A content security AI edge management system and method based on neural network
The neural network-based content security AI edge management system addresses the inadequacies of traditional methods by performing comprehensive content analysis and resource allocation to ensure secure online video content management.
Patent Information
- Application Number
- CN202411468669.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-21
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2044-10-21
AI Technical Summary
Traditional content filtering methods are inadequate for accurately and effectively controlling online video content, failing to address the security concerns associated with harmful and sensitive information.
A neural network-based content security AI edge management system that includes a source-end content security management module, security management analysis module, and security management allocation module, performing comprehensive analysis of video content from multiple aspects to determine safety and allocate resources efficiently.
The system ensures accurate and efficient security control of online videos by performing three-dimensional analysis of text, audio, and image content, and optimally distributing analysis resources among edge devices, thereby preventing the upload of sensitive content.
Smart Images
Figure CN118972628B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of content control technology, and more specifically, to a neural network-based content security AI edge control system and method. Background Art
[0002] With the rapid development of the Internet, the security of online content has become increasingly prominent. At present, online videos have become a carrier for the dissemination of more and more information content. Online videos can contain a lot of valuable information content, but they are also mixed with harmful and sensitive content information.
[0003] Traditional content filtering and control methods can only target image content and text content, and cannot perform accurate and effective security control when facing a large number of online videos. Therefore, how to accurately and effectively control the content in online videos and ensure that the upload of online videos meets security control requirements is an urgent problem that needs to be solved. Summary of the invention
[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a content security AI edge management system and method based on neural network.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A neural network-based content security AI edge management system, including a source-end content security management module, a security management analysis module, and a security management allocation module;
[0007] The source-end content security control module is used to obtain the edge control range of the edge device, and obtain the control data source based on the edge control range, periodically perform security analysis on the control data source, and obtain the edge control value of the control data source after performing security analysis on the videos to be uploaded at each control data source;
[0008] The security control analysis module controls the edge device to perform security analysis on the videos to be uploaded from all control data sources, obtain the control allocation value of the edge device, and determine whether to mark the edge device as an allocated edge device based on the comparison result between the control allocation value and the control allocation threshold, and obtain the allocated data source after the edge device is marked as an allocated edge device;
[0009] The security management and control allocation module is used to allocate the allocated data source end to the allocated device, and control the allocated device to subsequently perform security analysis on the video to be uploaded at the allocated data source end.
[0010] Furthermore, the edge control range of the edge device is obtained, and the control data source end is obtained based on the edge control range. Specifically, the edge device is controlled to draw a circle with its own position as the center and set the radius r to obtain the edge control range, and the data source end whose position is within the edge control range is marked as the control data source end.
[0011] Furthermore, a security analysis is performed on the control data source end periodically, specifically: the control data source end is sorted in descending order according to the numerical value of the control primary value, a preset time period is set, and at each time node of the preset time period, a security analysis is performed on the videos to be uploaded on the control data source end in the sorted order.
[0012] Furthermore, the primary control value of the control data source is obtained in the following way: obtain the i edge control values obtained by the control data source before the current time of the system, sort all the edge control values in the order of acquisition time, calculate the difference between the two adjacent edge control values after sorting and take the absolute value to obtain the edge control turbulence value, sum up all the edge control turbulence values and take the average to obtain the control turbulence mean Bz, sum up all the edge control values and take the average to obtain the edge control mean Ec, and use the formula The primary control value Mx of the control data source is obtained, where u1 is the control turbulence mean coefficient and u2 is the edge control mean coefficient.
[0013] Furthermore, the edge control value of the control data source is obtained in the following manner: obtain all videos to be uploaded on the control data source within a preset time period, obtain the content security analysis value of each video to be uploaded, set the content security analysis limit value, and when the content security analysis value is greater than or equal to the content security analysis limit value, mark the video to be uploaded as a security-sensitive video, and prevent the security-sensitive video from being uploaded; when the content security analysis value is less than the content security analysis limit value, mark the video to be uploaded as a safe and reasonable video, and calculate the difference between the total number of security-sensitive videos and the total number of safe and reasonable videos to obtain the edge control value of the control data source.
