Safe broadcast intelligent monitoring and early warning method, device and system based on AI pre-generated data model comparison

Abstract

The invention discloses a safe broadcast intelligent monitoring and early warning method, device and system based on AI pre-generated data model comparison. The method comprises the following steps: establishing a reference library containing an expected data model; collecting broadcast signals in real time to generate a same-dimension real-time broadcast data model; performing layered comparison on a data layer and a content layer of the two models according to timestamps; and if the comparison result exceeds the preset threshold, judging that the alarm is abnormal and triggering the alarm. According to the invention, upgrading from signal monitoring to content intelligent understanding and monitoring, instant fault discovery and accurate monitoring and judgment are realized, the false alarm and missing alarm rate is greatly reduced, the manual pressure is relieved, closed-loop intelligent guarantee of the whole broadcast process is realized, and the method is suitable for broadcast signal monitoring of broadcast television and new media.

Description

Technical Field

[0001] This invention relates to the field of broadcast television security technology, specifically to a method, device, and system for intelligent monitoring and early warning of broadcast security based on AI-pre-generated data model comparison. Background Technology

[0002] Safe broadcasting of radio and television is a core task of the broadcasting industry, and signal monitoring is a key means to ensure safe broadcasting. Existing traditional signal monitoring methods mainly rely on threshold judgments, such as monitoring physical signal indicators like black screens, still frames, and silence. These methods suffer from high false alarm rates and serious missed detections: for example, wind-blown leaves obscuring the camera lens can be mistakenly interpreted as a black screen, while a genuine still frame may be missed due to the static image. Furthermore, traditional monitoring methods can only achieve shallow physical and transmission layer monitoring, unable to understand whether the broadcast content is correct, and lack content recognition and logical judgment capabilities.

[0003] Existing technologies also include multi-channel signal comparison monitoring methods, such as main and backup signal comparison. However, this method can only detect faults in the signal transmission process and cannot solve the problem of discrepancies between the broadcast content and the pre-arranged program content. In other words, it cannot proactively detect misbroadcast anomalies such as "content A to be broadcast, content B to be broadcast" and the monitoring results lack correlation with the expected broadcast content, relying excessively on the operator's subjective judgment.

[0004] In addition, existing technologies include alarm optimization systems and emergency handling systems designed for safe broadcasting. These systems are all post-fault handling solutions, which can only detect anomalies and perform subsequent processing after faults such as signal interruption, still frames, and black screens occur. At this time, the fault has already actually occurred and affected the audience. They cannot fundamentally solve the problem of monitoring lag and make it difficult to predict and handle broadcast faults in advance.

[0005] In summary, existing broadcast and television security monitoring technologies suffer from technical problems such as monitoring lag, lack of content recognition capabilities, high false alarm and missed alarm rates, and excessive reliance on manual judgment. There is an urgent need for a monitoring and early warning solution that can achieve intelligent content understanding, real-time prediction, and accurate alarm. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method, device, and system for intelligent monitoring and early warning of safe broadcasting based on AI pre-generated data model comparison. It solves the problems of monitoring lag and identification of logical errors in content in existing technologies, and realizes a technological leap from "signal monitoring" to "intelligent content understanding and monitoring".

[0007] This invention discloses a method for intelligent monitoring and early warning of safe broadcasting based on AI pre-generated data model comparison, including: Step 1: Establish AI pre-generated benchmark data model library: Based on the broadcast schedule, retrieve the corresponding broadcast material library files, extract the video keyframe feature values ​​and audio feature values ​​of the material files, and combine the timestamp of the schedule and the expected program identifier to generate the expected data model for each broadcast period. At the same time, generate abnormal broadcast data models such as still frames, black screens, and padding. Step 2, Real-time signal acquisition and feature analysis: During the program broadcast, the broadcast audio and video signals are acquired in real time, and video feature values ​​and audio feature values ​​of the same dimension as the expected data model are extracted. Combined with the real-time broadcast timestamp, a real-time broadcast data model is generated. Step 3, Data Model Comparison: The real-time broadcast data model and the expected data model for the corresponding time period are compared layer by layer according to the timestamp. The layer comparison includes data layer comparison and content layer comparison. Step 4, Anomaly Detection and Alarm Triggering: Determine whether the comparison result exceeds the preset fault tolerance threshold. If it does, it is determined to be a broadcast anomaly, triggering an alarm command and executing the operations of anomaly information push and anomaly segment extraction.

