: first, determining the timestamps corresponding to the area parameter and perimeter parameter to ensure that the timestamps corresponding to the area parameter and perimeter parameter are consistent with the time axis in the optimized behavioral feature sequence; then, inserting the area parameter and perimeter parameter as two new feature dimensions into the optimized behavioral feature sequence in chronological order, so that the feature data at each time point not only includes the original vibration, opening and closing, and position features, but also adds area and perimeter features reflecting spatial distribution. During the fusion process, check the temporal matching between the new features and the original features to ensure that all feature data at the same time point can accurately correspond, ultimately forming an adjusted behavioral feature sequence containing spatial distribution features.

[0091] Step 4.5: Based on the adjusted behavioral feature sequence, input the preset machine learning-based behavioral classification model for pattern recognition analysis to obtain the output results of the behavioral classification model. Specifically, this includes: constructing the machine learning behavioral classification model, which needs to focus on accurately distinguishing between normal and illegal opening of the box, combining multi-dimensional feature data. First, prepare the dataset by collecting multi-dimensional feature sequences of normal and illegal opening under different scenarios. After cleaning and removing invalid data and labeling the categories, divide the dataset into training, validation, and test sets in a 7:2:1 ratio. Then, perform feature engineering optimization by selecting key features through feature importance assessment, encoding non-numerical features, and standardizing numerical features to eliminate dimensional differences. Next, select a model based on the temporal and multi-dimensional correlation of the box's behavioral features, prioritizing ensemble learning models such as random forests, and selecting LSTM deep learning models when temporal features are significant. Finally, train the model, setting the initial super... The parameters are iteratively trained using the training set, and then tuned to optimal overall performance using methods such as grid search, based on the performance on the validation set. Model evaluation and optimization are then conducted, with the test set used to verify metrics such as accuracy, precision, and recall. If the performance is not up to standard, adjustments are made retrospectively. Simultaneously, generalization ability is verified in different enclosures and environments, with incremental training performed as needed. Finally, the model is deployed in a lightweight manner, converted to a different format, and deployed to the enclosure data processing terminal to ensure real-time response. A performance monitoring mechanism is established to periodically collect new data and iterate the model iteratively to adapt to environmental changes and new illegal opening methods. The adjusted behavioral feature sequence is input into the model, which analyzes each feature layer by layer to identify patterns of association between features. For example, illegal opening is often accompanied by a large vibration area, a long vibration circumference, and specific opening and closing state changes. Through pattern recognition analysis, the model outputs a result reflecting the degree of matching between the current opening behavior and known patterns.

[0092] Step 4.6: Based on the output of the behavior classification model, obtain the probability assessment value of whether the current opening behavior belongs to illegal opening behavior. Specifically, the output of the behavior classification model includes the matching probability of the current opening behavior with various known behavior patterns, which explicitly includes the matching probability with illegal opening behavior patterns. Extract the probability value from the output result. The probability value is usually represented by a value between 0 and 1. The closer the value is to 1, the higher the matching degree between the current opening behavior and the illegal opening behavior pattern, and the greater the possibility that it belongs to illegal opening behavior. The closer the value is to 0, the closer the current opening behavior is to the normal opening mode. By extracting the probability value, obtain the probability assessment value of whether the current opening behavior belongs to illegal opening behavior.

[0093] Step 4.7: Based on the comparison between the probability assessment value and the preset threshold parameter, determine whether the current opening behavior is an illegal opening behavior. Specifically, the preset threshold parameter is determined according to the security level requirements of the container, historical false alarm rate, and false alarm rate data. For example, for containers storing high-value materials, the threshold can be set to a lower value to reduce the risk of false alarms; for containers that are frequently operated daily, the threshold can be set to a higher value to reduce false alarms. The obtained probability assessment value is compared with the preset threshold parameter. If the probability assessment value is greater than or equal to the preset threshold, the current opening behavior is determined to be an illegal opening behavior; if the probability assessment value is less than the preset threshold, the current opening behavior is determined to be a normal opening behavior, thereby realizing the determination of the nature of the container opening behavior.

