Aerial object ai video monitoring method and system for complex environment of nuclear power

Abstract

The application provides a high-altitude falling object AI video monitoring method and system for a complex nuclear power environment. First, the application obtains double-source video streams synchronously collected by a radiation-resistant camera and a thermal imager in a nuclear power operation area and identifies radiation interference frames. Based on the identification result, the application performs a radiation noise compensation operation on the video stream collected by the radiation-resistant camera. Second, the application performs a high-temperature steam distortion correction operation on the video stream collected by the thermal imager. Third, the application inputs the video streams after the radiation noise compensation operation and the high-temperature steam distortion correction operation into a pre-adapted model. Finally, when a dynamic target criterion meets a preset high-altitude falling object behavior mode, an alarm protocol is triggered. The technical solution provided by the application not only realizes full closed-loop management and control of nuclear power falling object risks through intelligent behavior matching, hierarchical response mechanisms, and precise instruction distribution, but also forms a complete safety closed loop of "monitoring-judgment-braking-isolation-alarm".

Description

Technical Field

[0001] This application relates to the field of intelligent nuclear power plant safety monitoring and multimodal perception fusion technology, and in particular to an AI video monitoring method and system for high-altitude falling objects in the complex environment of nuclear power plants. Background Technology

[0002] During the construction, maintenance, and operation of nuclear power plants, high-altitude work areas (such as reactor building domes, large equipment hoisting areas, and pipeline installation layers) are extremely high-risk locations. These areas not only pose a risk of personnel falling from heights, but more importantly, the accidental falling of tools, components, or debris can have serious consequences. Therefore, this technology must be able to overcome adverse conditions such as image noise interference caused by strong gamma-ray radiation and thermal imaging distortion caused by high-temperature steam, enabling real-time and accurate identification and early warning of falling object events, in order to meet the stringent requirements of nuclear power safety regulations for unmanned and intelligent monitoring of high-risk work areas.

[0003] Current research on monitoring falling objects from heights in nuclear power plant environments has explored deploying moving target analysis systems based on thermal imaging video. These systems primarily rely on thermal imagers to capture the difference in thermal radiation between objects and their environment, using background modeling and motion trajectory analysis algorithms to identify falling objects. Their advantages include insensitivity to changes in light, the ability to operate in low-light or smoky environments, and the ability to penetrate some steam interference to identify moving objects with significant temperature differences.

[0004] However, existing single-modal thermal imaging-based solutions have three significant drawbacks in the complex environment of nuclear power plants. First, the gamma rays generated by the high-intensity radiation field of nuclear power plants directly interfere with thermal imaging sensors, causing random noise or even large-area distortion in the images, severely affecting the accuracy of target detection. Second, the high-temperature steam jets commonly present in nuclear power plant operations cause severe distortion of the background thermal field distribution due to their thermal diffusion effect, making the thermal outlines of moving targets (especially objects with temperatures close to the environment or small in size) blurry or even completely obscured, resulting in a large number of false alarms or missed alarms. Third, single thermal imaging data is difficult to effectively distinguish between real falling objects and interference sources such as steam flow and thermal radiation fluctuations. Under the dual interference of radiation and steam, the constructed motion trajectories are often discontinuous and lack accuracy, failing to meet the stringent requirements of nuclear power safety for early warning reliability. Summary of the Invention

[0005] This application provides an AI video monitoring method and system for high-altitude falling objects in the complex environment of nuclear power plants, which solves the problems in the prior art that cannot overcome the direct interference of gamma rays on imaging sensors, are difficult to correct imaging distortion caused by thermal diffusion of high-temperature steam, and are prone to false alarms and missed alarms due to the lack of multimodal data fusion capabilities.

[0006] Firstly, this application provides an AI video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants, including:

[0007] Acquire dual-source video streams simultaneously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include the video stream collected by the radiation-resistant camera and the video stream collected by the thermal imager;

[0008] The dual-source video stream is subjected to radiated interference frames. Based on the identification results, a radiated noise compensation operation is performed on the video stream acquired by the radiation-resistant camera to generate a video stream after the radiated noise compensation operation.

[0009] Simultaneously, based on the recognition results, a high-temperature steam distortion correction operation is performed on the video stream acquired by the thermal imager to generate a video stream after the high-temperature steam distortion correction operation;

[0010] The video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation are input into a pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markers.

[0011] Based on the three-dimensional motion trajectory field, a dynamic target criterion constrained by nuclear power safety regulations is established. When the dynamic target criterion conforms to the preset high-altitude falling object behavior pattern, an alarm protocol covering multiple levels of safety control units is triggered.

[0012] Optionally, a dual-source video stream is acquired simultaneously from a radiation-resistant camera and a thermal imager in the nuclear power plant operation area. The dual-source video stream includes a video stream acquired by the radiation-resistant camera and a video stream acquired by the thermal imager, comprising:

[0013] According to the preset nuclear power safety layout diagram, the radiation-resistant camera and the thermal imager are deployed at the high-altitude operation hazard points in the nuclear power operation area;

[0014] Connect the radiation-resistant camera and the thermal imager to the nuclear power plant time synchronization controller, and use the nuclear power plant time synchronization controller to send a unified start signal;

[0015] In response to the unified start signal, the acquisition functions of the radiation-resistant camera and the thermal imager are activated simultaneously;

[0016] During the acquisition process, the radiation-resistant camera dynamically adjusts its exposure parameters according to the changes in radiation levels at its location, and the thermal imager adjusts its thermal sensitivity according to the surrounding steam temperature distribution to output a dual-source video stream, wherein the dual-source video stream is bound to a time stamp sequence.

[0017] Optionally, the dual-source video stream is subjected to radiated interference frame identification, and based on the identification result, radiated noise compensation is performed on the video stream acquired by the radiation-resistant camera to generate a video stream after radiated noise compensation, including:

[0018] Based on the time stamp sequence of the dual-source video stream, the video frames at the corresponding time points in the video stream acquired by the radiation-resistant camera are identified as candidate frames.

[0019] Detect the accumulated noise of pixel units caused by gamma rays in the candidate frame, and count the accumulated noise density of pixel units that exceeds the normal threshold of the visible light band from the detection results;

[0020] When the accumulated noise density of the pixel unit continuously reaches the radiative interference threshold and exceeds the preset number of frames, it is determined to be a radiative interference frame, and the total amount of radiative noise corresponding to the radiative interference frame is recorded.

[0021] Based on the radiated interference frame and the corresponding total amount of radiated noise, a compensation weight inversely proportional to the total amount of radiated noise is assigned to the radiated interference frame.

[0022] The pixel unit values ​​of the radiated interference frame are adjusted using the compensation weights to obtain the video stream after radiated noise compensation.

[0023] Optionally, the pixel unit values ​​of the radiated interference frame are adjusted using the compensation weights to obtain the video stream after radiated noise compensation, including:

[0024] Based on the spatial distribution relationship of pixel units in the radiated interference frame, the compensation region adjacent to the radiated noise points above the preset threshold is determined.

[0025] Extract the reference value of the pixel unit that is not affected by radiation interference within the compensation area;

[0026] The compensation weights are decomposed into horizontal and vertical weight components that are related to the diffusion direction of the radiated noise.