[0014] Furthermore, the content security analysis value of the video to be uploaded is obtained in the following manner: a preset frame extraction interval is set, frame extraction is performed on the video to be uploaded based on the preset frame extraction interval to obtain multiple video frames, all video frames are sorted in chronological order, two adjacent video frames after sorting are grouped together and marked as a content security analysis group, a group security value SJm of each content security analysis group is obtained, a group security value coefficient is set to PCw, a group security limit value is set, when the group security value is greater than or equal to the group security limit value, the content security analysis group is marked as a content sensitive group, when the group security value is less than the group security limit value, the content security analysis group is marked as a content normal group, the total number of content sensitive groups is marked as KDg, the total number of content normal groups is marked as ZMt, and the formula is used. The content security analysis value SLb of the video to be uploaded is obtained, where m=1, 2, ..., M, and M is the total number of content security analysis groups.
[0015] Furthermore, the group security value of the content security analysis group is obtained in the following way: two video frames in the content security analysis group are marked as the starting video frame and the ending video frame respectively, the word vector of the starting video frame and the word vector of the ending video frame are obtained, the audio between the starting video frame and the ending video frame in the uploaded video is intercepted to obtain the security group audio segment, the audio signal of the security group audio segment is extracted to obtain the audio segment signal feature, the image feature of the starting video frame and the ending video frame is extracted to obtain the image feature of the starting video frame and the image feature of the ending video frame, the text sensitivity times SKd of the content security analysis group are obtained, the content evaluation value WSr of the security group audio segment is obtained, the image security value TAZ of the content security analysis group is obtained, and the formula is used Get the group security value SJm of the content security analysis group, where s1 is the text sensitivity coefficient;
[0016] The word vector is obtained by: extracting the text in the video frame, preprocessing the extracted text, and obtaining the word vector of the preprocessed text through the Word2Vec model;
[0017] The text sensitivity times of the content security analysis group are obtained in the following manner: obtaining a content security model of the text type, using the word vector of the starting video frame and the word vector of the ending video frame in the content security analysis group as input data of the text type content security model, outputting a content evaluation value of the text type of the starting video frame and a content evaluation value of the text type of the ending video frame, setting a content evaluation threshold of the text type, and when the content evaluation value of the text type is greater than or equal to the content evaluation threshold, increasing the text sensitivity times by one, and marking the text sensitivity times as SKd;
[0018] The content rating value of the safety group audio segment is obtained by: obtaining a content security model of the audio type, using the audio segment signal feature as input data of the audio type content security model, and outputting the content rating value WSr of the safety group audio segment;
[0019] The image safety value of the content safety analysis group is obtained in the following manner: obtaining a content safety model of the image type, taking the image features of the starting video frame and the image features of the ending video frame in the content safety analysis group as input data of the content safety model of the image type, outputting a content evaluation value of the image type of the starting video frame and a content evaluation value of the image type of the ending video frame, summing the content evaluation values of the two image types to obtain an image evaluation superposition value RYe, performing difference calculation on the content evaluation values of the two image types to obtain an image evaluation fluctuation value CFp, and using the formula The image safety value TAZ of the content safety analysis group is obtained, wherein p1 is the image evaluation superposition coefficient and p2 is the image evaluation fluctuation coefficient.
[0020] Furthermore, the control allocation value of the edge device is obtained in the following manner: the edge control value of each control data source is obtained, and the edge control values of all control data sources are summed to obtain the control allocation value.
[0021] Further, based on the comparison result between the control allocation value and the control allocation threshold, it is determined whether the edge device needs to be marked as an allocated edge device. After the edge device is marked as an allocated edge device, the allocated data source end is obtained, specifically: the control allocation threshold is set, when the control allocation value is greater than or equal to the control allocation threshold, the edge device is marked as an allocated edge device, and the control data source end with the largest edge control value in the allocated edge device is marked as the allocated data source end, and when the control allocation value is less than the control allocation threshold, no processing is performed;
[0022] Assign the allocated data source end to the designated device, and control the designated device to perform security analysis on the videos to be uploaded from the allocated data source end. Specifically, take the position of the allocated data source end as the center, set the radius R to draw a circle, and obtain the edge allocation range. Mark the remaining edge devices within the edge allocation range as target allocation devices, mark the target allocation device with the smallest control allocation value as the designated device, and control the designated device to perform security analysis on the videos to be uploaded from the allocated data source end. When the control allocation value is less than the control allocation threshold, no processing will be performed.
