Safety management method for comprehensive security and protection integrated management system

Abstract

The invention discloses a security management method for a comprehensive security integrated management system, which relates to the technical field of security and comprises the following steps: converting security scene information into a standard data stream through an embedded processor; carrying out feature extraction on the standard data stream based on a convolutional neural network algorithm, analyzing the relevance of features in a time dimension through a long-short term memory model, generating a time sequence feature matrix, and storing the time sequence feature matrix into a feature database; inputting the time sequence characteristic matrix into a multi-layer perceptron network, performing classification calculation on the time sequence characteristic matrix according to a training sample library, and outputting a security level score which corresponds to a preset classification response rule; and calling the linkage control instruction according to the grading response rule, and feeding back an instruction execution state to the training sample library for optimizing the classification calculation parameters. According to the invention, from data acquisition, feature extraction and risk assessment to linkage disposal, accurate quantification of the security risk of heterogeneous data is realized.

Description

Technical Field

[0001] The present invention relates to the field of security technologies, and in particular to a security management method for an integrated security integration management system. Background Art

[0002] Traditional security management systems mainly rely on independent subsystems such as video surveillance, access control management, and alarm devices for security prevention and control. Information between the subsystems is isolated and the synergy is poor, making it difficult to achieve the fusion analysis and intelligent decision-making of multi-source data. Although some systems introduce artificial intelligence technologies for video analysis and behavior recognition, due to the lack of in-depth exploration of temporal features and analysis of scene relevance, the accuracy of security risk assessment is insufficient, and potential threats cannot be detected in a timely manner. At the same time, existing systems generally have problems such as a single response mechanism and low linkage disposal efficiency, making it difficult to meet the intelligent security management requirements in complex scenarios.

[0003] Currently, most integrated security management systems still use a fixed threshold judgment method based on rules for security level assessment. This method has poor adaptability to dynamically changing security risks and is prone to false alarms and missed alarms. In terms of feature extraction, traditional methods mainly focus on the static features of single-frame images, ignoring the evolution law of security events in the time dimension, which affects the accuracy of threat recognition. In addition, the linkage control strategies of existing systems are often preset and fixed, lacking an adaptive optimization mechanism and unable to dynamically adjust response measures according to the actual disposal effect. In terms of data utilization, a large amount of historical data has not been effectively transformed into knowledge and experience, and the learning ability and intelligent level of the system are limited in improvement.

[0004] These technical bottlenecks seriously restrict the effectiveness of integrated security integration management systems in practical applications. There is an urgent need to develop new intelligent security management methods to improve the perception, analysis, decision-making, and disposal capabilities of the system. Summary of the Invention

[0005] In view of the problems existing in the existing integrated security integration management systems in terms of feature extraction, temporal analysis, risk assessment, and linkage disposal, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to achieve intelligent perception of security scenarios, accurate assessment of security risks, and adaptive optimization of linkage disposal, so as to improve the intelligent level and actual combat effectiveness of the integrated security integration management system.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a security management method for an integrated security management system, which includes: a monitoring terminal collects security scenario information, and converts the security scenario information into a standard data stream through an embedded processor; based on a convolutional neural network algorithm, feature extraction is performed on the standard data stream, the relevance of the features in the time dimension is analyzed through a long short-term memory model, a temporal feature matrix is generated, and the temporal feature matrix is stored in a feature database; the temporal feature matrix is input into a multi-layer perceptron network, classification calculation is performed on the temporal feature matrix according to a training sample library, and a security level score is output, and the security level score corresponds to a preset hierarchical response rule; a linkage control instruction is called according to the hierarchical response rule, and the instruction execution status is fed back to the training sample library for optimizing classification calculation parameters.

[0009] As a preferred solution of the security management method for the integrated security management system of the present invention, wherein: the standard data stream includes an image information segment, a personnel information segment, and an alarm information segment; the security scenario information includes video data, audio data, and environmental parameter data; the conversion method of the standard data stream is to collect security scenario information through a multi-modal sensor array, and perform preprocessing and standardization conversion on the security scenario information through an embedded processor; the embedded processor includes a data preprocessing module, a priority arbitration module, a protocol conversion module, and a data encapsulation module; the priority arbitration module dynamically adjusts the processing priority of the security scenario information, and allocates processing time slices according to the priority order of the data buffer; the priority arbitration module divides the data buffer into a high-priority buffer, a medium-priority buffer, and a low-priority buffer; the data scheduled by the priority arbitration module is encoded into an environmental parameter data stream through the protocol conversion module, and packaged into a standard data stream through the data encapsulation module.