[0027] The grayscale shift of the pixel unit is suppressed proportionally along the horizontal axis according to the horizontal weight component, and the grayscale shift of the pixel unit is suppressed proportionally along the vertical axis according to the vertical weight component, so as to obtain the pixel unit value after horizontal and vertical suppression.

[0028] The pixel values ​​that have undergone lateral and longitudinal suppression are replaced with the pixel reference values ​​within the compensation area to obtain the video stream after radiative noise compensation.

[0029] Optionally, based on the recognition results, a high-temperature steam distortion correction operation is performed on the video stream acquired by the thermal imager to generate a video stream after the high-temperature steam distortion correction operation, including:

[0030] Based on the time stamp of the radiation interference frame, the video frame at the corresponding moment in the video stream acquired by the thermal imager is located as the steam-affected frame.

[0031] The high-temperature steam region and the isothermal background region in the steam-affected frame are divided;

[0032] The temperature distribution profile of the isothermal background region is extracted as a reference thermal field;

[0033] The thermal diffusion coverage coefficient of high-temperature steam on the background target is calculated based on the angle between the steam jet direction and the optical axis of the thermal imager.

[0034] Based on the thermal diffusion coverage coefficient, the pixel unit thermal field transmission value of the high-temperature steam region is adjusted in reverse to form a thermal field after radiation attenuation correction in the temperature distribution of the high-temperature steam region, so as to output the video stream after high-temperature steam distortion correction operation.

[0035] Optionally, the video stream after the radiation noise compensation operation and the video stream after the high-temperature steam distortion correction operation are input into a pre-trained nuclear power plant feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markers, including:

[0036] The pixel units of the video stream after the radiation noise compensation operation are bound to radiation dose markers to obtain video frames with radiation dose markers.

[0037] The thermal field markers of the video stream after the high-temperature steam distortion correction operation are superimposed with spatial gradient coordinates to obtain a video frame with thermal field gradient markers.

[0038] In the nuclear power characteristic adaptation model, the thermal radiation conflict index is calculated for the video frames with radiation dose markings and the video frames with thermal field gradient markings.

[0039] When the thermal radiation conflict index in the solution result is lower than the preset nuclear power environment conflict threshold, the video frame with radiation dose marker and the video frame with thermal field gradient marker are fused to generate a composite marker video frame.

[0040] Based on the displacement vector changes of the markers in the composite marker video frames, the depth positions in the nuclear power coordinate system are connected frame by frame to construct the three-dimensional motion trajectory field containing the radiation region markers.

[0041] Optionally, a dynamic target criterion constrained by nuclear power safety regulations is established based on the three-dimensional motion trajectory field. When the dynamic target criterion conforms to a preset high-altitude falling object behavior pattern, an alarm protocol covering multiple levels of safety control units is triggered, including:

[0042] The depth position change of the three-dimensional motion trajectory field is subjected to ultra-rapid descent feature extraction to generate a set of key parameters for dynamic target discrimination;

[0043] The set of key parameters is matched with the nuclear power equipment damage coefficient in the preset high-altitude falling object behavior pattern to verify the effectiveness of the dynamic target criterion and obtain the verification results.

[0044] When the set of key parameters in the verification results continuously satisfies the radiation acceleration constraint condition in the high-altitude falling object behavior mode, the multi-level control engine of the preset alarm protocol is activated.

[0045] The multi-level control engine generates a response link for a multi-level safety control unit, wherein the response link includes an emergency braking unit, a containment isolation unit, and a central control unit that are bound to the nuclear power safety classification matrix.

[0046] The trajectory coordinate sequence and safety risk label are extracted from the dynamic target criteria to construct a directional alarm instruction set;

[0047] The directional alarm instruction set is distributed to the corresponding level control unit through the response link to execute a closed-loop response.

[0048] Secondly, this application provides an AI video monitoring system for falling objects from heights in the complex environment of nuclear power plants, comprising:

[0049] The acquisition module is used to acquire dual-source video streams synchronously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include the video streams collected by the radiation-resistant camera and the video streams collected by the thermal imager.

[0050] The compensation module is used to identify radiated interference frames in the dual-source video stream, and perform radiated noise compensation operation on the video stream acquired by the radiation-resistant camera based on the identification results, thereby generating a video stream after radiated noise compensation.

[0051] The correction module is used to simultaneously perform high-temperature steam distortion correction on the video stream acquired by the thermal imager based on the recognition results, and generate a video stream after high-temperature steam distortion correction.

[0052] The construction module is used to input the video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation into the pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markings.

[0053] The triggering module establishes dynamic target criteria constrained by nuclear power safety regulations based on the three-dimensional motion trajectory field. When the dynamic target criteria meet the preset high-altitude falling object behavior pattern, it triggers an alarm protocol covering multiple levels of safety control units.

[0054] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the AI ​​video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants as described in the first aspect above.

[0055] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements the AI ​​video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants as described in the first aspect.

[0056] This application simultaneously acquires dual-source video streams from a radiation-resistant camera and a thermal imager, and performs precise identification and compensation corrections (radiation noise compensation operation and high-temperature steam distortion correction operation) for gamma-ray radiation interference and high-temperature steam distortion unique to the nuclear power environment, effectively overcoming the perception limitations of a single sensor in complex nuclear power environments. By inputting the corrected dual-source video streams into a nuclear power feature adaptation model for fusion processing, a three-dimensional motion trajectory field containing radiation area marking information can be constructed, significantly improving the identification accuracy and trajectory reconstruction capability of moving targets (especially falling objects from high altitudes) under conditions of strong radiation and steam interference.

[0057] Furthermore, in defining the triggering and execution mechanism of the alarm protocol, the accuracy and reliability of the alarm criteria are ensured by extracting the overspeed descent features of the three-dimensional motion trajectory field and rigorously matching and verifying them with preset high-altitude falling object behavior patterns (including nuclear power equipment damage coefficients and radiation-induced falling acceleration constraints), effectively avoiding false triggers. Secondly, after successful verification, the system can automatically activate a preset multi-level control engine, intelligently generating and executing a hierarchical response link including the emergency braking unit, containment isolation unit, and central control unit. By extracting key information (trajectory coordinate sequence and safety risk labels) from the dynamic target criteria to construct a directional alarm instruction set, and accurately distributing it to the corresponding level of control unit via the response link to execute a closed-loop response, the speed, accuracy, and systematic nature of nuclear power safety event response are greatly improved, providing strong technical support for the safety protection of high-risk areas in nuclear power plants.

[0058] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 The flowchart of an AI video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants provided in this application is shown.

[0061] Figure 2 This paper presents a schematic diagram of the structure of an AI video monitoring system for high-altitude falling objects in the complex environment of a nuclear power plant, as provided in this application.

[0062] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0063] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0064] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0065] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] Figure 1 This application provides a flowchart of an AI video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants, as shown in the following diagram. Figure 1 As shown, the method includes:

[0067] Step 101: Obtain dual-source video streams synchronously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include the video stream collected by the radiation-resistant camera and the video stream collected by the thermal imager.

