Radiation protection wall air tightness detection method based on image fusion

Abstract

The present application relates to the technical field of air tightness detection, in particular to a radiation protection wall air tightness detection method based on image fusion, comprising the following steps: acquiring multi-band images of the protection wall by multi-spectral imaging, collecting temperature and humidity and extracting wavelet transform features, constructing a time series set, calculating the change rate by difference, detecting extreme values by sliding window, outputting the change rate vector, clustering the Euclidean distance and included angle of spatial adjacent points, judging abnormal areas, modeling the polynomial trend line, and judging and outputting the positioning and tracking results. In the present application, the temperature and humidity data of the entire wall area are extracted in real time by fusing multi-spectral image information, the change rate analysis and trend line modeling are performed in combination with the sequence change trend and spatial distribution characteristics, the suspected air tightness abnormal area is more accurately and efficiently identified and positioned, the air tightness performance change is dynamically monitored and continuously tracked, the detection sensitivity and reliability are improved, the missed detection and misjudgment are reduced, the detection accuracy is ensured in complex environments, and the comprehensive evaluation and early risk warning of wall air tightness are effectively supported.

Description

Technical Field

[0001] This invention relates to the field of airtightness testing technology, and in particular to an image fusion-based method for testing the airtightness of radiation shielding walls. Background Technology

[0002] Radiation shielding wall airtightness testing technology belongs to the field of testing and measurement technology. Its core application scenario is to systematically test and scientifically evaluate the airtightness performance of radiation shielding walls during their service life. The core scope of this technology is clearly defined, mainly including five key aspects: first, structural integrity testing of the radiation shielding wall to ensure that the wall's foundation structure is free of damage or defects that could affect its sealing performance; second, airtightness parameter measurement to accurately obtain core data reflecting the wall's sealing degree; third, leak point identification to locate potential gas leaks within the wall; fourth, environmental pressure control to create a stable testing environment to ensure data accuracy; and fifth, sealing performance evaluation to comprehensively assess the sealing effect and risk of the wall based on various data. In terms of technical coverage, this field integrates physical testing methods with the application of specialized testing equipment, covering the entire process from data acquisition and in-depth parameter analysis to the development of testing standards, forming a relatively complete technical system.

[0003] In the technology system for testing the airtightness of radiation shielding walls, traditional testing algorithms are among the earliest and most fundamental methods. Their core principle involves applying a specific gas pressure to the radiation shielding wall and then using specialized pressure change measuring instruments to monitor the pressure changes within the wall in real time. The sealing performance of the wall is then assessed based on the pressure change data. In practice, traditional testing algorithms typically employ two pressure application methods: applying positive pressure to one side of the wall or applying negative pressure. Regardless of the method used, high-precision instruments such as digital pressure gauges and differential pressure gauges are employed to continuously record the pressure decay within the wall per unit time, or to calculate changes in gas flow rate using relevant devices. Testing personnel analyze the rate of pressure drop or directly calculate the gas leakage flow rate based on this real-time monitoring data, using this as a key basis for judging the wall's airtightness level—slow pressure decay and low gas leakage flow rate indicate good wall airtightness; conversely, it indicates a sealing defect in the wall.

[0004] However, traditional airtightness testing methods have significant limitations in practical applications, which to some extent affect the accuracy and reliability of the test results. The core problem lies in the fact that traditional methods rely solely on pressure changes and gas flow rate to determine wall airtightness, failing to adequately consider other factors that may influence the results. Specifically, in actual testing environments, interference from the external environment (such as irregular airflow within the testing site, sudden fluctuations in external air pressure, etc.), the complex structure of the radiation shielding wall itself (such as multiple layers of different materials spliced ​​together, internal pipes running through it, etc.), and localized temperature and humidity anomalies in the testing area can all interfere with the test data. For example, unexpected pressure fluctuations in the testing environment can directly lead to deviations in pressure decay data, making it difficult for testing personnel to accurately determine whether the pressure change is caused by wall leaks or environmental pressure fluctuations. Furthermore, when the wall structure is uneven, or has multiple layers or significant local temperature differences, the flow of gas within the wall becomes complex. In such cases, pressure measurements alone cannot accurately distinguish subtle leaks—pressure changes caused by temperature differences may be misjudged as leaks, or pressure changes caused by minute leaks may be missed. Ultimately, this makes it difficult to detect local leaks in a timely and accurate manner, or even leads to misjudgments, severely reducing the detection rate of minor airtightness anomalies. These problems not only affect the comprehensiveness of the test results but also introduce potential risks into safety measures developed based on the test results, failing to fully meet the safety requirements for radiation protection walls. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an image fusion-based method for detecting the airtightness of radiation shielding walls.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting the airtightness of a radiation shielding wall based on image fusion, comprising the following steps:

[0007] S1: Obtain multi-band images of the entire radiation shielding wall using a multispectral imaging device, collect temperature and humidity values ​​at each detection point, input the parameters into a wavelet transform algorithm for feature extraction, and construct a time series set of temperature and humidity features of the radiation shielding wall.

[0008] S2: By calling the time series set of temperature and humidity features of the radiation shielding wall, the continuous time series temperature and humidity parameter changes of each detection point are obtained by differential calculation, a two-dimensional rate of change vector is constructed, the vector is input into the sliding window extreme value detection algorithm for screening, and the temperature and humidity rate of change vector sequence of the radiation shielding wall detection points is output.

[0009] S3: By calling the temperature and humidity change rate vector sequence of the radiation shielding wall detection points, Euclidean distance and vector angle are applied to the temperature and humidity change rate vectors of spatially adjacent detection points to determine the clustering based on the synchronization trend, and the detection results of suspected airtightness anomalies are output.

[0010] S4: Based on the detection results of the suspected airtightness anomaly area, collect the continuous time-series temperature and humidity change rate vector of the detection points in the area, call the polynomial trend line fitting to model the vector trend line, input the trend line slope and fitting residual into the trend line consistency discrimination model, and output the location and dynamic tracking results of the airtightness anomaly of the radiation protection wall.

[0011] As a further aspect of the present invention, the time series set of temperature and humidity characteristics of the radiation shielding wall includes temperature distribution parameters, humidity distribution parameters, and time series index; the temperature and humidity change rate vector sequence of the radiation shielding wall detection points includes temperature change rate vector, humidity change rate vector, and time series change label; the detection results of the suspected airtightness anomaly area include the spatial coordinates of the detection point, the change rate synchronization group, and the anomaly clustering label; and the airtightness anomaly location and dynamic tracking results of the radiation shielding wall include the location of the airtightness anomaly detection point, dynamic trajectory parameters, and trend line fitting parameters.

[0012] As a further aspect of the present invention, step S1 includes:

[0013] The infrared and visible light images of the entire radiation shielding wall are continuously acquired by the multispectral imaging device within a preset time interval, and the real-time temperature and humidity values ​​of all preset detection points on the surface of the radiation shielding wall are monitored simultaneously to generate the original multimodal dataset.

[0014] For the real-time temperature and humidity value sequences in the original multimodal dataset, the multi-scale wavelet decomposition algorithm is applied to calculate the approximation coefficients and detail coefficients of each sequence at different decomposition levels, and a multi-scale feature coefficient set is obtained.

[0015] Calculate the energy of the approximation coefficients and the energy of the detail coefficients at each level in the multi-scale feature coefficient set. Use the energy of the approximation coefficients as the temperature distribution parameter and the humidity distribution parameter, and combine them with the time series information of the multi-band images to jointly establish the time series set of temperature and humidity features of the radiation shielding wall.

[0016] As a further aspect of the present invention, step S2 includes:

[0017] The temperature and humidity distribution parameters at two consecutive time points are extracted by calling the time series set of temperature and humidity characteristics of the radiation shielding wall.

[0018] The temperature distribution parameter at the next time point is calculated to be different from the temperature distribution parameter at the previous time point to obtain the temperature change difference. The humidity distribution parameter at the next time point is calculated to be different from the humidity distribution parameter at the previous time point to obtain the humidity change difference. The temperature change difference and the humidity change difference are combined into the two-dimensional rate of change vector.