[0010] As a preferred solution of the security management method for the integrated security management system of the present invention, wherein: the method for generating the temporal feature matrix M ij is as follows: feature extraction is performed on the standard data stream based on a convolutional neural network and input into a bidirectional long short-term memory model; the feature vectors are arranged in time order to construct a temporal feature matrix M ij , where the number of rows i of the temporal feature matrix M ij is the feature dimension, and the number of columns j is the time step; principal component analysis is performed on the temporal feature matrix M ij for dimensionality reduction, and the dimensionality-reduced temporal feature matrix is stored in the feature database, and an index structure based on timestamps is established.

[0011] As a preferred solution of the security management method for the integrated security management system of the present invention, wherein: for the temporal feature matrix Mij Perform principal component analysis for dimensionality reduction, specifically including the following steps: For the time series feature matrix M ij Perform centering processing, calculate the mean vector μ of the feature dimension j , and subtract the original data by the mean vector μ j to obtain the centered matrix M' ij , where the mean vector μ j is obtained by taking the arithmetic mean of each row of the time series feature matrix M ij ; Calculate the covariance matrix C of the centered matrix M' ij , perform eigenvalue decomposition, and obtain the eigenvalues λ and the corresponding eigenvectors v; Arrange the eigenvalues λ in descending order, calculate the contribution rate r i and the cumulative contribution rate R i ; Select k eigenvectors corresponding to when the cumulative contribution rate R i reaches the preset threshold, construct the projection matrix P, and perform dimensionality reduction transformation on the centered matrix M' ij ; Store the feature matrix Y, the mean vector μ, and the projection matrix P in the feature database.

[0012] As a preferred solution of the security management method for the integrated security integration management system described in the present invention, wherein: The specific formula of the covariance matrix C is as follows:

[0013] C = (1 / n) × M' ij × M' ij T

[0014] where n is the number of samples, M' ij is the centered matrix, M' ij T is the transpose matrix of the centered matrix M' ij .

[0015] The contribution rate r i and the cumulative contribution rate R i are as follows:

[0016]

[0017] where λ i is the i-th eigenvalue, and n is the total number of eigenvalues.

[0018] The specific formula of the dimensionality reduction transformation is as follows:

[0019] Y = PT × M' ij

[0020] where Y is the feature matrix after dimensionality reduction, and PT is the transpose matrix of the projection matrix P.

[0021] As a preferred solution of the security management method for a comprehensive security integrated management system described in the present invention, wherein: the time series feature matrix is ​​input into a multi-layer perceptron network, the time series feature matrix is ​​classified and calculated according to a training sample library, and a security level score is output, and the security level score corresponds to a preset hierarchical response rule, including: the system constructs a multi-layer perceptron network for classification calculation, and trains and optimizes the multi-layer perceptron network based on the training sample library; the time series feature matrix is ​​normalized and input into the trained multi-layer perceptron network for classification calculation; the probability distribution of each security level is obtained through network forward propagation, and a weighted score is calculated by combining the probability value and the preset weight, and the security level score is output; a corresponding relationship is established between the security level score and the preset hierarchical response rule, and a linkage control instruction set is triggered according to the hierarchical response rule.

[0022] As a preferred solution of the security management method for a comprehensive security integrated management system described in the present invention, wherein: the hierarchical response rules include first level, second level, third level, fourth level and fifth level; the first level includes starting multi-system linkage instructions, switching security equipment to high-frequency sampling mode, activating fire alarm devices, and releasing access control system interlocks, and playing evacuation information; the second level includes calling security reinforcement instructions, turning on the intelligent tracking function, and recording key area trajectory information; the third level includes increasing the monitoring sampling frequency and recording abnormal target information; the fourth level includes increasing the monitoring frequency and turning on density detection; the fifth level includes maintaining the monitoring state and recording environmental parameters.