[0068] Optionally, step 101 may specifically include the following steps:

[0069] Step 1011: According to the preset nuclear power safety layout diagram, deploy the radiation-resistant camera and the thermal imager at the high-altitude work hazard points in the nuclear power operation area;

[0070] Step 1012: Connect the radiation-resistant camera and the thermal imager to the nuclear power plant time synchronization controller, and use the nuclear power plant time synchronization controller to send a unified start signal;

[0071] Step 1013: In response to the unified start signal, the acquisition functions of the radiation-resistant camera and the thermal imager are activated simultaneously.

[0072] Step 1014: During the acquisition process, the radiation-resistant camera dynamically adjusts the exposure parameters according to the changes in the radiation level at its location, and the thermal imager adjusts the thermal sensitivity according to the surrounding steam temperature distribution to output a dual-source video stream, wherein the dual-source video stream is bound to a time stamp sequence.

[0073] In the above scheme, the dual-source video stream refers to a synchronized data stream composed of a visible light video stream acquired by a radiation-resistant camera and an infrared thermal imaging video stream acquired by a thermal imager. This stream is used to complementaryly capture the optical and thermodynamic characteristics of target objects in the complex environment of a nuclear power plant. The radiation-resistant camera is a specially designed camera device that resists gamma-ray radiation interference and adapts to changes in radiation levels by dynamically adjusting exposure parameters, ensuring stable imaging in areas with high radiation. The thermal imager is a device that generates a temperature distribution image by detecting the infrared radiation of an object. It can automatically adjust its thermal sensitivity based on the surrounding steam temperature distribution and can penetrate smoke but is susceptible to thermal diffusion from high-temperature steam. The preset nuclear power plant safety layout diagram is a safety monitoring point deployment map drawn based on hazardous points for high-altitude operations in nuclear power plants (such as the reactor building roof and large equipment hoisting areas), used to guide equipment installation locations. The nuclear power plant time synchronization controller is a dedicated clock synchronization device that ensures complete alignment of the time signature sequences of the dual-source video streams by issuing a unified start signal, providing a foundation for subsequent frame-level data fusion. The unified start signal refers to the electronic command issued by the nuclear power plant time synchronization controller, which simultaneously triggers the radiation-resistant camera and thermal imager to begin acquisition, avoiding time discrepancies between the dual-source video streams. Acquisition is the core operating mode of the equipment; the radiation-resistant camera records video in the visible light band, while the thermal imager records video of temperature changes in the infrared band. Dynamic adjustment of exposure parameters based on radiation level changes means that the radiation-resistant camera automatically adjusts the aperture and exposure time according to the real-time radiation dose, preventing overexposure or complete darkness caused by radiation pulses. Ambient steam temperature distribution is the real-time steam thermal field data detected by the thermal imager within the operating area, used to identify the location and temperature gradient of high-temperature steam clusters. Thermal sensitivity refers to the thermal imager automatically adjusting the thermal sensing accuracy of the infrared sensor based on steam temperature; sensitivity is reduced in high-temperature areas to avoid saturation distortion, and increased in low-temperature areas to enhance detail. The time stamp sequence refers to the precise timestamp (e.g., millisecond level) bound to each frame of video data, ensuring accurate matching of images from the same moment in the dual-source video streams.

[0074] In this embodiment, firstly, step 1011 determines the high-altitude operation hazard points (such as the reactor top cover hoisting port) based on the preset nuclear power plant safety layout diagram. The radiation-resistant camera and thermal imager are then installed in pairs on the same bracket, ensuring that their fields of view overlap and cover the target area. Secondly, step 1012 connects the two devices to the nuclear power plant time synchronization controller using cables. The controller generates a unified start signal with nanosecond-level precision and sends it synchronously to both devices. Next, step 1013 allows both devices to synchronously receive the start signal and simultaneously activate their respective acquisition functions: the radiation-resistant camera starts visible light video recording, and the thermal imager starts infrared video recording, both operating in parallel at the same frame rate (e.g., 30fps). Finally, in step 1014 during the acquisition process: First, the radiation-resistant camera's built-in radiation dose sensor monitors the ambient radiation value in real time. When the radiation level increases sharply, it automatically shortens the exposure time and reduces the aperture (e.g., the exposure time is reduced by 15% for every 10% increase in radiation value) to prevent snow noise in the image. Second, the thermal imager scans the temperature distribution of the image and automatically reduces the local thermal sensitivity when identifying high-temperature steam areas (>100℃) (e.g., the sensitivity in the steam area drops to 70%), while increasing the sensitivity to 130% in low-temperature background areas (<50℃) to balance thermal field details. Third, both devices add a unified time stamp sequence (e.g., "20230820_10:51:22.003") to each frame of video, outputting a strictly time-aligned dual-source video stream.

[0075] For example, during the hoisting operation of a steam generator on the roof (Area A) of a nuclear power plant reactor building: according to the nuclear power plant safety layout diagram, a radiation-resistant camera (model X) and a thermal imager (model Y) are deployed 5 meters above the hoisting opening in Area A (danger point), with a horizontal distance of 0.5 meters between them and their lenses angled downwards at 30° towards the work surface; the equipment is connected via fiber optic cable to a nuclear power plant time synchronization controller (brand Z) located in the central control room, and the controller sends a unified start signal before the operation begins; the two devices start recording synchronously, with the radiation-resistant camera capturing the hoisting steel cable... Visible light imaging and thermal imager real-time tracking of steam pipe heat dissipation temperature; when a sudden radiation leak occurs during hoisting (radiation level rises to 500 μSv / h), the radiation-resistant camera automatically adjusts the exposure time from 1 / 60 second to 1 / 250 second to avoid overexposure; at the same time, the thermal imager detects steam jet on the left (120℃), reduces the thermal sensitivity of that area from 100% to 65%, and outputs a dual-source video stream with millisecond-level timestamps (such as the time stamp "20230820_11:03:45.124" for frame 1032).

[0076] This solution achieves highly reliable acquisition of dual-modal video data in the complex environment of nuclear power plants through three core technologies: precise deployment of dual-source equipment, nanosecond-level time synchronization, and dynamic parameter adaptive adjustment. The equipment deployment covers all high-altitude hazardous points, eliminating monitoring blind spots; the time synchronization controller ensures frame-level alignment of the dual-source video streams, providing a precise time reference for subsequent radiation compensation and steam correction; the radiation-resistant camera uses dynamic exposure to avoid image distortion caused by radiation interference, and the thermal imager's sensitivity adaptively suppresses thermal field distortion caused by high-temperature steam; the final output dual-source video stream with time stamps combines optical detail integrity with thermodynamic stability, providing high-quality input for subsequent AI analysis.

[0077] Step 102: Identify radiated interference frames in the dual-source video stream, and perform radiated noise compensation operation on the video stream acquired by the radiation-resistant camera based on the identification results to generate a video stream after radiated noise compensation operation.

[0078] Optionally, step 102 may specifically include the following steps:

[0079] Step 1021: Based on the time stamp sequence of the dual-source video stream, locate the video frame at the corresponding time point in the video stream acquired by the radiation-resistant camera as a candidate frame;

[0080] Step 1022: Detect the accumulated noise of pixel units caused by gamma rays in the candidate frame, and count the accumulated noise density of pixel units that exceeds the normal threshold of the visible light band from the detection results.

[0081] Step 1023: When the accumulated noise density of the pixel unit continuously reaches the radiation interference threshold and exceeds the preset number of frames, it is determined to be a radiation interference frame, and the total amount of radiation noise corresponding to the radiation interference frame is recorded.