[0019] Set the sliding window size and the extreme value judgment threshold, move the sliding window on the time series composed of the two-dimensional rate of change vector, determine whether the maximum value of the vector magnitude in the window exceeds the extreme value judgment threshold, filter out the two-dimensional rate of change vectors that exceed the threshold, and obtain the temperature and humidity rate of change vector sequence of the radiation protection wall detection point.

[0020] As a further aspect of the present invention, step S3 includes:

[0021] The temperature and humidity change rate vector sequence of the radiation shielding wall detection points is invoked, and the set of spatially adjacent detection points for each detection point is determined based on the spatial coordinates of the detection points.

[0022] Calculate the Euclidean distance and cosine similarity between the temperature and humidity change rate vector of the target detection point and the temperature and humidity change rate vectors of each detection point in the set of spatially adjacent detection points, and generate multidimensional associated feature parameters.

[0023] Set an upper limit threshold for distance and a lower limit threshold for similarity. Determine whether the Euclidean distance is less than the upper limit threshold for distance and whether the vector cosine similarity is greater than the lower limit threshold for similarity. Detection points that meet the conditions are identified as synchronously changing point pairs.

[0024] Density clustering algorithm is applied to aggregate and analyze all the synchronous change point pairs. The set of detection points that are spatially connected and have the same synchronous change trend is marked as the rate of change synchronization group, and the detection results of the suspected airtightness anomaly area are generated.

[0025] As a further aspect of the present invention, step S4 includes:

[0026] Based on the detection results of the suspected airtightness anomaly area, extract the vector sequence of temperature and humidity change rate of all detection points in the synchronous change rate group within the continuous time period of the radiation shield wall detection point;

[0027] For the temperature and humidity change rate vector sequence at each detection point, a polynomial fitting is performed using the least squares method to establish a unique polynomial trend line for the temperature and humidity change trend at each detection point, and the slope of the trend line and the fitting residual are calculated.

[0028] Based on the preset baseline slope and residual threshold, the deviation of the trend line slope and the dispersion of the fitted residual are analyzed. When both the slope deviation and the residual dispersion exceed the preset threshold, the detection point is determined to be an airtightness anomaly point.

[0029] The location information of all the airtightness anomalies is integrated, and their location changes over a continuous period of time are recorded to generate the location and dynamic tracking results of the airtightness anomalies of the radiation shielding wall.

[0030] As a further aspect of the present invention, the determination of the synchronization trend is achieved by calculating a synchronization trend discrimination index. Finish;

[0031] The synchronous trend discrimination indicator The calculation formula is:

[0032] ;

[0033] in, For target detection points Detection points adjacent to space Indicators for judging synchronous trends between them;

[0034] These are the Euclidean distance weighting coefficients. The cosine value of the angle between the vectors is the weighting coefficient, and ;

[0035] For testing points temperature and humidity change rate vector With the testing point temperature and humidity change rate vector The Euclidean distance between them;

[0036] The preset maximum normalized distance;

[0037] Calculate the synchronous trend discriminant index Then, it is compared with a preset synchronization trend determination threshold. If the threshold is... If the value is greater than the synchronization trend determination threshold, then the two detection points are determined to have a synchronization trend.

[0038] As a further aspect of the present invention, the trend line consistency discrimination model calculates the trend line deviation index. accomplish;

[0039] The trendline deviation index The calculation formula is:

[0040] ;

[0041] in, For the first The trend line deviation index of the detection points in the suspected airtightness anomaly area;

[0042] This is the weighting factor for the relative slope deviation of the trend line. This is the weighting factor for the root mean square error term of the normalized fitting residuals;

[0043] For the first The slope of the polynomial trend line fitting for each suspected airtightness anomaly detection point;

[0044] Let be the slope of the baseline trend line for temperature and humidity changes in the radiation shielding wall under normal airtight conditions, and its absolute value. Not zero;

[0045] For the first Each detection point at the time point The difference between the actual temperature and humidity change rate vector and the fitted value of the polynomial trend line, i.e., the fitting residual;

[0046] This represents the total number of continuous time-series temperature and humidity rate vectors used for fitting;

[0047] is the root mean square error of the baseline fitting residual under normal airtightness conditions, and Not zero;

[0048] Calculate the trendline deviation index Then, it is compared with a preset trendline consistency threshold. If the... If the value is greater than the trend line consistency threshold, then the detection point is finally confirmed as the airtightness anomaly detection point.