[0023] Furthermore, a content security AI edge management method based on a neural network includes the following steps:
[0024] Step 1: Obtain the edge control scope of the edge device, and obtain the control data source based on the edge control scope;
[0025] Step 2: Periodically perform security analysis on the control data source;
[0026] Step 3: After performing security analysis on the videos to be uploaded at each control data source, obtain the edge control value of the control data source;
[0027] Step 4: After the edge device performs security analysis on the videos to be uploaded from all control data sources, the control edge device obtains the control allocation value of the edge device;
[0028] Step 5: Based on the comparison result between the control allocation value and the control allocation threshold, determine whether the edge device needs to be marked as an allocated edge device, and after the edge device is marked as an allocated edge device, obtain the allocated data source end;
[0029] Step 6: Allocate the allocated data source to the assigned device, and control the assigned device to subsequently perform security analysis on the video to be uploaded from the allocated data source.
[0030] Compared with the prior art, the present invention has the following beneficial effects:
[0031] 1. The method of the present invention performs security control on the video at the data source end by combining a neural network with edge control, forming a control linkage between multiple edge devices to avoid uneven distribution of control resources in edge devices;
[0032] 2. Set up a source-side content security management and control module to conduct a three-dimensional comprehensive analysis of each video to be uploaded from the three aspects of text, audio, and image, and efficiently and accurately determine whether the content of each video to be uploaded in the data source is safe. Set up a security management and control analysis module and a security management and control allocation module to manage and analyze edge devices, quickly and effectively allocate analysis resources between edge devices, and ensure that edge devices can efficiently analyze the video content on the data source. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is a flowchart of a content security AI edge control method based on neural network;
[0034] Figure 2 This is the operation flow chart of the source-side content security control module;
[0035] Figure 3 Assign module operation flow chart for security control;
[0036] Figure 4 This is a module block diagram of a neural network-based content security AI edge management system. DETAILED DESCRIPTION
[0037] Example 1: Reference Figure 1 , a content security AI edge management method based on neural network, comprising the following steps:
[0038] Step 1: Obtain the edge control scope of the edge device, and obtain the control data source based on the edge control scope;
[0039] Step 2: Periodically perform security analysis on the control data source;
[0040] Step 3: After performing security analysis on the videos to be uploaded at each control data source, obtain the edge control value of the control data source;
[0041] Step 4: After the edge device performs security analysis on the videos to be uploaded from all control data sources, the control edge device obtains the control allocation value of the edge device;
[0042] Step 5: Based on the comparison result between the control allocation value and the control allocation threshold, determine whether the edge device needs to be marked as an allocated edge device, and after the edge device is marked as an allocated edge device, obtain the allocated data source end;
[0043] Step 6: Allocate the allocated data source to the assigned device, and control the assigned device to subsequently perform security analysis on the video to be uploaded from the allocated data source.
[0044] Through the above method, the present invention can perform security management on the video at the data source end by combining neural network and edge management, forming a management linkage between multiple edge devices, and avoiding uneven distribution of management resources in edge devices.
[0045] Example 2: Reference Figure 2-Figure 4 , a neural network-based content security AI edge management and control system, including a source content security management and control module, a security management and control analysis module, and a security management and control allocation module.
[0046] Source-end content security control module: The edge device uses its own position as the center of the circle and sets the radius r to draw a circle (r is the length of the radius, and the length can be set according to needs) to obtain the edge control range, mark the data source end within the edge control range as the controlled data source end, sort the controlled data source ends in descending order according to the value of the control primary value, set a preset time period (the period length of the preset time period, that is, the length between two time nodes, can be set and adjusted according to needs), and at each time node of the preset time period, perform security analysis on the videos to be uploaded on the controlled data source end in the sorted order. After performing security analysis on the videos to be uploaded on each controlled data source end, obtain the edge control value of the controlled data source end.