[0082] Step 1024: Based on the radiated interference frame and the corresponding total amount of radiated noise, assign a compensation weight to the radiated interference frame that is inversely proportional to the total amount of radiated noise.

[0083] Step 1025: Adjust the pixel unit values ​​of the radiated interference frame using the compensation weight to obtain the video stream after radiated noise compensation operation.

[0084] Step 1025 may specifically include the following steps:

[0085] Based on the spatial distribution of pixel units in the radiated noise frame, a compensation region adjacent to radiated noise points above a preset threshold is determined. Reference values ​​of pixel units unaffected by radiated noise are extracted from the compensation region. The compensation weight is decomposed into a horizontal weight component and a vertical weight component related to the diffusion direction of the radiated noise. The grayscale shift of pixel units is suppressed proportionally along the horizontal axis according to the horizontal weight component, and simultaneously, the grayscale shift of pixel units is suppressed proportionally along the vertical axis according to the vertical weight component, to obtain pixel unit values ​​after horizontal and vertical suppression. These pixel unit values ​​are then replaced with the reference values ​​of pixel units within the compensation region to obtain the video stream after radiated noise compensation.

[0086] In the above scheme, radiated interference frames refer to video frames in the radiation-resistant camera video stream that exhibit abnormal snow-like noise due to gamma ray interference. These frames are characterized by noise density exceeding the normal threshold and persisting for multiple frames. Radiated noise compensation is a technique for repairing images affected by radiated interference. By analyzing noise distribution characteristics, it replaces damaged areas with surrounding normal pixel values. The video stream after radiated noise compensation refers to the video stream that has undergone noise repair processing, where radiated snow-like noise is effectively suppressed. Video frames corresponding to specific time points refer to images selected from the radiation-resistant camera video stream at the same moment as the thermal imager, based on the time markers of the dual-source video streams. Candidate frames are alternative video frames for detecting radiated interference, selected frame by frame in chronological order. Pixel unit cumulative noise refers to white or colored spots in the image formed by gamma rays striking the sensor, similar to the snow-like noise on an old-fashioned television. The normal threshold for the visible light band refers to the allowable noise density standard for an image under normal conditions (e.g., no more than 100 noise points per frame). Pixel unit cumulative noise density is the percentage of radiated noise per unit image area (e.g., 1% of the image is covered by noise). The radiated interference threshold is the noise density threshold for determining severe image interference (e.g., noise covering more than 5% of the image). The preset frame number is the required number of consecutive frames of interference (e.g., only after 3 consecutive frames exceeding the threshold is it considered interference). The corresponding total radiated noise is the sum of all noise in a single frame. The compensation weight is a control parameter for the repair intensity; the more noise, the greater the repair intensity (the smaller the weight value). The pixel unit value is the color value of each pixel in the image (e.g., RGB value). Spatial distribution relationship refers to the arrangement of noise in the image (e.g., concentrated in the upper left corner or scattered). The preset threshold for radiated noise refers to noise with abnormal brightness or color values ​​exceeding the set standard (e.g., pure white spots). The compensation area is a selected area of ​​normal surrounding image around the noise, used to provide a repair reference value. The pixel unit baseline value is the average color value of undisturbed pixels within the compensation area. The horizontal weight component refers to the horizontal repair intensity parameter (used to handle horizontally elongated noise). The vertical weight component refers to the vertical repair intensity parameter (used to handle vertically elongated noise). Suppressing the grayscale shift of the pixel unit reduces the color difference between noise pixels and normal values. The pixel unit values ​​after horizontal and vertical suppression refer to the pixel color values ​​after being repaired in the horizontal and vertical directions, respectively.

[0087] In this embodiment, firstly, in step 1021, based on the time stamp sequence of the dual-source video stream (e.g., "11:03:45.124"), a video frame with the exact same timestamp is located in the radiation-resistant camera video as a candidate frame. Secondly, in step 1022, a noise detection algorithm is used to scan the candidate frames: identifying pixels with brightness values ​​exceeding the normal range (e.g., pure white RGB(255,255,255)), and calculating the proportion of these noise points to the total number of pixels in the image to obtain the noise density. Next, in step 1023, the system continuously monitors multiple frames: if the noise density continuously exceeds a threshold value (e.g., density > 5% for 3 consecutive frames), the frame is marked as a radiation interference frame, and the total number of noise points in this frame is recorded (e.g., 1200 noise points). Then, in step 1024, a compensation weight is calculated based on the total noise amount: the more noise, the smaller the weight (e.g., a total noise amount of 1200 corresponds to a weight of 0.3). Finally, pixel repair is performed in step 1025: First, in the radiated interference frame, a 3×3 pixel compensation area is drawn with each noise point as the center; Second, the average color value of the undisturbed pixels in the compensation area is calculated as the reference value (e.g., RGB(120,115,110)); Third, the compensation weight is split into a horizontal component (0.15) and a vertical component (0.15); Fourth, for the noise pixels: Horizontally: reduce the color difference between the pixel and the reference value by a horizontal weight of 0.15 (e.g., the original RGB(255,255,255) is suppressed to become RGB(230,230,230)); Vertically: reduce the difference again by a vertical weight of 0.15 (finally becoming RGB(210,210,210)); Fifth, replace the repaired pixel value with the reference value (RGB(120,115,110)) and output the repaired video frame.

[0088] Following the specific implementation of the previous step, in the A area scenario of step 101: when radiation leakage causes dense snowflakes to appear in frame 1032 of the radiation-resistant camera (timestamp "20230820_11:03:45.124"): locate the corresponding candidate frame based on the timestamp; detect 300 pure white noise points (density 8%) in the upper left corner of the image; when the noise density exceeds the critical value of 5% for 3 consecutive frames, it is determined to be a radiation interference frame, and the total number of noise points is recorded as 900; calculate the compensation weight = 1 / (1+900 / 10). 0)=0.1 (Example Algorithm); For each noise point: 1. Take the surrounding 3×3 area and calculate the normal pixel mean RGB(100,105,95); 2. Decompose the weight into horizontal / vertical values ​​of 0.05 each; 3. The original noise value RGB(255,255,255) is suppressed horizontally to RGB(240,240,240), and then suppressed vertically to RGB(230,230,230); 4. Finally, it is replaced with the baseline value RGB(100,105,95), and the snowflakes disappear after repair.

[0089] This solution effectively eliminates gamma ray interference in video footage through a triple mechanism of precise time positioning, continuous multi-frame verification, and dynamic weight compensation: frame-level matching based on time identifiers ensures consistency between dual-source data; the judgment logic for continuous exceedance of multiple frames avoids misjudgment due to instantaneous interference; spatial distribution analysis and weight decomposition technology achieve precise noise repair; and the final output compensated video stream restores the true details of the image, providing high-quality visible light data for subsequent high-temperature steam correction.

[0090] Step 103: Simultaneously, based on the recognition results, perform a high-temperature steam distortion correction operation on the video stream acquired by the thermal imager to generate a video stream after the high-temperature steam distortion correction operation.