[0023] In the second aspect, an embodiment of the present invention provides a security management system for a comprehensive security integrated management system, which includes: a conversion module, which is used to monitor the terminal to collect security scene information, and convert the security scene information into a standard data stream through an embedded processor; an extraction module, which extracts features from the standard data stream based on a convolutional neural network algorithm, analyzes the correlation of the features in the time dimension through a long short-term memory model, generates a time series feature matrix, and stores the time series feature matrix in a feature database; a classification module, which is used to input the time series feature matrix into a multi-layer perceptron network, classify and calculate the time series feature matrix according to a training sample library, and output a security level score, which corresponds to a preset hierarchical response rule; a calling module, which is used to call a linkage control instruction according to the hierarchical response rule, and feed back the instruction execution status to the training sample library for optimizing the classification calculation parameters.

[0024] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program instructions are executed by the processor, the steps of the security management method for the integrated security integration management system as described in the first aspect of the present invention are implemented.

[0025] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program instructions are executed by the processor, the steps of the security management method for the integrated security integration management system as described in the first aspect of the present invention are implemented.

[0026] The beneficial effects of the present invention are as follows: multi-source security scenario information is collected through a multi-modal sensor array, and the hierarchical caching and priority scheduling mechanism of the embedded processor is used to achieve real-time preprocessing and standardized conversion of data, effectively solving the problem of unified access to heterogeneous data, and achieving a comprehensive improvement in the system's perception ability and a significant increase in data processing efficiency; by combining the spatial feature extraction ability of the convolutional neural network and the temporal correlation analysis ability of the bidirectional long short-term memory model, and using principal component analysis for dimensionality reduction optimization, multi-dimensional feature expression of the security scenario and mining of the temporal evolution law are realized, breaking through the limitation of traditional methods that only focus on static features, and enabling the system to have stronger scenario understanding and prediction capabilities; by constructing a multi-layer perceptron network for classification calculation and combining a weighted scoring mechanism of probability distribution and preset weights, accurate quantitative assessment of security risks is realized, improving the risk identification accuracy and early warning timeliness of the system. By establishing a hierarchical response rule and a linkage control mechanism, and feeding back the instruction execution status to the training sample library, adaptive optimization and continuous learning evolution of the disposal measures are realized, solving the problems of low linkage efficiency and insufficient optimization ability of traditional systems. Description of the Drawings

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0028] Figure 1 It is a flowchart of the security management method for the integrated security integration management system in Embodiment 1. Detailed Embodiments

[0029] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0030] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0031] Secondly, as used herein, "an embodiment" or "embodiments" refer to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The appearances of "in an embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other.

[0032] Embodiment 1

[0033] Referring to Figure 1 , this is the first embodiment of the present invention. This embodiment provides a security management method for an integrated security management system, including,

[0034] S1: The monitoring terminal collects security scenario information and converts the security scenario information into a standard data stream through an embedded processor.

[0035] Specifically, the method for converting the standard data stream is to collect security scenario information through a multi-modal sensor array and perform preprocessing and standardization conversion on the security scenario information through an embedded processor;

[0036] It should be noted that the monitoring terminal includes a video monitor, a personnel access scanner, and an infrared detector; the standard data stream includes an image information segment, a personnel information segment, and an alarm information segment; the security scenario information includes video data, audio data, and environmental parameter data.

[0037] Furthermore, the embedded processor includes a data preprocessing module, a priority arbitration module, a protocol conversion module, and a data encapsulation module; the priority arbitration module dynamically adjusts the processing priority of the security scenario information and allocates processing time slices according to the priority order of the data buffer; the priority arbitration module divides the data buffer into a high-priority buffer, a medium-priority buffer, and a low-priority buffer;

[0038] It should be noted that when an abnormal sound signal is detected, the audio data is allocated to the high-priority buffer; when a severe displacement signal is detected, the video data is allocated to the high-priority buffer; when an environmental parameter exceeds the limit, the environmental parameter data is allocated to the high-priority buffer.

[0039] Further, the data scheduled by the priority arbitration module is encoded into an environmental parameter data stream by the protocol conversion module and packed into a standard data stream by the data encapsulation module.