[0094] In this embodiment of the invention, the area and perimeter parameters of the convex hull are used as new spatial feature dimensions and fused into the optimized behavioral feature sequence. The integrated and adjusted behavioral feature sequence is then input into a preset machine learning-based behavioral classification model for pattern recognition analysis. The probability assessment value of whether the current opening behavior is illegal is obtained from the model output. The nature of the opening behavior is determined by comparing this probability assessment value with a preset threshold parameter. Therefore, this method overcomes the technical problems of existing technologies, such as the single dimension of behavioral features, the difficulty in distinguishing between normal and illegal opening based on simple signals, and the resulting high false alarm and false negative rates. This method enriches the dimensions and information content of behavioral features, enabling the machine learning model to more comprehensively capture the feature patterns of illegal opening behavior, improve the accuracy of determining the nature of the box opening behavior, and effectively reduce false alarms and false negatives.

[0095] In a preferred embodiment of the present invention, step 5 above may include:

[0096] Step 5.1: Based on the illegal opening judgment result output by the behavior classification model, generate the corresponding alarm signal and activate the emergency response protocol. Specifically, after the behavior classification model outputs the illegal opening judgment result, the data processing terminal immediately generates two types of alarm signals. One type is a local alarm signal, which triggers the audible and visual alarm device equipped on the cabinet, which emits a warning sound through a high-decibel buzzer and flashes a high-brightness LED light to deter illegal operation behavior on site. The other type is a remote alarm signal, which simultaneously activates the preset emergency response protocol.

[0097] Step 5.2: Based on the optimized behavioral feature sequence and convex hull geometric parameter data, extract the timestamp of the illegal opening behavior and the current position information of the enclosure. Specifically, this includes: extracting the timestamp corresponding to the first occurrence of vibration data or position offset data exceeding the threshold from the optimized behavioral feature sequence. The timestamp is the starting time of the illegal opening behavior. At the same time, extract the timestamp and position feature data within the continuous abnormal period. Combined with the enclosure reference position information collected and stored by the positioning equipment in the early stage, determine the current position information of the enclosure when the illegal opening behavior occurs. If the enclosure has moved, the key coordinate points in the position offset trajectory need to be extracted simultaneously to ensure that the occurrence time and the current position information can correspond to the occurrence process of the illegal behavior.

[0098] Step 5.3: Based on the timestamp of the occurrence, the current position information of the enclosure, and the geometric parameters of the convex hull, generate a complete record of the illegal opening behavior and store it in the safety event database. Specifically, this includes: integrating the extracted timestamp of the illegal opening behavior, the current position information of the enclosure, and the calculated geometric parameters of the convex hull, including the area, perimeter, and vertex coordinates of the convex hull; and supplementing key abnormal data in the optimized behavior feature sequence, such as the maximum vibration amplitude and the number of opening and closing state transitions, to form a complete record of the illegal opening behavior that includes the time of the behavior occurrence, the position of the enclosure, the spatial features, and the abnormal data; and storing the illegal opening behavior record in the safety event database according to a preset structured format.

[0099] Step 5.4: Based on the alarm signal, send a security alarm notification containing detailed information about the unauthorized opening behavior to the monitoring center and relevant responsible personnel through a preset communication interface. Specifically, based on the generated remote alarm signal, send a security alarm notification to two types of entities through the communication interface deployed on the enclosure: First, to the security management platform of the monitoring center, the notification content is displayed in the form of a pop-up window, including the time of the unauthorized opening behavior, the specific location of the enclosure, the abnormal range reflected by the convex hull parameters, and key vibration and location data, so that the monitoring center personnel can grasp the situation on site in real time; Second, to relevant responsible personnel, the notification is sent via SMS and security APP push, with concise and clear content including the core information of the incident and the contact person for handling, ensuring that the responsible personnel can quickly obtain information and go to the scene to handle the situation, avoiding response delays.