[0047] Set up a source-side content security management module to conduct a three-dimensional comprehensive analysis of each video to be uploaded from the three aspects of text, audio, and image, and efficiently and accurately determine whether the content of each video to be uploaded in the data source is safe.
[0048] Security control analysis module: After the edge device performs security analysis on the videos to be uploaded from all control data sources (only for all control data sources that need to be analyzed at this time node), it obtains the control allocation value of the edge device and sets the control allocation threshold (system threshold, set and adjusted according to needs). When the control allocation value is greater than or equal to the control allocation threshold, the edge device is marked as an allocated edge device, and the control data source with the largest edge control value in the allocated edge devices is marked as the allocated data source. When the control allocation value is less than the control allocation threshold, no processing is performed.
[0049] Security control allocation module: With the position of the allocated data source as the center, set the radius R to draw a circle to obtain the edge allocation range, mark the remaining edge devices (except the allocated edge devices) within the edge allocation range as target allocation devices, mark the target allocation device with the smallest control allocation value as the fixed allocation device, and control the fixed allocation device to perform security analysis on the videos to be uploaded at the allocated data source. When the control allocation value is less than the control allocation threshold, no processing is performed.
[0050] By setting up a security control analysis module and a security control allocation module, you can perform control and analysis on edge devices, quickly and effectively allocate analysis resources between edge devices, and ensure that edge devices can efficiently analyze video content at the data source.
[0051] The control allocation value of the edge device is obtained in the following way: the edge control value of each control data source is obtained, and the edge control values of all control data sources are summed to obtain the control allocation value.
[0052] The primary control value of the control data source is obtained in the following way: obtain the i edge control values obtained by the control data source before the current time of the system, sort all edge control values in the order of acquisition time, calculate the difference between the two adjacent edge control values after sorting and take the absolute value to obtain the edge control turbulence value, sum up all edge control turbulence values and take the average to obtain the control turbulence mean Bz, sum up all edge control values and take the average to obtain the edge control mean Ec, and use the formula The primary control value Mx of the control data source is obtained, where u1 is the control turbulence mean coefficient, u2 is the edge control mean coefficient, the value of u1 is 0.32, and the value of u2 is 0.28.
[0053] The edge control value of the control data source is obtained in the following way: obtain all videos to be uploaded on the control data source within a preset time period (i.e., all videos to be uploaded generated between the current time node and the previous time node), obtain the content security analysis value of each video to be uploaded, set the content security analysis limit value (system threshold, set and modified according to needs), when the content security analysis value is greater than or equal to the content security analysis limit value, mark the video to be uploaded as a security-sensitive video, and prevent the security-sensitive video from being uploaded; when the content security analysis value is less than the content security analysis limit value, mark the video to be uploaded as a safe and reasonable video, calculate the difference between the total number of security-sensitive videos and the total number of safe and reasonable videos, and obtain the edge control value of the control data source.
[0054] The content security analysis value of the video to be uploaded is obtained in the following way: a preset frame extraction interval is set (the interval length of the preset frame extraction interval, that is, the number of video frames between two video frames, can be set according to the threshold adjustment requirements), and the video to be uploaded is subjected to frame extraction based on the preset frame extraction interval to obtain multiple video frames, and all video frames are sorted in chronological order, and two adjacent video frames after sorting are grouped together and marked as a content security analysis group, and the group security value SJm of each content security analysis group is obtained, and the group security value coefficient is set to PCw, w=1, 2, 3, ...w, PC1<PC2<...<PCw, and each group security value coefficient PC w corresponds to a range of group security value SJm. The range of group security value SJm includes (0, SJ1], (SJ1, SJ2], …, (SJM-1, SJM], when SJm∈(SJ1, SJ2], the group security value coefficient is PC2, set the group security limit value (system threshold, modified according to actual needs), when the group security value is greater than or equal to the group security limit value, mark the content security analysis group as a content sensitive group, when the group security value is less than the group security limit value, mark the content security analysis group as a content normal group, mark the total number of content sensitive groups as KDg, and the total number of content normal groups as ZMt, using the formula The content security analysis value SLb of the video to be uploaded is obtained, where m=1, 2, ..., M, and M is the total number of content security analysis groups.