[0040] S2: Based on the convolutional neural network algorithm, feature extraction is performed on the standard data stream, the correlation of the features in the time dimension is analyzed through the long short-term memory model, a time series feature matrix is generated, and the time series feature matrix is stored in the feature database.

[0041] Specifically, the method for generating the time series feature matrix is to perform feature extraction on the standard data stream based on the convolutional neural network and input it into the bidirectional long short-term memory model.

[0042] It should be noted that the convolutional neural network includes five convolutional layer blocks, and each convolutional layer block is provided with a residual unit; the residual unit retains the shallow features through skip connections; the first convolutional layer block extracts the basic texture features, the convolutional kernel size is 7×7, and the stride is 2; the second to fifth convolutional layer blocks sequentially extract the middle-level object contour features and high-level semantic features, the convolutional kernel size is 3×3, and the number of channels of the feature map is adjusted through a 1×1 convolutional kernel; a max pooling layer is set between each convolutional layer block, the pooling kernel size is 2×2, and the spatial dimension of the feature map is reduced; after each convolutional layer, a batch normalization layer and a ReLU activation function are connected to reduce the risk of overfitting.

[0043] Further, the construction steps of the bidirectional long short-term memory model are as follows: The bidirectional long short-term memory model includes two hidden layers, the forward and the backward, and each hidden layer is set with 256 neurons; the memory unit of the long short-term memory model includes an input gate, a forget gate, and an output gate, where the input gate controls the writing of new information, and the threshold is set to 0.6; the forget gate controls the retention of historical information, and the threshold is set to 0.4; the output gate controls the output of information, and the threshold is set to 0.5; the bidirectional long short-term memory model performs forward and backward feature extraction on 30 consecutive frames of data, and dynamically adjusts the feature weights through the forget gate to generate a 512-dimensional feature vector reflecting the time series correlation.

[0044] Furthermore, the feature vectors are arranged in chronological order to construct a time series feature matrix M ij , where the number of rows i of the time series feature matrix M ij is the feature dimension, and the number of columns j is the time step. Perform principal component analysis dimensionality reduction on the time series feature matrix M ij and store the reduced-dimensional time series feature matrix in the feature database, and establish an index structure based on timestamps.

[0045] Specifically, performing principal component analysis dimensionality reduction on the time series feature matrix M ij specifically includes the following steps: Performing principal component analysis dimensionality reduction on the time series feature matrix M ijPerform centering processing and calculate the mean vector μ of the feature dimension j , and subtract the mean vector μ from the original data j to obtain the centered matrix M' ij , where the mean vector μ j is obtained by taking the arithmetic mean of each row of the time series feature matrix M ij . The specific formula for the centered matrix M' ij is as follows:

[0046] M' ij = M ij - μ j

[0047] Furthermore, calculate the covariance matrix C of the centered matrix M' ij , perform eigenvalue decomposition, and obtain the eigenvalues λ and the corresponding eigenvectors v; the specific formula for the covariance matrix C is as follows:

[0048] C = (1 / n) × M' ij × M' ij T

[0049] where n is the number of samples, M' ij is the centered matrix, and M' ij T is the transpose matrix of the centered matrix M' ij .

[0050] Even further, sort the eigenvalues λ in descending order, and calculate the contribution rate r i of the eigenvalues and the cumulative contribution rate R i , and the specific formulas are as follows:

[0051]

[0052] where λ i is the i-th eigenvalue and n is the total number of eigenvalues.

[0053] Specifically, select k eigenvectors corresponding to when the cumulative contribution rate Ri reaches a preset threshold, construct the projection matrix P, and perform dimensionality reduction transformation on the centered matrix M' ij through the projection matrix P. The specific formula is as follows:

[0054] Y = PT × M' ij

[0055] where Y is the feature matrix after dimensionality reduction, and PT is the transpose matrix of the projection matrix P.

[0056] It should be noted that the preset threshold is finally determined to be 95% cumulative contribution rate based on the data distribution characteristics of the original dimensional feature matrix in large-scale security scenarios, the system's real-time processing efficiency requirements, and the results of multiple sets of control experiments.

[0057] Furthermore, store the feature matrix Y, the mean vector μ, and the projection matrix P in the feature database.