[0100] Step 5.5: Based on the illegal opening behavior records, update the security status assessment of the enclosure and trigger the corresponding equipment security protection mechanism. Specifically, this includes: updating the enclosure's security level according to the illegal opening behavior records stored in the security event database and the preset security status assessment standards. If it is the first illegal opening and no equipment damage has occurred, the security level is adjusted from normal to attention. If multiple illegal openings occur within a short period or equipment damage has occurred, the security level is adjusted to high risk, and the corresponding equipment security protection mechanism is triggered. For enclosures at the attention level, the data collection frequency is increased, such as increasing the vibration sensor collection frequency from once per second to five times per second, and shortening the status monitoring interval. For enclosures at the high risk level, in addition to increasing the monitoring frequency, surrounding security equipment is also linked, and higher-priority handling reminders are sent to maintenance personnel to ensure that the enclosure receives more stringent security protection in the future and reduce the risk of illegal opening from happening again.

[0101] In this embodiment of the invention, an alarm signal is generated based on the illegal opening determination result of the behavior classification model, and an emergency response protocol is initiated. The timestamp of the illegal behavior and the current position information of the enclosure are extracted from the optimized behavior feature sequence and convex hull geometric parameter data. A complete record of illegal opening behavior is generated and stored in the security event database. A security alarm notification with detailed information is sent to the monitoring center and relevant responsible personnel through a preset communication interface. At the same time, the enclosure safety status assessment is updated based on the behavior record, and the equipment safety protection mechanism is triggered. Therefore, this invention overcomes the technical problems of existing technologies, such as the lack of a closed-loop response mechanism, difficulty in quickly generating alarm information and retaining key evidence, delayed emergency response and inability to promptly link relevant parties, and lack of continuous safety status assessment and active protection. Thus, it achieves a complete closed-loop response of illegal opening behavior identification, alarm, recording, notification and protection, ensuring timely alarm, traceable evidence, rapid linkage and handling by relevant parties, and dynamic improvement of the enclosure safety protection level.

[0102] like Figure 2 As shown, embodiments of the present invention also provide an intelligent recognition system for illegal opening of a cabinet, comprising:

[0103] The acquisition module is used to monitor the physical state data of the enclosure in real time. The physical state data includes vibration signals, opening and closing status signals, and position information. Feature extraction is performed on the physical state data to obtain an initial behavioral feature sequence.

[0104] The analysis module is used to dynamically select three specific structural points on the surface of the enclosure as a set of reference positions based on the behavioral feature sequence. The specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. Based on the selected three specific structural points, a dynamically changing area is constructed. The dynamically changing area is divided into several sub-regions. Based on the dynamic change characteristics of each sub-region, a behavioral feature correction coefficient is obtained.

[0105] The calculation module is used to calibrate the initial behavior feature sequence based on the behavior feature correction coefficient to obtain an optimized behavior feature sequence; extract the discrete vibration points and position offset points of the box from the optimized behavior feature sequence; construct a convex hull boundary covering the discrete vibration points and position offset points and calculate the area and perimeter parameters; supplement the area and perimeter parameters into the optimized behavior feature sequence to form an adjusted behavior feature sequence, and input it into a preset behavior classification model to determine whether the opening behavior is illegal opening;

[0106] The processing module is used to generate an alarm signal and record relevant information about the illegal opening behavior when it is determined that the current opening behavior of the cabinet is illegal.