[0055] The group security value of the content security analysis group is obtained in the following way: the two video frames in the content security analysis group are marked as the starting video frame and the ending video frame respectively (the starting video frame is the video frame with the front ranking, and the ending video frame is the video frame with the back ranking), the word vector of the starting video frame and the word vector of the ending video frame are obtained, the audio between the starting video frame and the ending video frame in the uploaded video is intercepted to obtain the security group audio segment, the audio signal of the security group audio segment is extracted (the features extracted from the audio signal include but are not limited to spectrum features, time domain features, and Mel-frequency cepstral coefficients), and the audio segment signal features are obtained, the image features of the starting video frame and the ending video frame are extracted to obtain the image features of the starting video frame and the ending video frame, the text sensitivity times SKd of the content security analysis group are obtained, the content evaluation value WSr of the security group audio segment is obtained, and the image security value TAZ of the content security analysis group is obtained. The group security value SJm of the content security analysis group is obtained, where s1 is the text sensitivity coefficient, and the value of s1 is 1.38.
[0056] The word vector is obtained in the following way: extract the text in the video frame (the text can be extracted through image text recognition OCR technology), preprocess the extracted text (the preprocessing method includes but is not limited to word segmentation, stop word removal, and word form restoration, and the specific method is selected according to needs), and obtain the word vector of the preprocessed text through the Word2Vec model.
[0057] The text sensitivity times of the content security analysis group are obtained in the following way: obtain the content security model of the text type, use the word vector of the starting video frame and the word vector of the ending video frame in the content security analysis group as the input data of the text type content security model respectively, output the content evaluation value of the text type of the starting video frame and the content evaluation value of the text type of the ending video frame, set the content evaluation threshold of the text type (system threshold, modified according to actual needs), when the content evaluation value of the text type is greater than or equal to the content evaluation threshold, increase the text sensitivity times by one, and mark the text sensitivity times as SKd (because the content security analysis group only contains the starting video frame and the ending video frame, the text sensitivity times is at most two).
[0058] The content rating value of the security group audio segment is obtained in the following manner: obtain the content security model of the audio type, use the audio segment signal feature as input data of the audio type content security model, and output the content rating value WSr of the security group audio segment.
[0059] The image safety value of the content safety analysis group is obtained in the following manner: obtaining a content safety model of the image type, taking the image features of the starting video frame and the image features of the ending video frame in the content safety analysis group as input data of the content safety model of the image type, outputting a content evaluation value of the image type of the starting video frame and a content evaluation value of the image type of the ending video frame, summing the content evaluation values of the two image types to obtain an image evaluation superposition value RYe, performing difference calculation on the content evaluation values of the two image types to obtain an image evaluation fluctuation value CFp, and using the formula The image safety value TAZ of the content safety analysis group is obtained, where p1 is the image evaluation overlap coefficient, p2 is the image evaluation fluctuation coefficient, the value of p1 is 0.97, and the value of p2 is 0.65.
[0060] This application involves a total of three types of content security models, namely text type, audio type and image type. Different content security models are used to obtain content evaluation values of different types of content. The construction methods of different content security models are similar. It is only necessary to replace the training data accordingly. The value ranges of the content evaluation values of the three types of content security models are consistent. Therefore, this application does not introduce the construction methods of the three types of content security models one by one. Only the construction method of one type of content security model is introduced, and the construction methods of the other types of content security models can be known by the same logic. The following takes the text type as an example to introduce the construction method of the text type content security model. Formula: Get a group of word vectors (if it is audio type, get a group of audio segment signal features; if it is image type, get a group of image features). The word vectors can be actually obtained or virtually set. Use a group of word vectors as training data for the neural network model, assign content evaluation values to the training data, divide the training data into training set and verification set in a ratio of 1:3, perform neural network iterative training on the training set and verification set, and obtain a content security model of text type; the value range of the content evaluation value is (5~10), the closer the content evaluation value is to 10, the more sensitive the text content is, and the closer the content evaluation value is to 5, the less sensitive the text content is.