[0058] S3 Input the time-series feature matrix into a multi-layer perceptron network, perform classification calculations on the time-series feature matrix according to the training sample library, and output a security level score, which corresponds to a preset hierarchical response rule.

[0059] Specifically, the system constructs a multi-layer perceptron network for classification calculations, and trains and optimizes the multi-layer perceptron network based on the training sample library; normalizes the time-series feature matrix and inputs it into the trained multi-layer perceptron network for classification calculations; obtains the probability distribution of each security level through network forward propagation, calculates the weighted score by combining the probability values and preset weights, and outputs the security level score; establishes a corresponding relationship between the security level score and the preset hierarchical response rule, and triggers the linkage control instruction set according to the hierarchical response rule.

[0060] It should be noted that the multi-layer perceptron network is a five-layer perceptron network structure. The number of input layer nodes matches the dimension of the time-series feature matrix. The first hidden layer is configured with 1024 neurons, and the ReLU activation function is used to extract feature combinations; the second hidden layer is configured with 512 neurons, and the tanh activation function is used to enhance the non-linear expression ability; the third hidden layer is configured with 256 neurons, and the PReLU activation function is used to extract high-order feature patterns; the output layer uses the Softmax function to map to the probability distribution of 5 security levels. A residual connection mechanism is used between layers to prevent gradient disappearance, and Batch Normalization is used to accelerate training convergence.

[0061] Furthermore, the training sample library contains 500,000 sets of labeled historical case data, which are divided into a training set and a validation set according to a ratio of 80:20. A weighted cross-entropy loss function is used, and different weight coefficients are set for different security levels: the weight for the particularly serious level is 1.5, the weight for the serious level is 1.3, the weight for the relatively large level is 1.1, the weight for the general level is 1.0, and the weight for the minor level is 0.9. The Adam optimizer is used for parameter updates, the initial learning rate is set to 0.001, and it decays to 0.8 times the original every 50 epochs.

[0062] Furthermore, a corresponding relationship is established between the security level score and a preset hierarchical response rule. 90 - 100 points correspond to the first level, and the first-level response plan is activated; 80 - 89 points correspond to the second level, and the second-level response plan is activated; 70 - 79 points correspond to the third level, and the third-level response plan is activated; 60 - 69 points correspond to the fourth level, and the fourth-level response plan is activated; 0 - 59 points correspond to the fifth level, and the fifth-level response plan is activated. The system automatically triggers the processing flow of the corresponding level through the scoring threshold.

[0063] S4: Invoke the linkage control instruction according to the hierarchical response rule, and feedback the instruction execution status to the training sample library for optimizing the classification calculation parameters.

[0064] Specifically, trigger the linkage control instruction set according to the hierarchical response rule. The hierarchical response rule includes the first level, the second level, the third level, the fourth level, and the fifth level; the first level includes activating the multi-system linkage instruction, switching the security equipment to the high-frequency sampling mode, activating the fire alarm device, releasing the interlock of the access control system, and playing the evacuation information; the second level includes invoking the security reinforcement instruction, enabling the intelligent tracking function, and recording the trajectory information of key areas; the third level includes increasing the monitoring sampling frequency,

[0065] starting the trajectory analysis, and recording the abnormal target information; the fourth level includes increasing the basic monitoring frequency and enabling the density detection; the fifth level includes maintaining the basic monitoring state and recording the environmental parameters.

[0066] Further, the system records the execution status information of the linkage control instruction. The status information includes the instruction issuance timestamp, the device response confirmation flag, the execution action completion rate, the response time statistics, and the device operation parameters; at the same time, record the on-site disposal effect, including the personnel evacuation speed, the hazard source control situation, the device linkage efficiency, and the system response delay; in addition, it also includes the instruction execution exception information, such as the device failure type, the network communication status, and the manual intervention record.

[0067] Furthermore, the execution status information is sorted in a preset format and stored in the training sample library. The sample data structure includes four parts: the scene feature vector, the security level annotation, the response measure record, and the processing effect evaluation; through the sliding window mechanism, keep the sample data of the most recent 90 days in an active state; evaluate the sample quality, eliminate the noise samples and duplicate samples; establish a sample weight system, and assign different weights according to the case typicality and timeliness.