[0008] As a further improvement of the present invention, in step 1, the serial list includes broadcast material ID, material editing point, material broadcast time, and channel name information, and the expected program identifier includes station logo and advertising label information.

[0009] As a further improvement of the present invention, in step 1 or step 2... The video feature value is a binary string calculated based on the image content. The extraction process is as follows: the image is reduced to a fixed size, converted to grayscale, and then discrete cosine transform is performed. The 8×8 matrix in the upper left corner of the transformation result is retained, and the average value of the coefficients of the matrix is ​​calculated. The coefficients are compared with the average value and then binarized. The video feature value is obtained by combining them in order. The audio feature value is a hash value generated based on the audio acoustic features. The extraction process is as follows: a short-time Fourier transform is performed on the audio segment to generate a spectrum, energy peak feature points are extracted from the spectrum, and the time and frequency information of the feature points are used to generate audio feature values ​​through a hash function.

[0010] As a further improvement of the present invention, the fixed size of the image reduction is 32×32 or 8×8, and the binary string is 64 bits.

[0011] As a further improvement of the present invention, in step 2, a 100G fiber optic interface is used to realize the real-time acquisition of broadcast audio and video signals, and the feature value extraction rule of the same dimension is completely consistent with the feature value extraction rule of the expected data model in step 1.

[0012] As a further improvement of the present invention, in step 3, the data layer comparison is to compare the technical indicators of bitrate, IP stream information, and field blanking period data; the content layer comparison includes comparing screen similarity, comparing audio features, and comparing the consistency between the identified subtitle text and the expected text using deep hashing algorithms or convolutional neural networks.

[0013] As a further improvement of the present invention, in step 4... The preset fault tolerance thresholds include one or more of the following: image similarity is less than 90%, audio is silent for more than 2 seconds, and the actual advertising time period does not match the pre-model advertising time period. The alarm command includes an audible and visual alarm, and the abnormal information push refers to pushing the abnormal data analysis results to the alarm terminal and the SDN centralized control system. As a further improvement of the present invention, the video feature values ​​have perceptual robustness, with similar images having highly similar feature values ​​and different images having completely different feature values; the audio feature values ​​have robustness, with different versions of the same audio segment generating the same or highly similar feature values.

[0014] This invention also discloses a smart monitoring and early warning device for safe broadcasting based on AI pre-generated data model comparison, used to implement the above method, including: The broadcast control data interface module is used to connect to the broadcast control system and obtain broadcast sequence information and corresponding broadcast files; The AI-based expected model generation service module is used to pre-generate audio and video feature values ​​of materials for each broadcast period based on the broadcast schedule, forming expected data models and abnormal broadcast data models. The AI ​​data model generation service module is used to collect broadcast audio and video signals in real time, generate audio and video feature values ​​of materials for each broadcast period, and form a real-time broadcast data model. The comparison engine and alarm service module is used to perform hierarchical comparison between the real-time broadcast data model and the expected data model for the corresponding time period, and determine whether the comparison result exceeds the preset fault tolerance threshold. If it exceeds the threshold, it is determined to be a broadcast abnormality and an alarm command is triggered, while the monitoring log is recorded. The SDN interface module is used to connect to the SDN centralized control system and report alarm information and corresponding abnormal data analysis results.

[0015] As a further improvement of the present invention, the core processing chip of the device adopts the NVIDIA Jetson T5000 series chip, based on the Blackwell GPU architecture, and supports running mainstream large models such as Llama3, Qwen, and Stable Diffusion.

[0016] As a further improvement of the present invention, the AI ​​expected model generation service module is configured with a feature value extraction unit to extract video keyframe feature values ​​and audio feature values ​​respectively; wherein, the extraction rules of the feature value extraction unit are consistent with the extraction rules described in the above method.

[0017] As a further improvement of the present invention, the AI ​​data model generation service module is equipped with a 100G fiber optic acquisition interface for real-time acquisition of broadcast audio and video signals, and its feature value extraction rules are completely consistent with the feature value extraction rules of the AI ​​expected model generation service module.

[0018] As a further improvement of the present invention, the comparison engine and alarm service module includes a data layer comparison unit, a content layer comparison unit, and an anomaly determination unit. The content layer comparison unit is equipped with the operation logic of a deep hash algorithm and a convolutional neural network. The comparison engine and alarm service module is connected to an alarm terminal, which supports audible and visual alarms, abnormal fragment storage, and abnormal information display functions.