[0091] Optionally, step 103 may specifically include the following steps:

[0092] Step 1031. Based on the time stamp of the radiation interference frame, locate the video frame at the corresponding time in the video stream acquired by the thermal imager as the steam-affected frame;

[0093] Step 1032: Divide the high-temperature steam region and the isothermal background region in the steam-affected frame;

[0094] Step 1033: Extract the temperature distribution profile of the isothermal background region as the reference thermal field;

[0095] Step 1034: Calculate the thermal diffusion coverage coefficient of high-temperature steam on the background target based on the angle between the steam jet direction and the optical axis of the thermal imager.

[0096] Step 1035: Based on the thermal diffusion coverage coefficient, the pixel unit thermal field transmission value of the high-temperature steam region is adjusted in reverse so that the temperature distribution of the high-temperature steam region forms a thermal field after radiation attenuation correction, so as to output the video stream after high-temperature steam distortion correction operation.

[0097] In the above scheme, high-temperature steam distortion correction is a technique to repair temperature display distortion caused by high-temperature steam in thermal imaging images. It restores the true thermal field distribution by analyzing the steam coverage pattern. The video stream after high-temperature steam distortion correction refers to the thermal imaging video stream repaired by steam interference, accurately displaying the true temperature of objects obscured by steam. Steam-affected frames are the frames in the thermal imaging video stream corresponding to the radiation interference frames, where high-temperature steam may cause thermal field distortion. High-temperature steam regions refer to continuous high-temperature pixel blocks in the image whose temperature is significantly higher than the environment (e.g., >100℃) and conforms to the steam pattern. Isothermal background regions are background areas in the image with stable temperatures and no drastic changes (e.g., walls or equipment surfaces at 50℃±2℃). Temperature distribution contours refer to the shape boundary lines formed by connecting the temperature values ​​of pixels in the isothermal background region, reflecting the original form of the stable thermal field. The reference thermal field is a stable temperature distribution model extracted from the isothermal background region, serving as a reference benchmark for correcting steam distortion. Steam jet direction refers to the flow angle of high-temperature steam in the image (e.g., diffusing from the lower right to the upper left at a 45° angle). The angle between the optical axis of a thermal imager refers to the angle between the direction of the steam jet and the centerline of the thermal imaging lens (e.g., a 30° deflection). The thermal diffusion coverage coefficient is a parameter characterizing the degree to which steam obscures the thermal field of the background target (the larger the angle, the higher the coefficient). The pixel unit thermal field transmission value refers to the ability of each pixel to transmit true temperature information (this value decreases when obscured by steam). The thermal field after radiation attenuation correction refers to the true temperature distribution image restored by increasing the thermal field transmission value.

[0098] In this embodiment, firstly, in step 1031, based on the radiation interference frame time stamp (e.g., "11:03:45.124") determined in step 102, the frame at the same moment in the thermal imaging video stream is located as the steam-affected frame. Secondly, in step 1032, a temperature threshold segmentation algorithm is used: areas with temperatures higher than a set value (e.g., 100℃) and connected pixels are marked as high-temperature steam areas, while other areas with small temperature fluctuations are marked as isothermal background areas. Next, in step 1033, the temperature distribution contour of the isothermal background area is extracted: the average pixel temperature (e.g., 52℃) and boundary shape within the area are calculated to construct a baseline thermal field model. Then, in step 1034, the steam jet direction (e.g., from the lower right to the upper left of the image) is determined through image analysis, the angle θ (e.g., 30°) between the steam jet and the center line of the thermal imager lens is measured, and the thermal diffusion coverage coefficient w is calculated. Finally, in step 1035, for each pixel in the high-temperature steam region: first, adjust the thermal field transmission value in reverse according to the thermal diffusion coverage coefficient k: new transmission value = original transmission value / k (e.g., the original value of 0.7 is corrected to 0.7 / 0.87≈0.8); second, use the improved transmission value to restore the true temperature: new temperature = (pixel measurement value - steam interference) × new transmission value; finally, output the corrected thermal field image to form the video stream after the high-temperature steam distortion correction operation.

[0099] Following the specific implementation of the previous step, in the A area scene of step 102: when steam interference occurs in frame 1032 of the thermal imager (timestamp "20230820_11:03:45.124"): firstly, locate the steam-affected frame according to the same time stamp; secondly, divide the 120×80 pixel area in the lower right corner as the high-temperature steam area (110℃), and the remaining wall area as the isothermal background area (50℃±1℃); then extract the wall temperature distribution contour as the reference thermal field (a rectangular area with an average temperature of 50℃); then measure the angle θ=40° between the steam jet direction and the lens optical axis, and calculate k=cos(40°)≈0.77; subsequently, for each pixel in the steam area: the original transmission value of 0.75 is increased to 0.75 / 0.77≈0.97; finally, after correction, the temperature of the bolt covered by steam is restored from the blurry 80℃ to the true 95℃.

[0100] This step effectively eliminates the interference of high-temperature steam on thermal imaging through three technologies: time synchronization positioning, thermal field partitioning modeling, and physical angle compensation. Based on the time stamp of the radiation interference frame, the thermal imaging frame is accurately located and associated; the isothermal background area provides a reliable thermal field reference; the physical model of the steam jet angle realizes the quantitative calculation of the thermal diffusion coefficient; the true temperature distorted by the steam is restored by improving the thermal field transmission value; and finally, the corrected video stream clearly presents the thermal characteristics of the concealed target.

[0101] Step 104: Input the video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation into the pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markings.

[0102] Optionally, step 104 may specifically include the following steps:

[0103] Step 1041: Bind the pixel units of the video stream after the radiation noise compensation operation to radiation dose markers to obtain video frames with radiation dose markers.

[0104] Step 1042: Superimpose spatial gradient coordinates onto the thermal field markers of the video stream after the high-temperature steam distortion correction operation to obtain a video frame with thermal field gradient markers.

[0105] Step 1043: In the nuclear power characteristic adaptation model, perform thermal radiation conflict index calculation on the video frames with radiation dose markings and the video frames with thermal field gradient markings;

[0106] Step 1044: When the thermal radiation conflict index in the solution result is lower than the preset nuclear power environment conflict threshold, the video frame with radiation dose marker and the video frame with thermal field gradient marker are fused to generate a composite marker video frame.

[0107] Step 1045: Based on the displacement vector changes of the markers in the composite marker video frames, connect the depth positions of the nuclear power coordinate system frame by frame to construct the three-dimensional motion trajectory field containing the radiation region markers.