[0068] Specifically, optimize the classification calculation parameters based on the updated training sample library. Adopt the online learning method, use the sample data within the sliding window to calculate the parameter gradient; dynamically adjust the weight coefficients of each security level in the cross-entropy loss function; update the convolution kernel parameters of the feature extraction network; optimize the memory unit parameters of the time series analysis model. Continuously improve the model performance through incremental learning, enabling the system to gradually adapt to new scenario features and threat patterns.

[0069] Furthermore, this embodiment also provides a security management system for an integrated security management system, including: a conversion module for monitoring the terminal to collect security scenario information and converting the security scenario information into a standard data stream through an embedded processor; an extraction module for extracting features from the standard data stream based on the convolutional neural network algorithm, analyzing the correlation of the features in the time dimension through a long short-term memory model, generating a time series feature matrix, and storing the time series feature matrix in a feature database; a classification module for inputting the time series feature matrix into a multi-layer perceptron network, performing classification calculation on the time series feature matrix according to the training sample library, and outputting a security level score, where the security level score corresponds to a preset hierarchical response rule; a calling module for calling a linkage control instruction according to the hierarchical response rule and feeding back the instruction execution status to the training sample library for optimizing the classification calculation parameters.

[0070] This embodiment also provides a computer device applicable to the security management method of an integrated security management system, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the security management method for an integrated security management system as proposed in the above embodiment.

[0071] This computer device can be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0072] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, the following steps are implemented: monitoring terminals collect security scenario information, and converting the security scenario information into a standard data stream through an embedded processor; extracting features from the standard data stream based on the convolutional neural network algorithm, analyzing the correlation of the features in the time dimension through a long short-term memory model, generating a time series feature matrix, and storing the time series feature matrix in a feature database; inputting the time series feature matrix into a multi-layer perceptron network, performing classification calculation on the time series feature matrix according to a training sample library, and outputting a security level score, where the security level score corresponds to a preset hierarchical response rule; calling a linkage control instruction according to the hierarchical response rule, and feeding back the instruction execution status to the training sample library for optimizing classification calculation parameters.

[0073] In summary, by collecting multi-source security scenario information through a multi-modal sensor array and using the hierarchical caching and priority scheduling mechanism of an embedded processor, real-time preprocessing and standardization conversion of data are achieved, effectively solving the problem of unified access to heterogeneous data, and comprehensively improving the system's perception ability and significantly enhancing the data processing efficiency; by combining the spatial feature extraction ability of a convolutional neural network and the time series correlation analysis ability of a bidirectional long short-term memory model, and using principal component analysis for dimensionality reduction optimization, multi-dimensional feature expression of security scenarios and mining of time series evolution laws are realized, breaking through the limitations of traditional methods that only focus on static features, and enabling the system to have stronger scene understanding and prediction capabilities; by constructing a multi-layer perceptron network for classification calculation and combining a weighted scoring mechanism of probability distribution and preset weights, accurate quantitative assessment of security risks is achieved, improving the system's risk identification accuracy and early warning timeliness. By establishing a hierarchical response rule and a linkage control mechanism, and feeding back the instruction execution status to the training sample library, adaptive optimization and continuous learning and evolution of disposal measures are realized, solving the problems of low linkage efficiency and insufficient optimization ability of traditional systems.

[0074] Embodiment 2

[0075] Referring to Table 1, this is the second embodiment of the present invention. This embodiment provides a security management method for an integrated security management system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0076] Specifically, a 30-day system test was conducted in a large commercial complex, where 215 high-definition cameras, 168 audio collectors, and 286 environmental sensors (including smoke sensors, temperature sensors, infrared sensors, etc.) were deployed inside the building; an embedded processing unit configured with an Inteli7-9750H processor and 16GB of memory was used as an edge computing node, and a deep learning model developed based on the TensorFlow framework was deployed. The improved ResNet50 network structure was used for feature extraction, the bidirectional LSTM network (hidden layer dimension 256) was used for time series analysis, and the multi-layer perceptron adopted a four-layer structure (input layer - 1024 - 512 - 256 - output layer).