[0061] The above formulas are all dimensionless and numerical calculations are performed, and the preset parameters in the formulas are set by technicians in this field according to actual conditions.
[0062] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.
[0063] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0064] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0065] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0066] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0067] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.
[0068] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A content security AI edge management system based on neural network, characterized in that: It includes source content security control module, security control analysis module, and security control allocation module; The source-end content security control module is used to obtain the edge control range of the edge device, and obtain the control data source based on the edge control range, periodically perform security analysis on the control data source, and obtain the edge control value of the control data source after performing security analysis on the videos to be uploaded at each control data source; The security control analysis module controls the edge device to perform security analysis on the videos to be uploaded from all control data sources, obtain the control allocation value of the edge device, and determine whether to mark the edge device as an allocated edge device based on the comparison result between the control allocation value and the control allocation threshold, and obtain the allocated data source after the edge device is marked as an allocated edge device; The security control allocation module is used to allocate the allocated data source end to the allocated device, and control the allocated device to subsequently perform security analysis on the video to be uploaded at the allocated data source end; The edge control value of the control data source end is obtained in the following manner: obtain all videos to be uploaded on the control data source end within a preset time period, obtain the content security analysis value of each video to be uploaded, set the content security analysis limit value, when the content security analysis value is greater than or equal to the content security analysis limit value, mark the video to be uploaded as a security-sensitive video, and prevent the security-sensitive video from being uploaded; when the content security analysis value is less than the content security analysis limit value, mark the video to be uploaded as a safe and reasonable video, calculate the difference between the total number of security-sensitive videos and the total number of safe and reasonable videos, and obtain the edge control value of the control data source end; The content security analysis value of the video to be uploaded is obtained in the following way: a preset frame extraction interval is set, frame extraction is performed on the video to be uploaded based on the preset frame extraction interval to obtain multiple video frames, all video frames are sorted in chronological order, two adjacent video frames after sorting are grouped together and marked as a content security analysis group, and the group security value SJm of each content security analysis group is obtained, the group security value coefficient is set to PCw, and the group security limit value is set. When the group security value is greater than or equal to the group security limit value, the content security analysis group is marked as a content sensitive group. When the group security value is less than the group security limit value, the content security analysis group is marked as a content normal group. The total number of content sensitive groups is marked as KDg, and the total number of content normal groups is marked as ZMt. Using the formula Obtaining a content security analysis value SLb of the video to be uploaded, where m=1, 2, ..., M, and M is the total number of content security analysis groups; The group security value of the content security analysis group is obtained in the following way: the two video frames in the content security analysis group are marked as the starting video frame and the ending video frame respectively, the word vector of the starting video frame and the word vector of the ending video frame are obtained, the audio between the starting video frame and the ending video frame in the uploaded video is intercepted to obtain the security group audio segment, the audio signal of the security group audio segment is extracted to obtain the audio segment signal feature, the image feature of the starting video frame and the ending video frame is extracted to obtain the image feature of the starting video frame and the image feature of the ending video frame, the text sensitivity times SKd of the content security analysis group are obtained, the content evaluation value WSr of the security group audio segment is obtained, and the image security value TAZ of the content security analysis group is obtained. Get the group security value SJm of the content security analysis group, where s1 is the text sensitivity coefficient; The word vector is obtained by: extracting the text in the video frame, preprocessing the extracted text, and obtaining the word vector of the preprocessed text through the Word2Vec model; The text sensitivity times of the content security analysis group are obtained in the following manner: obtaining a content security model of the text type, using the word vector of the starting video frame and the word vector of the ending video frame in the content security analysis group as input data of the text type content security model, outputting a content evaluation value of the text type of the starting video frame and a content evaluation value of the text type of the ending video frame, setting a content evaluation threshold of the text type, and when the content evaluation value of the text type is greater than or equal to the content evaluation threshold, increasing