[0019] The present invention also discloses a safe broadcast intelligent monitoring and early warning system based on AI pre-generated data model comparison, including the above-mentioned device, and further including a broadcast control system, an SDN centralized control system and a broadcast material library connected in communication with it; The broadcast control system is used to issue broadcast sequence lists and broadcast control instructions to the intelligent monitoring and early warning device; The broadcast material library is used to store audio and video material files of programs, advertisements, and padding, which can be retrieved by the intelligent monitoring and early warning device; The SDN centralized control system is used to receive alarm information and abnormal data analysis results reported by the intelligent monitoring and early warning device, and realize centralized management and control of the broadcast system.

[0020] As a further improvement of the present invention, the intelligent monitoring and early warning device can be deployed in single or multiple device modes according to the channel scale of the broadcast system. When deployed in single device mode, it integrates file monitoring and stream monitoring functions. When deployed in multiple devices mode, it can respectively realize the generation of expected data model and the generation of real-time broadcast data model.

[0021] As a further improvement of the present invention, the system is constructed with a low-power, high-density multi-channel AI signal monitoring architecture, which supports real-time monitoring and model comparison of multiple channels of audio and video signals simultaneously.

[0022] As a further improvement of the present invention, a high-speed data transmission channel is established between the broadcast material library and the intelligent monitoring and early warning device to ensure rapid retrieval of broadcast material files and extraction of feature values.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses AI to pre-generate expected data models to achieve forward-looking monitoring of "expected comparison". It can detect broadcast anomalies in the first millisecond or even when an abnormal trend appears. In contrast, existing technologies can only detect anomalies by judging thresholds after a fault occurs. This invention fundamentally solves the problem of monitoring lag, allowing sufficient time for on-duty personnel to respond to emergencies and preventing broadcast accidents from escalating.

[0024] 2. This invention breaks away from the shallow mode of traditional signal monitoring that only stays at the physical and transmission layers, and realizes deep monitoring of the data layer and content layer. It can not only detect the presence or absence of signals, but also accurately identify anomalies at the content level such as misplays, still frames, black screens, missing subtitles, audio misalignment, and incorrect advertising. It truly realizes intelligent content understanding that "knows what to play and judges whether the playback is correct".

[0025] 3. The AI ​​model of this invention combines contextual information such as timestamps and material IDs. Through high-precision comparison of audio and video feature values, it can accurately distinguish between "normal still frames (such as end credits)" and "faulty still frames", "normal screen occlusion" and "black screen fault", which greatly reduces the false alarm rate. At the same time, based on the full-dimensional comparison of the content layer, it avoids the problem of missed detection in traditional threshold monitoring and improves the accuracy of fault judgment.

[0026] 3. This invention realizes the transformation from "passive signal reception" to "active prediction and detection". The system can automatically complete the entire process of model generation, signal acquisition, comparison and judgment, and abnormal alarm without the need for subjective judgment by the operator. It liberates the human from tedious real-time monitoring work and only requires emergency response after an alarm, which greatly reduces the workload of the on-duty personnel.

[0027] 4. This invention uses the scheduling data of the broadcast control system as the direct input for the generation of the AI ​​model, connecting the business flow and data flow of "pre-broadcast scheduling" and "in-broadcast monitoring", and realizing closed-loop intelligent protection of the entire broadcast and television broadcast process; at the same time, the integrated hardware and software design of the system is adapted to the multi-channel monitoring needs of the broadcasting industry, providing a feasible technical solution for the intelligent transformation of the broadcasting industry.

[0028] 5. The feature extraction rules and fault tolerance thresholds of this invention can be customized according to the broadcasting needs of the broadcasting industry. Custom abnormal broadcasting data models can be added to support the monitoring of different types of programs such as variety shows, news, advertisements, and TV dramas. At the same time, it can be extended to broadcasting scenarios such as new media and online audiovisual, and has good technical scalability and scenario adaptability. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall process of the intelligent monitoring and early warning method disclosed in this invention; Figure 2This is a schematic diagram of the functional modules of the intelligent monitoring and early warning system of the present invention; Figure 3 This is a schematic diagram of the hardware configuration of the intelligent monitoring and early warning device of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] This invention provides a method, device, and system for intelligent monitoring and early warning of safe broadcasting based on AI-pre-generated data model comparison. It utilizes the broadcast sequence list and material library files of the broadcast control system to pre-generate an expected data model as a judgment benchmark before the broadcast. During the broadcast, signals are collected in real time and a real-time broadcast data model is generated. The broadcast anomalies are immediately determined through precise feature-by-feature comparison of the two-layer model. At the same time, the integrated design of hardware and software realizes intelligent, automated, and precise monitoring and early warning.