[0049] As a further aspect of the present invention, the specific implementation of the sliding window extreme value detection algorithm is as follows:

[0050] The sliding window size is set to an integer greater than 2, where the integer is the length of the time series segment for extreme value detection;

[0051] The movement step size of the sliding window is set to a positive integer, which is a parameter that controls the degree of overlap between two adjacent detections;

[0052] For each detection point, the complete time series is composed of the two-dimensional rate of change vector. Starting from the initial moment, vector subsequences with a length equal to the sliding window size are successively extracted according to the moving step size.

[0053] Within each of the vector subsequences, the Euclidean norm of all two-dimensional rate of change vectors is calculated, and the vector with the largest norm value is identified as the local extremum vector of the current window.

[0054] The Euclidean norm of the local extreme value vector is compared with the preset extreme value determination threshold, which is the upper limit of normal temperature and humidity fluctuations obtained based on historical data statistics.

[0055] If the Euclidean norm of the local extremum vector is greater than the extremum determination threshold, the vector is retained and the time-series change label is added to it. Finally, all the retained vectors together constitute the temperature and humidity change rate vector sequence of the radiation protection wall detection points.

[0056] As a further aspect of the present invention, the application of the multi-scale wavelet decomposition algorithm to calculate the approximation coefficients and detail coefficients of each sequence at different decomposition levels is specifically as follows:

[0057] The Daubechies wavelet basis function is selected to perform multi-level decomposition on the real-time temperature value sequence and the real-time humidity value sequence at each detection point. The number of decomposition layers is adaptively determined or preset according to the non-stationary characteristics of the signal.

[0058] At each decomposition level, the signal is decomposed into the approximation coefficients representing the low-frequency trend of the signal and the detail coefficients representing the high-frequency abrupt changes of the signal.

[0059] Extract the approximation coefficients from the last level of decomposition;

[0060] Extract the detail coefficients from all decomposition levels;

[0061] The extracted approximation coefficients and the detail coefficients of all levels are combined to form a feature vector, generating a multi-scale temperature and humidity dynamic feature vector for the detection point.

[0062] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0063] In this invention, temperature and humidity data of the entire wall area are extracted in real time by fusing multispectral image information. Further, the rate of change analysis and trend line modeling are performed by combining the sequence change trend and spatial distribution characteristics. This enables more accurate and efficient identification and location of suspected airtightness anomalies. In dynamic monitoring, continuous tracking and accurate reflection of changes in the airtightness performance of the wall are achieved, thereby improving the sensitivity and reliability of detection, significantly reducing the probability of missed detection and false judgment, and ensuring the accuracy of detection results even in complex environments. This effectively supports the comprehensive evaluation of the airtightness status of the wall and early risk warning, and strengthens the overall sealing performance assessment and dynamic management capabilities. Attached Figure Description

[0064] Figure 1 This is a flowchart of the overall process for the image fusion-based radiation shielding wall airtightness detection method of the present invention;

[0065] Figure 2 This is a flowchart illustrating the assembly of the time-series set of temperature and humidity characteristics of the radiation shielding wall according to the present invention.

[0066] Figure 3 This is a flowchart of the temperature and humidity change rate vector sequence screening process at the radiation shielding wall detection points of the present invention.

[0067] Figure 4 This is a flowchart of the detection process for suspected airtightness anomalies in this invention.