the text sensitivity times by one, and marking the text sensitivity times as SKd; The content rating value of the safety group audio segment is obtained by: obtaining a content security model of the audio type, using the audio segment signal feature as input data of the audio type content security model, and outputting the content rating value WSr of the safety group audio segment; The image safety value of the content safety analysis group is obtained in the following manner: obtaining a content safety model of the image type, taking the image features of the starting video frame and the image features of the ending video frame in the content safety analysis group as input data of the content safety model of the image type, outputting a content evaluation value of the image type of the starting video frame and a content evaluation value of the image type of the ending video frame, summing the content evaluation values of the two image types to obtain an image evaluation superposition value RYe, performing difference calculation on the content evaluation values of the two image types to obtain an image evaluation fluctuation value CFp, and using the formula The image safety value TAZ of the content safety analysis group is obtained, wherein p1 is the image evaluation overlap coefficient and p2 is the image evaluation fluctuation coefficient; The control allocation value of the edge device is obtained by: obtaining the edge control value of each control data source, summing up the edge control values of all control data sources, and obtaining the control allocation value; Based on the comparison result between the control allocation value and the control allocation threshold, determine whether the edge device needs to be marked as an allocated edge device. After the edge device is marked as an allocated edge device, obtain the allocated data source end, specifically: set the control allocation threshold. When the control allocation value is greater than or equal to the control allocation threshold, mark the edge device as an allocated edge device, and mark the control data source end with the largest edge control value in the allocated edge device as the allocated data source end. When the control allocation value is less than the control allocation threshold, no processing is performed; Assign the allocated data source end to the designated device, and control the designated device to perform security analysis on the videos to be uploaded from the allocated data source end. Specifically, take the position of the allocated data source end as the center, set the radius R to draw a circle, and obtain the edge allocation range. Mark the remaining edge devices within the edge allocation range as target allocation devices, mark the target allocation device with the smallest control allocation value as the designated device, and control the designated device to perform security analysis on the videos to be uploaded from the allocated data source end. When the control allocation value is less than the control allocation threshold, no processing will be performed.
2. According to the neural network-based content security AI edge management system of claim 1, it is characterized in that: Obtain the edge control range of the edge device, and obtain the controlled data source based on the edge control range. Specifically, control the edge device to draw a circle with its own position as the center and set the radius r to obtain the edge control range, and mark the data source within the edge control range as the controlled data source.
3. According to the neural network-based content security AI edge management system of claim 1, it is characterized in that: Periodically perform security analysis on the control data source, specifically: sort the control data source in descending order according to the value of the control primary value, set a preset time period, and perform security analysis on the videos to be uploaded on the control data source in the sorted order at each time node of the preset time period; The primary control value of the control data source is obtained in the following way: obtain the i edge control values obtained by the control data source before the current time of the system, sort all edge control values in the order of acquisition time, calculate the difference between the two adjacent edge control values after sorting and take the absolute value to obtain the edge control turbulence value, sum up all edge control turbulence values and take the average to obtain the control turbulence mean Bz, sum up all edge control values and take the average to obtain the edge control mean Ec, and use the formula The primary control value Mx of the control data source is obtained, where u1 is the control turbulence mean coefficient and u2 is the edge control mean coefficient.
4. A content security AI edge management method based on a neural network, applied to a content security AI edge management system based on a neural network as described in claim 1, characterized in that: The steps include: Step 1: Obtain the edge control scope of the edge device, and obtain the control data source based on the edge control scope; Step 2: Periodically perform security analysis on the control data source; Step 3: After performing security analysis on the videos to be uploaded at each control data source, obtain the edge control value of the control data source; Step 4: After the edge device performs security analysis on the videos to be uploaded from all control data sources, the control edge device obtains the control allocation value of the edge device; Step 5: Based on the comparison result between the control allocation value and the control allocation threshold, determine whether the edge device needs to be marked as an allocated edge device, and after the edge device is marked as an allocated edge device, obtain the allocated data source end; Step 6: Allocate the allocated data source to the assigned device, and control the assigned device to subsequently perform security analysis on the video to be uploaded from the allocated data source.