[0032] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1 As shown, this invention provides a method for intelligent monitoring and early warning of safe broadcasting based on AI-pre-generated data model comparison, including: S1. Establish an AI pre-generated benchmark data model library: Based on the broadcast schedule (which includes broadcast material ID, material editing point, material broadcast time, channel name, etc.), the corresponding audio and video files in the broadcast material library are retrieved; the video keyframe feature values ​​and audio feature values ​​of the material files are extracted, and the expected data models for each broadcast time slot are generated by combining the timestamp of the schedule and the expected program identifiers (station logo, advertising labels, etc.); at the same time, abnormal broadcast data models of types such as still frames, black screens, and padding are generated in advance to provide a reference for anomaly judgment.

[0033] S2. Real-time signal acquisition and feature analysis: During the broadcast, the audio and video signals are collected in real time through a 100G fiber optic interface. Video and audio feature values ​​of the same dimension are extracted according to the rules that are completely consistent with the expected data model. Combined with the real-time timestamp of the broadcast, a real-time broadcast data model is generated to ensure the effectiveness and accuracy of the model comparison.

[0034] S3, Intelligent Data Model Comparison: The real-time broadcast data model and the expected data model for the corresponding time period are compared layer by layer based on timestamps: Data layer comparison: Compare technical indicators such as bit rate, IP stream information, and field blanking period data to detect anomalies in the signal transmission process; Content layer comparison: Using deep hashing algorithms or convolutional neural networks to compare the similarity of images, compare audio features, and compare the consistency between the identified subtitle text and the expected text, anomalies are detected during the content broadcast process.

[0035] S4. Anomaly Detection and Alarm Triggering: A preset fault tolerance threshold (such as image similarity below 90%, audio silence exceeding 2 seconds, actual advertising time not matching pre-model advertising time, etc.) is set to determine whether the model comparison result exceeds the threshold; if it exceeds the threshold, it is determined to be a broadcast abnormality, and an alarm command is immediately triggered to perform sound and light alarm, abnormal segment interception, and abnormal information push operation. At the same time, the alarm information and abnormal data analysis results are reported to the SDN centralized control system.

[0036] The video feature value is a 64-bit binary string, which has perceptual robustness. The feature values ​​of similar images are highly similar, while the feature values ​​of different images are completely different. The extraction process is as follows: reduce the image to a fixed size of 32×32 / 8×8 → convert it to grayscale → perform discrete cosine transform (DCT) → retain the 8×8 matrix in the upper left corner of the DCT result → calculate the average value of the matrix coefficients → compare the coefficients with the average value and then binarize → combine them in order to obtain the 64-bit video feature value.

[0037] Audio feature values ​​are hash values ​​based on acoustic features, which are robust. Different versions of the same audio segment generate the same or highly similar feature values. The extraction process is as follows: perform a short-time Fourier transform on the audio segment to generate a spectrogram → extract the energy peak feature points on the spectrogram → use the time and frequency information of the feature points to generate audio feature values ​​through a hash function.

[0038] like Figure 2 , 3 As shown, this invention provides a smart monitoring and early warning device for safe broadcasting based on AI pre-generated data model comparison. This monitoring and early warning device is the hardware carrier for implementing the above method, and its functional modules include: Broadcast control data interface module: Serves as a communication bridge between the device and the broadcast control system, enabling rapid acquisition of broadcast sequence information and corresponding broadcast files; AI Expected Model Generation Service Module: Configure feature value extraction unit to pre-generate expected data models and abnormal broadcast data models for each broadcast period based on the broadcast schedule; AI data model generation service module: Configured with a 100G fiber optic acquisition interface and a feature value extraction unit of the same specification, it can acquire audio and video signals in real time and generate real-time broadcast data models. The comparison engine and alarm service module includes a data layer comparison unit, a content layer comparison unit, and an anomaly judgment unit. It executes layered comparison logic, judges broadcast anomalies and triggers alarm commands, and records monitoring logs at the same time. SDN interface module: Enables communication between the device and the SDN centralized control system, and completes the reporting of alarm information and abnormal data analysis results.