[0108] In the above scheme, the pre-trained nuclear power feature adaptation model refers to an artificial intelligence model trained using historical nuclear power monitoring data, used to identify the motion characteristics of objects under the special environment (radiation, high temperature) of a nuclear power plant. The three-dimensional motion trajectory field containing radiation area markers is a data field recording the path of an object in three-dimensional space, additionally annotating the radiation intensity information of the area the object passes through. Radiation dose markers refer to radiation value labels (e.g., low / medium / high radiation areas) attached to visible light video pixels, derived from radiation sensor data from radiation-resistant cameras. Video frames with radiation dose markers refer to visible light images where each pixel is accompanied by a radiation value (e.g., red markers indicate high radiation areas). Thermal field markers refer to temperature feature labels for each pixel in the thermal imaging video (e.g., high-temperature objects, low-temperature background). Spatial gradient coordinates refer to spatial location information describing the direction and magnitude of temperature changes (e.g., temperature decreases from the upper left to the lower right). Video frames with thermal field gradient markers refer to thermal imaging images where each pixel is accompanied by the direction of temperature change. The thermal radiation conflict index solution is an algorithm that compares whether the visible light radiation value at the same location contradicts the temperature value in the thermal imaging (e.g., high-radiation areas should not contain low-temperature objects). The preset nuclear power plant environmental conflict threshold refers to the maximum reasonable range within which radiation and temperature values ​​can differ. A composite marker video frame is a frame that integrates radiation dose markers and thermal field gradient markers, simultaneously displaying the object's radiation properties and temperature change characteristics. Displacement vector change refers to the direction and distance an object moves within a continuous frame (e.g., a bolt moving 10 pixels from the upper left to the lower right). The nuclear power plant coordinate system depth position is a three-dimensional coordinate system established with a fixed point in the nuclear power plant as the origin, indicating the distance of objects from the camera (e.g., Z-axis coordinate 3.5 meters).

[0109] In this embodiment, firstly, step 1041 reads the visible light video stream after radiation noise compensation and adds a real-time radiation value label to each pixel (e.g., the radiation value of 500 μSv / h in the upper left corner of the image is marked in red). Secondly, step 1042 calculates the direction and magnitude of temperature change between each pixel and its surrounding pixels (e.g., the temperature decreases at a rate of 30°, decreasing by 5°C per second) on the thermal imaging video stream after high-temperature steam correction, generating a thermal gradient map with arrows. Next, in step 1043, the two types of marked images are input into the nuclear power feature adaptation model: the model compares the radiation mark and the thermal field gradient mark of the pixel at the same location (if a low-temperature object appears in a high-radiation area, a conflict is determined); the thermal radiation conflict index is calculated: Iconflict=Tmax|Tactual−TExpected| (where T is the temperature value). Example: the expected temperature in the high-radiation area should be >80℃, but the actual detected temperature is 50℃, so the conflict index =|50-80| / 100=0.3; then in step 1044, when the conflict index is <threshold (e.g., 0.2), the two types of marks are superimposed: the visible light pixels retain the radiation mark (color block), and the thermal imaging pixels retain the gradient arrow, generating a composite image. Finally, step 1045 tracks the movement of markers in the composite image: identify the positional changes of the same marker in consecutive frames (e.g., a bolt marker moves from (100,200) to (110,205)); calculate the depth position based on the displacement vector (faster movement means closer distance, slower movement means farther distance); connect the position points of each frame in the nuclear power coordinate system (X horizontal, Y vertical, Z depth) to form a three-dimensional trajectory field (e.g., trajectory point sequence (1m,5m,3m)→(1.1m,4.9m,2.8m)).

[0110] Following the specific implementation of the previous step, in the A area scene of step 103: In the repaired visible light image, the fallen bolt is marked as a high-radiation area (red); the corrected thermal imaging shows that the bolt temperature is 95℃, and its thermal field gradient arrow is downward (cooling down); the model calculates the conflict index: the expected temperature of the high-radiation area is >80℃, and the actual temperature of 95℃ is in line with expectations (index 0.05 < threshold 0.2); a composite frame is generated with the red bolt marked with a downward arrow; the bolt is tracked for 3 consecutive frames: the position of the bolt in the first frame is (100, 200, Z=3.5m), the displacement in the second frame is (110, 205, Z=3.2m) (vector: 10 pixels to the right / 5 pixels down), and the displacement in the third frame is (120, 210, Z=2.9m), forming a three-dimensional trajectory of downward movement along the Z-axis.

[0111] This solution utilizes three technologies—dual-source data labeling, intelligent conflict detection, and dynamic trajectory construction—to achieve precise motion analysis in the complex environment of nuclear power plants: radiation dose labeling reveals the radioactive risk of objects, and thermal field gradient labeling reflects temperature change patterns; a nuclear power characteristic adaptation model filters out interference signals that contradict radiation and temperature (such as low-temperature false targets caused by steam); composite marker video frames synchronously present the physical properties and thermodynamic behavior of objects; and depth position calculation based on displacement vectors constructs a realistic three-dimensional trajectory, providing reliable input for safety criteria.

[0112] Step 105: Establish dynamic target criteria for nuclear power safety regulations based on the three-dimensional motion trajectory field. When the dynamic target criteria meet the preset high-altitude falling object behavior pattern, trigger an alarm protocol covering multiple levels of safety control units.

[0113] Optionally, step 105 may specifically include the following steps:

[0114] Step 1051: Extract the ultra-rapid descent feature from the depth position change of the three-dimensional motion trajectory field to generate a set of key parameters for dynamic target criteria;

[0115] Step 1052: Match the set of key parameters with the nuclear power equipment damage coefficient in the preset high-altitude falling object behavior mode to verify the effectiveness of the dynamic target criterion and obtain the verification result;

[0116] Step 1053: When the set of key parameters in the verification results continuously satisfies the radiation acceleration constraint condition in the high-altitude falling object behavior mode, the multi-level control engine of the preset alarm protocol is activated.

[0117] Step 1054: The multi-level control engine generates a response link for a multi-level safety control unit, wherein the response link includes an emergency braking unit, a containment isolation unit, and a central control unit that are bound to the nuclear power safety classification matrix.

[0118] Step 1055: Extract the trajectory coordinate sequence and safety risk label from the dynamic target criterion to construct a directional alarm instruction set;

[0119] Step 1056: Distribute the directional alarm instruction set to the corresponding level control unit through the response link to execute a closed-loop response.

[0120] In the above scheme, nuclear power safety regulations are the specific requirements of nuclear power plant safety standards for monitoring falling objects (such as alarm thresholds for the size and speed of falling objects). Dynamic target criteria refer to the real-time risk assessment basis generated based on the object's motion characteristics (position, speed). Preset high-altitude falling object behavior patterns are a pre-defined library of typical falling object characteristics (such as tools falling in an accelerating linear motion). Alarm protocols covering multi-level safety control units refer to alarm mechanisms that link different safety-level devices (emergency braking / isolation / central control). Overspeed descent feature extraction is the process of identifying objects falling at speeds exceeding safety standards (such as falling more than 1 meter per second). The key parameter set refers to a data set containing core motion characteristics such as object position, speed, and acceleration. The nuclear equipment damage coefficient refers to the risk level of damage caused by an object impact to different equipment (such as a coefficient of 0.3 for small parts impacting pipes). Verification results are the conclusions drawn from matching the key parameter set with the behavior patterns (valid / invalid alarms). Radiation falling acceleration constraints refer to the physical laws that radioactive objects should satisfy when falling (such as continuous acceleration due to gravity). The multi-level control engine is the core program that drives the coordinated operation of different safety devices. A response chain is the logical sequence in which safety devices are activated after an alarm is triggered (e.g., braking before isolation). A nuclear power plant safety classification matrix is ​​a rule table defining control measures corresponding to different risk levels (e.g., activating all equipment for high-risk situations). An emergency braking unit refers to a device that urgently stops high-altitude work equipment (e.g., crane emergency stop). A containment isolation unit is a system that encloses hazardous areas (e.g., automatically closing explosion-proof doors). A central control unit is a central control system that receives alarm information and dispatches personnel. A trajectory coordinate sequence refers to a continuous record of the position of an object during its fall (e.g., one coordinate point per second). A safety risk label indicates the risk level of an object (e.g., "high-radiation bolt"). A directional alarm command set refers to alarm commands customized for different control units (e.g., sending "XY area emergency stop" to the braking unit). Control units at corresponding levels are response devices matched according to risk levels (e.g., only notifying the central control unit for low-risk situations).