[0107] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligently recognizing unauthorized opening of a cabinet, characterized in that, The method includes: Real-time monitoring of the enclosure's physical state data, including vibration signals, opening and closing status signals, and position information; feature extraction is performed on the physical state data to obtain an initial behavioral feature sequence; Based on the behavioral feature sequence, three specific structural points on the surface of the enclosure are dynamically selected as a set of reference positions. These specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. A dynamically changing area is constructed based on the three selected specific structural points. The dynamically changing area is then divided into several sub-regions. Based on the dynamic change characteristics of each sub-region, a behavioral feature correction coefficient is obtained. Based on the behavioral feature correction coefficient, the initial behavioral feature sequence is calibrated to obtain the optimized behavioral feature sequence. From the obtained optimized behavior feature sequence, the discrete vibration points and position offset points of the box are extracted; the convex hull boundary covering the discrete vibration points and position offset points is constructed and the area and perimeter parameters are calculated; the area and perimeter parameters are added to the optimized behavior feature sequence to form an adjusted behavior feature sequence, which is then input into a preset behavior classification model to determine whether the opening behavior is illegal. When the current opening behavior of the enclosure is determined to be illegal, an alarm signal is generated and relevant information about the illegal opening behavior is recorded.

2. The intelligent identification method for illegal opening of a cabinet according to claim 1, characterized in that, Real-time monitoring of the enclosure's physical state data, including vibration signals, opening / closing status signals, and position information; feature extraction from the physical state data to obtain an initial behavioral feature sequence, including: Vibration sensors, opening / closing status sensors, and positioning devices deployed on the enclosure are used to collect vibration signals, opening / closing status signals, and position information in real time. Time-domain and frequency-domain analyses are performed on vibration signals to extract vibration characteristics including amplitude, frequency components, and duration. Perform state transition detection on the opening and closing state signal, and extract the timestamps and state change sequences of the opening and closing events; Differential calculations are performed on the location information to extract the position offset and motion trajectory features; The extracted vibration features, opening and closing event features, and positional motion features are fused and sequenced to obtain the initial behavioral feature sequence.

3. The intelligent identification method for illegal opening of a cabinet according to claim 2, characterized in that, Based on behavioral feature sequences, three specific structural points on the surface of the enclosure are dynamically selected as a set of reference positions. These specific structural points include the enclosure lock mounting point, the enclosure hinge connection point, and the enclosure seam monitoring point. A dynamically changing region is constructed based on these three specific structural points. This dynamically changing region is then divided into several sub-regions. Based on the dynamic characteristics of each sub-region, a behavioral feature correction coefficient is obtained, including: Based on the initial behavioral characteristic sequence, the intensity distribution and positional shift trend of the vibration signal are analyzed to dynamically determine the specific spatial coordinates of the enclosure lock installation point, enclosure hinge connection point, and enclosure seam monitoring point within the current monitoring cycle. Based on the spatial coordinates of three specific structural points, the minimum enclosing region formed by the three specific structural points is calculated, and the dynamically changing region range is constructed by expanding the preset tolerance threshold. The dynamically changing area is divided into several sub-regions by spatial grid, and the frequency of vibration events and the magnitude of positional shifts within each sub-region are analyzed based on behavioral feature sequences. Based on the dynamic variation characteristics of the vibration event frequency and position offset amplitude in each sub-region, a dynamic weighting coefficient is assigned to each sub-region. By combining the dynamic weight coefficients and spatial distribution relationships of all sub-regions, the behavioral feature correction coefficients are calculated.

4. The intelligent identification method for illegal opening of a cabinet according to claim 3, characterized in that, Based on the behavioral feature correction coefficient, the initial behavioral feature sequence is calibrated to obtain the optimized behavioral feature sequence, including: Based on the obtained behavioral feature correction coefficients, the initial behavioral feature sequence is adaptively weighted and calibrated to obtain the weighted and adjusted behavioral feature sequence. Based on the above weighted and adjusted behavioral feature sequence, a time window-based moving average filtering method is used for smoothing to suppress random noise introduced during data acquisition and generate a smoothed feature sequence. Based on the smoothed feature sequence described above, normalization calculations are performed to unify the dimensions and numerical ranges of each feature dimension, thereby generating the optimized behavioral feature sequence.