[0039] The core processing chip of the device is the NVIDIA Jetson T5000 series. This chip is based on the Blackwell GPU architecture, equipped with high-bandwidth memory and a complete software ecosystem. It can migrate the large model inference capabilities in the cloud to edge devices, support the running of mainstream large models such as Llama3, Qwen, and StableDiffusion, and provide powerful computing support for audio and video feature extraction and model comparison. At the same time, it builds a low-power, high-density multi-channel AI signal monitoring architecture.

[0040] This invention provides a smart monitoring and early warning system for safe broadcasting based on AI pre-generated data model comparison. This monitoring and early warning system is an integrated system architecture, including the aforementioned smart monitoring and early warning device, as well as a broadcast control system, an SDN centralized control system, and a broadcast material library connected to it. Broadcast control system: Provides broadcast sequence lists and broadcast control instructions to the system, and is the data source for generating the expected data model; Broadcast Material Library: Stores all audio and video material files, including programs, advertisements, and background clips, providing raw materials for feature value extraction; SDN centralized control system: Receives information reported by intelligent monitoring and early warning devices to realize centralized management and emergency response of the broadcasting and television system; Intelligent monitoring and early warning device: The core execution unit of the system, realizing the entire process of model generation, signal acquisition, model comparison, and abnormal alarm. Example 1:

[0041] This embodiment uses the monitoring of advertisement broadcasts on radio and television channels as an example to illustrate the specific implementation process of the method of the present invention: Pre-generated expected data model: At 9:00 AM, the AI ​​monitoring system reads the broadcast schedule for the day through the broadcast control data interface module and obtains the information that a 15-second advertisement for brand A will be broadcast at 9:05:00 AM. The system retrieves the complete video file of brand A's advertisement from the broadcast material library based on the advertisement material ID. Extract the 64-bit feature values ​​of the video keyframes and the audio hash feature values ​​of the advertisement, and combine them with the timestamp of 9:05:00-9:05:15 and the A brand advertisement tag to generate the expected data model for this time period. At the same time, pre-generate abnormal broadcast data models such as black screen and still frame.

[0042] Real-time acquisition and model generation: At 9:05:00, when the advertisement is broadcast, the system acquires the broadcast audio and video signals of the channel in real time through the 100G fiber optic interface. According to the rules that are completely consistent with the expected data model, the system extracts the video and audio feature values ​​of the real-time signal and generates a real-time broadcast data model by combining the real-time timestamp.

[0043] Intelligent Model Comparison: The comparison engine performs a layered comparison between the real-time broadcast data model and the expected data model of Brand A's advertisement. The data layer comparison results show that the bitrate and IP stream information are normal. The content layer comparison results show that the audio-visual feature values ​​of the real-time signal match the expected feature values ​​of Brand A's advertisement by only 30%, while the match rate with the feature values ​​of Brand B's advertisement is as high as 95%.

[0044] Anomaly detection and alarm: If the system determines that the comparison result exceeds the preset 90% image similarity fault tolerance threshold, it will immediately determine it as a broadcast anomaly of "advertisement misplay"; trigger an audio and visual alarm, and at the same time capture the audio and video segments of the abnormal period, push the abnormal information of "advertisement misplay, real-time features highly match with Brand B's advertisement" to the alarm terminal of the on-duty personnel, and report it to the SDN centralized control system through the SDN interface module.

[0045] Emergency Response: The anomaly was detected only 0.5 seconds after the advertisement was broadcast. Upon receiving the alarm, the on-duty personnel quickly intervened, switched the signal, and prevented the broadcast accident from escalating, thus ensuring broadcast safety.

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring and early warning of safe broadcasting based on AI pre-generated data model comparison, characterized in that, include: Step 1: Based on the broadcast schedule, retrieve the corresponding broadcast material library file, extract the video keyframe feature values ​​and audio feature values ​​of the material file, and combine the timestamp of the schedule and the expected program identifier to generate the expected data model for each broadcast period. Step 2: During the program broadcast, the broadcast audio and video signals are collected in real time, and the video feature values ​​and audio feature values ​​of the same dimension as the expected data model are extracted. The real-time broadcast data model is generated by combining the real-time broadcast timestamp. Step 3: Compare the real-time broadcast data model with the expected data model for the corresponding time period in layers based on timestamps. The layer comparison includes data layer comparison and content layer comparison. Step 4, Anomaly Detection and Alarm Triggering: Determine whether the comparison result exceeds the preset fault tolerance threshold. If it does, it is determined to be a broadcast anomaly, triggering an alarm command and executing the operations of anomaly information push and anomaly segment extraction.