[0121] In this embodiment, step 1051 first analyzes the changes in the depth position of the object in the three-dimensional motion trajectory field (e.g., the Z-coordinate decreases from 5 meters to 2 meters), calculates the falling distance per second, and extracts overspeed features (e.g., speed > safety standard value). Next, the key parameters (speed 1.5 m / s, object type bolt) are matched with a preset behavior pattern: the nuclear power equipment damage coefficient database is queried (bolt damage coefficient to pipe 0.3); it is verified whether it conforms to the characteristics of a radiated falling object (continuous acceleration and vertical trajectory). Then, in step 1053, when the verification shows a continuous increase in speed (e.g., 0.8 m / s → 1.2 m / s → 1.5 m / s) and the acceleration constraint is met, the multi-level control engine is activated. Then, in step 1054, the engine generates a response link based on the safety classification matrix: high-risk event: simultaneously triggers the emergency braking unit (stops the crane), the containment isolation unit (closes the explosion-proof door in area A), and the central control unit (alarm sound + pop-up window); low-risk event: only the central control unit is activated. After extracting the trajectory coordinates (e.g., (1m, 5m, 3m) → (1m, 4m, 2m)) and safety tags (“high-radiation bolts”) in step 1055, an instruction set is constructed: braking unit instruction: “Stop XY coordinate crane”; isolation unit instruction: “Close gate 3 in area A”. Finally, the response link distribution instruction is executed in step 1056: high-risk instructions are sent to all three levels of units simultaneously, while low-risk instructions are sent only to the central unit.

[0122] Following the specific implementation of the previous step, in the bolt falling scenario of step 104: First, it is detected that the bolt drops from Z=3.5m to Z=0.5m within 3 seconds (speed 1m / s→1.5m / s); second, it matches the behavior pattern library (bolt failure coefficient 0.3) and verifies that it meets the characteristics of accelerated falling; then, the speed continues to exceed the limit, and the multi-level control engine is activated; then, the safety classification matrix determines it to be high risk (the landing point is close to the steam pipe), and a three-level response link is generated; subsequently, the trajectory coordinate sequence and the safety label "high radiation bolt" are extracted, and instructions are constructed: braking unit: "Stop crane B"; isolation unit: "Close the east gate of area A"; central control: "Alarm: Bolt fell at (120,210,0.5m)". Finally, the instructions are synchronously sent to the three control units for execution.

[0123] This solution achieves closed-loop management of nuclear power plant falling object risks through three major technologies: intelligent behavior matching, hierarchical response mechanism, and precise command distribution. Overspeed descent feature extraction accurately captures dangerous movements; dual verification of equipment damage coefficient and behavior pattern improves alarm reliability; multi-level control engine automatically schedules safety equipment according to risk level; directional alarm command set ensures precise execution of response measures; and ultimately forms a complete safety closed loop of "monitoring-judgment-braking-isolation-alarm".

[0124] Figure 2 This application provides a structural schematic diagram of an AI video monitoring system for falling objects from heights in the complex environment of nuclear power plants, as shown in the following figure. Figure 2 As shown, the system includes:

[0125] The acquisition module 21 is used to acquire dual-source video streams synchronously collected by a radiation-resistant camera and a thermal imager in the nuclear power operation area, wherein the dual-source video streams include the video streams collected by the radiation-resistant camera and the video streams collected by the thermal imager.

[0126] The compensation module 22 is used to identify radiated interference frames in the dual-source video stream, and perform radiated noise compensation operation on the video stream acquired by the radiation-resistant camera based on the identification result, thereby generating a video stream after radiated noise compensation operation.

[0127] The correction module 23 is used to simultaneously perform high-temperature steam distortion correction operation on the video stream acquired by the thermal imager based on the recognition result, and generate a video stream after high-temperature steam distortion correction operation;

[0128] Construction module 24 is used to input the video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation into a pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markings.

[0129] The trigger module 25 is used to establish dynamic target criteria for nuclear power safety regulations based on the three-dimensional motion trajectory field. When the dynamic target criteria meet the preset high-altitude falling object behavior pattern, an alarm protocol covering multiple levels of safety control units is triggered.

[0130] Figure 2 The aforementioned AI video monitoring system for high-altitude falling objects in the complex environment of nuclear power plants can perform... Figure 1 The implementation principle and technical effects of the AI ​​video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants, as described in the above embodiment, will not be repeated here. The specific operation methods of each module and unit in the AI ​​video monitoring system for high-altitude falling objects in the complex environment of nuclear power plants described in the above embodiment have been described in detail in the embodiments of the relevant method, and will not be elaborated upon here.

[0131] In one possible design, Figure 2 The high-altitude falling object AI video monitoring system of the embodiment shown can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0132] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0133] The processing component 32 is used for the above Figure 1The embodiment describes an AI video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants.

[0134] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0135] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0136] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0137] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0138] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0139] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0140] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is an AI video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants.

[0141] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0142] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A high-altitude falling object AI video monitoring method for the complex environment of nuclear power plants, characterized in that, include: Acquire dual-source video streams simultaneously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include the video stream collected by the radiation-resistant camera and the video stream collected by the thermal imager; The dual-source video stream is subjected to radiated interference frames. Based on the identification results, a radiated noise compensation operation is performed on the video stream acquired by the radiation-resistant camera to generate a video stream after the radiated noise compensation operation. Simultaneously, based on the recognition results, a high-temperature steam distortion correction operation is performed on the video stream acquired by the thermal imager to generate a video stream after the high-temperature steam distortion correction operation; The video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation are input into a pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markers. Based on the three-dimensional motion trajectory field, a dynamic target criterion constrained by nuclear power safety regulations is established. When the dynamic target criterion conforms to the preset high-altitude falling object behavior pattern, an alarm protocol covering multiple levels of safety control units is triggered. The step of inputting the video stream after the radiation noise compensation operation and the video stream after the high-temperature steam distortion correction operation into a pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markers includes: The pixel units of the video stream after the radiation noise compensation operation are bound to radiation dose markers to obtain video frames with radiation dose markers. The thermal field markers of the video stream after the high-temperature steam distortion correction operation are superimposed with spatial gradient coordinates to obtain a video frame with thermal field gradient markers. In the nuclear power characteristic adaptation model, the thermal radiation conflict index is calculated for the video frames with radiation dose markings and the video frames with thermal field gradient markings. When the thermal radiation conflict index in the solution result is lower than the preset nuclear power environment conflict threshold, the video frame with radiation dose marker and the video frame with thermal field gradient marker are fused to generate a composite marker video frame. Based on the displacement vector changes of the markers in the composite marker video frames, the depth positions in the nuclear power coordinate system are connected frame by frame to construct the three-dimensional motion trajectory field containing the radiation region markers.