5. The intelligent identification method for illegal opening of a cabinet according to claim 4, characterized in that, From the obtained optimized behavioral feature sequence, the discrete vibration points and position offset points of the box are extracted; By constructing a convex hull boundary covering discrete vibration points and position offset points, and calculating the area and perimeter parameters, including: Based on the optimized behavioral feature sequence, data points exceeding the preset vibration amplitude threshold and position offset threshold are selected and used as discrete vibration points and position offset points, respectively. Based on the spatial coordinates of all extracted discrete vibration points and position offset points, the outer points are sequentially connected after polar angle sorting to obtain the minimum convex polygon boundary that completely contains all points. Based on the coordinate sequence of each vertex of the convex polygon boundary, the area parameter is obtained by the standard mathematical method for calculating the polygon area, and the perimeter parameter is obtained by accumulating the lengths of each boundary line segment.

6. The intelligent identification method for illegal opening of a cabinet according to claim 5, characterized in that, The area and perimeter parameters are added to the optimized behavior feature sequence to form an adjusted behavior feature sequence. This sequence is then input into a preset behavior classification model to determine whether the opening behavior is illegal, including: The area parameter and perimeter parameter are added as new feature dimensions and fused into the optimized behavioral feature sequence to obtain an adjusted behavioral feature sequence that includes spatial distribution features. Based on the adjusted behavioral feature sequence, a pre-defined machine learning-based behavioral classification model is input for pattern recognition analysis to obtain the output results of the behavioral classification model. Based on the output of the behavior classification model, the probability assessment value of the current opening behavior being an illegal opening behavior is obtained; Based on the comparison between the probability assessment value and the preset threshold parameter, it is determined whether the current opening behavior is an illegal opening behavior.

7. The intelligent identification method for illegal opening of a cabinet according to claim 6, characterized in that, When the current opening behavior of the enclosure is determined to be illegal, an alarm signal is generated, and relevant information about the illegal opening behavior is recorded, including: Based on the illegal opening determination result output by the behavior classification model, a corresponding alarm signal is generated and an emergency response protocol is activated. Based on the optimized behavioral feature sequence and convex hull geometric parameter data, the timestamp of the illegal opening behavior and the current position information of the box are extracted; Based on the timestamp of the occurrence, the current location information of the enclosure, and the geometric parameters of the convex hull, a complete record of the illegal opening behavior is generated and stored in the security event database. Based on the alarm signal, a security alarm notification containing detailed information about the unauthorized opening behavior is sent to the monitoring center and relevant responsible personnel through a preset communication interface; Based on the record of unauthorized opening, update the security status assessment of the enclosure and trigger the corresponding equipment security protection mechanism.

8. An intelligent recognition system for illegal opening of a cabinet, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to monitor the physical status data of the enclosure in real time. The physical status data includes vibration signals, opening and closing status signals and position information. Feature extraction is performed using physical state data to obtain an initial sequence of behavioral features; The analysis module is used to dynamically select three specific structural points on the surface of the enclosure as a set of reference positions based on the behavioral feature sequence. The specific structural points include the enclosure lock installation point, the enclosure hinge connection point, and the enclosure seam monitoring point. Based on the selected three specific structural points, a dynamically changing area is constructed. The dynamically changing area is divided into several sub-regions. Based on the dynamic change characteristics of each sub-region, a behavioral feature correction coefficient is obtained. The calculation module is used to calibrate the initial behavior feature sequence based on the behavior feature correction coefficient to obtain an optimized behavior feature sequence; and to extract the discrete vibration points and position offset points of the box from the optimized behavior feature sequence. By constructing a convex hull boundary covering discrete vibration points and position offset points and calculating area and perimeter parameters; the area and perimeter parameters are added to the optimized behavior feature sequence to form an adjusted behavior feature sequence, which is then input into a preset behavior classification model to determine whether the opening behavior is illegal; The processing module is used to generate an alarm signal and record relevant information about the illegal opening behavior when it is determined that the current opening behavior of the cabinet is illegal.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images