2. The method according to claim 1, characterized in that, In step 1, the serial list includes broadcast material ID, material editing point, material broadcast time, and channel name information, and the expected program identifier includes station logo and advertising label information.

3. The method according to claim 1, characterized in that, In step 1 or step 2, The video feature value is a binary string calculated based on the image content. The extraction process is as follows: the image is reduced to a fixed size, converted to grayscale, and then discrete cosine transform is performed. The 8×8 matrix in the upper left corner of the transformation result is retained, and the average value of the coefficients of the matrix is ​​calculated. The coefficients are compared with the average value and then binarized. The video feature value is obtained by combining them in order. The audio feature value is a hash value generated based on the audio acoustic features. The extraction process is as follows: a short-time Fourier transform is performed on the audio segment to generate a spectrum, energy peak feature points are extracted from the spectrum, and the time and frequency information of the feature points are used to generate audio feature values ​​through a hash function.

4. The method according to claim 1, characterized in that, In step 3, the data layer comparison involves comparing technical indicators such as bitrate, IP stream information, and field blanking period data; the content layer comparison includes comparing screen similarity using deep hashing algorithms or convolutional neural networks, comparing audio feature values, and comparing the consistency between the identified subtitle text and the expected text.

5. The method according to claim 1, characterized in that, In step 4, The preset fault tolerance thresholds include one or more of the following: image similarity is less than 90%, audio is silent for more than 2 seconds, and the actual advertising time period does not match the pre-model advertising time period. The alarm commands include audible and visual alarms, and the abnormal information push refers to pushing the abnormal data analysis results to the alarm terminal and the SDN centralized control system.

6. A smart monitoring and early warning device for safe broadcasting based on AI pre-generated data model comparison, used to implement the method as described in any one of claims 1 to 5, characterized in that, include: The broadcast control data interface module is used to connect to the broadcast control system and obtain broadcast sequence information and corresponding broadcast files; The AI-based expected model generation service module is used to pre-generate audio and video feature values ​​of materials for each broadcast period based on the broadcast schedule, forming expected data models and abnormal broadcast data models. The AI ​​data model generation service module is used to collect broadcast audio and video signals in real time, generate audio and video feature values ​​of materials for each broadcast period, and form a real-time broadcast data model. The comparison engine and alarm service module is used to perform hierarchical comparison between the real-time broadcast data model and the expected data model for the corresponding time period, and determine whether the comparison result exceeds the preset fault tolerance threshold. If it exceeds the threshold, it is determined to be a broadcast abnormality and an alarm command is triggered, while the monitoring log is recorded. The SDN interface module is used to connect to the SDN centralized control system and report alarm information and corresponding abnormal data analysis results.

7. The apparatus according to claim 6, characterized in that, The core processing chip of the device uses the NVIDIA Jetson T5000 series chip, which is based on the Blackwell GPU architecture and supports running mainstream large models such as Llama3, Qwen, and StableDiffusion.

8. The apparatus according to claim 7, characterized in that, The comparison engine and alarm service module includes a data layer comparison unit, a content layer comparison unit, and an anomaly determination unit. The content layer comparison unit is equipped with deep hashing algorithm and convolutional neural network operation logic. The comparison engine and alarm service module is connected to an alarm terminal, which supports audible and visual alarms, abnormal fragment storage, and abnormal information display functions.

9. A smart monitoring and early warning system for safe broadcasting based on AI-pre-generated data model comparison, characterized in that, The apparatus includes the device according to any one of claims 6-8, and further includes a broadcast control system, an SDN centralized control system, and a broadcast material library that are communicatively connected thereto; The broadcast control system is used to issue broadcast sequence lists and broadcast control instructions to the intelligent monitoring and early warning device; The broadcast material library is used to store audio and video material files of programs, advertisements, and padding, which can be retrieved by the intelligent monitoring and early warning device; The SDN centralized control system is used to receive alarm information and abnormal data analysis results reported by the intelligent monitoring and early warning device, and realize centralized management and control of the broadcast system.

10. The system according to claim 9, characterized in that, The intelligent monitoring and early warning device can be deployed in single or multiple device modes according to the channel scale of the broadcast system. When deployed in single device mode, it integrates file monitoring and stream monitoring functions. When deployed in multiple devices mode, it can generate expected data models and real-time broadcast data models respectively.

Images