[0077] Furthermore, the system collects scene data in real time through a multi-modal sensor array. The sampling frequency of the environmental sensors is set to 10Hz, the video capture frame rate is 25fps, and the audio sampling rate is 16kHz. Through the priority arbitration module, the alarm information is assigned to the high-priority buffer (processing delay < 50ms), the data related to abnormal behaviors is assigned to the medium-priority buffer (processing delay < 200ms), and the regular monitoring data is assigned to the low-priority buffer (processing delay < 500ms).

[0078] Specifically, after feature extraction of the collected data, a feature matrix (original dimension 1024) is generated through time series analysis, and it is reduced to 256 dimensions through principal component analysis (cumulative contribution rate reaches 95%). The system is trained based on 30,000 groups of labeled samples, using the Adam optimizer, with the learning rate set to 0.001 and the batch size of 64. During the training process, the generalization ability of the model is ensured through 5-fold cross-validation.

[0079] Furthermore, as shown in Table 1, in terms of the recognition accuracy of abnormal events, the system of the present invention reaches 94.2%, which is 15.7 percentage points higher than that of the traditional system. This is mainly due to the synergistic effect of multi-modal data fusion and time series feature analysis, enabling the system to more comprehensively understand the scene context information. The average response time is significantly reduced. The system of the present invention only needs 320ms to complete the whole process from perception to response, which is 530ms faster than the traditional system.

[0080] Table 1 Test data table

[0081] Evaluation Index Traditional System System of the Present Invention Accuracy Rate of Abnormal Event Recognition (%) 78.5 94.2 Average Response Time (ms) 850 320 False Alarm Rate (%) 12.3 3.8 Missed Alarm Rate (%) 15.6 4.5 System Resource Occupancy Rate (%) 85 58 Processing Throughput (events / second) 125 280

[0082] Furthermore, in terms of reliability indicators, the false alarm rate and the missed alarm rate of the present invention are 3.8% and 4.5% respectively, which are 8.5 and 11.1 percentage points lower than those of the traditional system. This indicates that the feature extraction and risk assessment methods proposed by the present invention have stronger robustness and can effectively cope with various security risks in complex scenarios. In terms of system resource utilization, through the improved data processing flow and dimensionality reduction optimization, the present invention controls the resource occupancy rate at 58%, and at the same time increases the processing throughput to 280 events / second, demonstrating excellent performance efficiency ratio.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A security management method for a comprehensive security integrated management system, characterized by: include, The monitoring terminal collects security scene information and converts the security scene information into a standard data stream through an embedded processor; Extracting features from the standard data stream based on a convolutional neural network algorithm, analyzing the relevance of the features in the time dimension through a long short-term memory model, generating a time series feature matrix, and storing the time series feature matrix in a feature database; Input the time series feature matrix into a multi-layer perceptron network, classify and calculate the time series feature matrix according to a training sample library, and output a security level score, wherein the security level score corresponds to a preset graded response rule; The linkage control instruction is called according to the hierarchical response rule, and the instruction execution status is fed back to the training sample library for optimizing the classification calculation parameters.

2. The safety management method for a comprehensive security integrated management system according to claim 1, characterized in that: The standard data stream includes an image information segment, a personnel information segment and an alarm information segment; the security scene information includes video data, audio data and environmental parameter data; the conversion method of the standard data stream is: Collect security scene information through a multimodal sensor array, and pre-process and standardize the security scene information through an embedded processor; The embedded processor includes a data preprocessing module, a priority arbitration module, a protocol conversion module and a data encapsulation module; The priority arbitration module dynamically adjusts the processing priority of the security scene information and allocates processing time slices according to the priority order of the data buffer area; The priority arbitration module divides the data cache area into a high priority cache area, a medium priority cache area and a low priority cache area; The data scheduled by the priority arbitration module is encoded into an environmental parameter data stream by the protocol conversion module, and is packaged into a standard data stream by the data encapsulation module.

3. The safety management method for a comprehensive security integrated management system according to claim 2, characterized in that: The generated time series feature matrix M ij The method is, Extracting features from the standard data stream based on a convolutional neural network and inputting the features into a bidirectional long short-term memory model; Arrange the feature vectors in chronological order to construct a time series feature matrix M ij , where the time series feature matrix M ij The number of rows i is the feature dimension, and the number of columns j is the time step; For the time series feature matrix M ij Perform principal component analysis to reduce the dimension, store the reduced time series feature matrix into the feature database, and establish an index structure based on timestamp.