2. The method according to claim 1, characterized in that, Acquire dual-source video streams simultaneously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include a video stream collected by the radiation-resistant camera and a video stream collected by the thermal imager, including: According to the preset nuclear power safety layout diagram, the radiation-resistant camera and the thermal imager are deployed at the high-altitude operation hazard points in the nuclear power operation area; Connect the radiation-resistant camera and the thermal imager to the nuclear power plant time synchronization controller, and use the nuclear power plant time synchronization controller to send a unified start signal; In response to the unified start signal, the acquisition functions of the radiation-resistant camera and the thermal imager are activated simultaneously; During the acquisition process, the radiation-resistant camera dynamically adjusts its exposure parameters according to the changes in radiation levels at its location, and the thermal imager adjusts its thermal sensitivity according to the surrounding steam temperature distribution to output a dual-source video stream, wherein the dual-source video stream is bound to a time stamp sequence.

3. The method according to claim 1, characterized in that, The process involves identifying radiated interference frames in the dual-source video stream, performing radiated noise compensation on the video stream acquired by the radiation-resistant camera based on the identification results, and generating a video stream after radiated noise compensation, including: Based on the time stamp sequence of the dual-source video stream, the video frames at the corresponding time points in the video stream acquired by the radiation-resistant camera are identified as candidate frames. Detect the accumulated noise of pixel units caused by gamma rays in the candidate frame, and count the accumulated noise density of pixel units that exceeds the normal threshold of the visible light band from the detection results; When the accumulated noise density of the pixel unit continuously reaches the radiative interference threshold and exceeds the preset number of frames, it is determined to be a radiative interference frame, and the total amount of radiative noise corresponding to the radiative interference frame is recorded. Based on the radiated interference frame and the corresponding total amount of radiated noise, a compensation weight inversely proportional to the total amount of radiated noise is assigned to the radiated interference frame. The pixel unit values ​​of the radiated interference frame are adjusted using the compensation weights to obtain the video stream after radiated noise compensation.

4. The method according to claim 3, characterized in that, The pixel values ​​of the radiated interference frame are adjusted using the compensation weights to obtain the video stream after radiated noise compensation, including: Based on the spatial distribution relationship of pixel units in the radiated interference frame, the compensation region adjacent to the radiated noise points above the preset threshold is determined. Extract the reference value of the pixel unit that is not affected by radiation interference within the compensation area; The compensation weights are decomposed into horizontal and vertical weight components that are related to the diffusion direction of the radiated noise. The grayscale shift of the pixel unit is suppressed proportionally along the horizontal axis according to the horizontal weight component, and the grayscale shift of the pixel unit is suppressed proportionally along the vertical axis according to the vertical weight component, so as to obtain the pixel unit value after horizontal and vertical suppression. The pixel values ​​that have undergone lateral and longitudinal suppression are replaced with the pixel reference values ​​within the compensation area to obtain the video stream after radiative noise compensation.

5. The method according to claim 1, characterized in that, Simultaneously, based on the recognition results, a high-temperature steam distortion correction operation is performed on the video stream acquired by the thermal imager to generate a video stream after the high-temperature steam distortion correction operation, including: Based on the time stamp of the radiation interference frame, the video frame at the corresponding moment in the video stream acquired by the thermal imager is located as the steam-affected frame. The high-temperature steam region and the isothermal background region in the steam-affected frame are divided; The temperature distribution profile of the isothermal background region is extracted as the reference thermal field; The thermal diffusion coverage coefficient of high-temperature steam on the background target is calculated based on the angle between the steam jet direction and the optical axis of the thermal imager. Based on the thermal diffusion coverage coefficient, the pixel unit thermal field transmission value of the high-temperature steam region is adjusted in reverse to form a thermal field after radiation attenuation correction in the temperature distribution of the high-temperature steam region, so as to output the video stream after high-temperature steam distortion correction operation.

6. The method according to claim 1, characterized in that, Based on the three-dimensional motion trajectory field, a dynamic target criterion constrained by nuclear power safety regulations is established. When the dynamic target criterion matches a preset high-altitude falling object behavior pattern, an alarm protocol covering multiple levels of safety control units is triggered, including: The depth position change of the three-dimensional motion trajectory field is subjected to ultra-rapid descent feature extraction to generate a set of key parameters for dynamic target discrimination; The set of key parameters is matched with the nuclear power equipment damage coefficient in the preset high-altitude falling object behavior pattern to verify the effectiveness of the dynamic target criterion and obtain the verification results. When the set of key parameters in the verification results continuously satisfies the radiation acceleration constraint condition in the high-altitude falling object behavior mode, the multi-level control engine of the preset alarm protocol is activated. The multi-level control engine generates a response link for a multi-level safety control unit, wherein the response link includes an emergency braking unit, a containment isolation unit, and a central control unit that are bound to the nuclear power safety classification matrix. The trajectory coordinate sequence and safety risk label are extracted from the dynamic target criteria to construct a directional alarm instruction set; The directional alarm instruction set is distributed to the corresponding level control unit through the response link to execute a closed-loop response.

7. A high-altitude falling object AI video monitoring system for the complex environment of nuclear power plants, characterized in that, include: The acquisition module is used to acquire dual-source video streams synchronously collected by a radiation-resistant camera and a thermal imager in the nuclear power plant operation area, wherein the dual-source video streams include the video streams collected by the radiation-resistant camera and the video streams collected by the thermal imager. The compensation module is used to identify radiated interference frames in the dual-source video stream, and perform radiated noise compensation operation on the video stream acquired by the radiation-resistant camera based on the identification results, thereby generating a video stream after radiated noise compensation. The correction module is used to simultaneously perform high-temperature steam distortion correction on the video stream acquired by the thermal imager based on the recognition results, and generate a video stream after high-temperature steam distortion correction. The construction module is used to input the video stream after the radiation noise compensation operation and the video stream after the high temperature steam distortion correction operation into the pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markings. The triggering module establishes dynamic target criteria constrained by nuclear power safety regulations based on the three-dimensional motion trajectory field. When the dynamic target criteria meet the preset high-altitude falling object behavior pattern, it triggers an alarm protocol covering multiple levels of safety control units. The step of inputting the video stream after the radiation noise compensation operation and the video stream after the high-temperature steam distortion correction operation into a pre-trained nuclear power feature adaptation model to construct a three-dimensional motion trajectory field containing radiation region markers includes: The pixel units of the video stream after the radiation noise compensation operation are bound to radiation dose markers to obtain video frames with radiation dose markers. The thermal field markers of the video stream after the high-temperature steam distortion correction operation are superimposed with spatial gradient coordinates to obtain a video frame with thermal field gradient markers. In the nuclear power characteristic adaptation model, the thermal radiation conflict index is calculated for the video frames with radiation dose markings and the video frames with thermal field gradient markings. When the thermal radiation conflict index in the solution result is lower than the preset nuclear power environment conflict threshold, the video frame with radiation dose marker and the video frame with thermal field gradient marker are fused to generate a composite marker video frame. Based on the displacement vector changes of the markers in the composite marker video frames, the depth positions in the nuclear power coordinate system are connected frame by frame to construct the three-dimensional motion trajectory field containing the radiation region markers.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the AI ​​video monitoring method for high-altitude falling objects in the complex environment of nuclear power plants as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an AI video monitoring method for high-altitude falling objects in complex nuclear power plant environments as described in any one of claims 1 to 6.

Images