4. The safety management method for a comprehensive security integrated management system according to claim 3, characterized in that: For the time series feature matrix M ij Perform principal component analysis to reduce dimension, which includes the following steps: For the time series feature matrix M ij Perform centralization and calculate the mean vector μ of the feature dimension j , and subtract the mean vector μ from the original data j Get the centralization matrix M i ' j , where the mean vector μ j By ij The arithmetic mean of each row is obtained; Calculate the centralization matrix M i ' j The covariance matrix C of is decomposed into eigenvalues ​​to obtain the eigenvalues ​​λ and the corresponding eigenvectors v; Arrange the eigenvalues ​​λ in descending order and calculate the contribution rate r of the eigenvalues i and the cumulative contribution rate R i ; Select the cumulative contribution rate R i When the preset threshold is reached, the corresponding k eigenvectors are used to construct the projection matrix P, and the centralization matrix M i ' j Perform dimensionality reduction transformation; The feature matrix Y, mean vector μ and projection matrix P are stored in a feature database.

5. The safety management method for a comprehensive security integrated management system according to claim 4, characterized in that: The specific formula of the covariance matrix C is as follows: C=(1 / n)×M i ' j ×M i ' j T Where n is the number of samples, M i ' j is the centralization matrix, M i ' j T is the central matrix M i ' j The transposed matrix of The contribution rate r of the eigenvalue i and the cumulative contribution rate R i The specific formula is as follows: Among them, λ i is the i-th eigenvalue, n is the total number of eigenvalues; The specific formula of the dimensionality reduction transformation is as follows: Y=PT×M i ' j Among them, Y is the feature matrix after dimensionality reduction, and PT is the transposed matrix of the projection matrix P.

6. The safety management method for a comprehensive security integrated management system according to claim 4, characterized in that: The time series feature matrix is ​​input into a multi-layer perceptron network, and the time series feature matrix is ​​classified and calculated according to a training sample library, and a security level score is output, and the security level score corresponds to a preset graded response rule, including: The system constructs a multi-layer perceptron network for classification calculations, and optimizes the training of the multi-layer perceptron network based on the training sample library; Normalize the time series feature matrix and input it into the trained multi-layer perceptron network for classification calculation; The probability distribution of each security level is obtained through network forward propagation, and the weighted score is calculated by combining the probability value and the preset weight to output the security level score; A correspondence is established between the security level score and the preset hierarchical response rules, and the linkage control instruction set is triggered according to the hierarchical response rules.

7. The safety management method for a comprehensive security integrated management system according to claim 6, characterized in that: The hierarchical response rules include the first level, the second level, the third level, the fourth level and the fifth level; the first level includes starting a multi-system linkage command, switching the security equipment to the high-frequency sampling mode, activating the fire alarm device, releasing the access control system interlock, and playing evacuation information; the second level includes calling the security reinforcement command, turning on the intelligent tracking function, and recording the trajectory information of key areas; the third level includes increasing the monitoring sampling frequency and recording abnormal target information; the fourth level includes increasing the monitoring frequency and turning on density detection; the fifth level includes maintaining the monitoring status and recording environmental parameters.

8. A security management system for a comprehensive security integrated management system, based on the security management method for a comprehensive security integrated management system according to any one of claims 1 to 7, characterized in that: Also includes, A conversion module, used for monitoring the terminal to collect security scene information, and converting the security scene information into a standard data stream through an embedded processor; An extraction module, which extracts features from the standard data stream based on a convolutional neural network algorithm, analyzes the relevance of the features in the time dimension through a long short-term memory model, generates a time series feature matrix, and stores the time series feature matrix in a feature database; A classification module, used for inputting the time series feature matrix into a multi-layer perceptron network, performing classification calculation on the time series feature matrix according to a training sample library, and outputting a security level score, wherein the security level score corresponds to a preset graded response rule; The calling module is used to call the linkage control instruction according to the hierarchical response rule, and feed back the instruction execution status to the training sample library for optimizing the classification calculation parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the security management method for a comprehensive security integrated management system described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the security management method for a comprehensive security integrated management system described in any one of claims 1 to 7 are implemented.

Images