[0068] Figure 5 This is a flowchart of the process for locating and dynamically tracking abnormal air tightness of the radiation shielding wall according to the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0070] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0071] Please see Figure 1 and Figure 2 This invention provides a technical solution: a method for detecting the airtightness of a radiation shielding wall based on image fusion, comprising the following steps:

[0072] S1: Obtain multi-band images of the entire radiation shielding wall using a multispectral imaging device, collect temperature and humidity values ​​at each detection point, input the parameters into a wavelet transform algorithm for feature extraction, and construct a time series set of temperature and humidity features of the radiation shielding wall.

[0073] The infrared and visible light images of the entire radiation shielding wall are continuously acquired by the multispectral imaging device within a preset time interval, and the real-time temperature and humidity values ​​of all preset detection points on the surface of the radiation shielding wall are monitored simultaneously to generate the original multimodal dataset.

[0074] For the real-time temperature and humidity sequences in the original multimodal dataset, the multi-scale wavelet decomposition algorithm is applied to calculate the approximation coefficients and detail coefficients of each sequence at different decomposition levels, and a multi-scale feature coefficient set is obtained.

[0075] The Daubechies wavelet basis function is selected to perform multi-level decomposition on the real-time temperature and humidity value sequences of each detection point. The number of decomposition levels is adaptively determined or preset according to the non-stationary characteristics of the signal.

[0076] At each decomposition level, the signal is decomposed into approximate coefficients representing the low-frequency trend of the signal and detail coefficients representing the high-frequency abrupt changes of the signal.

[0077] Extract the approximate coefficients of the last level of decomposition;

[0078] Extract the detail coefficients of all decomposition levels;

[0079] The extracted approximation coefficients and the detail coefficients of all levels are combined to form a feature vector, generating a multi-scale temperature and humidity dynamic feature vector for the detection point.

[0080] Calculate the energy of the approximation coefficients and the energy of the detail coefficients at each level in the multi-scale feature coefficient set. Use the approximation coefficient energy as temperature distribution parameters and humidity distribution parameters, and combine it with the temporal information of multi-band images to jointly establish a temporal set of temperature and humidity characteristics of the radiation shielding wall.

[0081] The time series set of temperature and humidity characteristics of the radiation shielding wall includes temperature distribution parameters, humidity distribution parameters, and time series index.

[0082] In the specific implementation, at 08:00 on July 25, 2025, a multispectral imaging device was activated to collect data on a radiation shielding wall measuring 10 meters × 5 meters. The wall surface was pre-laid with a 10×5 grid, totaling 50 detection points, with a spacing of 1 meter between the detection points. The multispectral imaging device scanned the wall continuously for 24 hours at preset time intervals of 10 minutes, acquiring infrared (8-14μm) and visible light (400-700nm) images. Simultaneously, high-precision temperature and humidity sensors (temperature accuracy ±0.1℃, humidity accuracy ±1%RH) attached to each detection point synchronously recorded real-time temperature and humidity values. These images and data together generated the original multimodal dataset.

[0083] Feature extraction was performed on the time-series data in the original multimodal dataset. Taking detection point P23 as an example, the seven consecutive real-time temperature values ​​collected between 08:00 and 09:00 are [22.5, 22.6, 22.5, 22.7, 22.8, 22.6, 22.9] (unit: ℃), and the real-time humidity values ​​are [45.1, 45.3, 45.2, 45.5, 45.6, 45.4, 45.7] (unit: %RH). The Daubechies4 (db4) wavelet basis function was selected to perform a three-level decomposition on the temperature sequence. This decomposition level was based on Fourier analysis of more than 1000 hours of normal wall temperature and humidity historical data, which determined that the main fluctuation frequency was concentrated in the low-frequency region. The three-level decomposition was sufficient to separate the background trend from the instantaneous disturbance. The decomposition process is as follows: The first-level decomposition decomposes the original 7-point temperature sequence into approximation coefficient cA1 and detail coefficient cD1; the second-level decomposition decomposes cA1 into approximation coefficient cA2 and detail coefficient cD2; the third-level decomposition decomposes cA2 into approximation coefficient cA3 and detail coefficient cD3. After calculation, the approximation coefficient cA3=[32.15,32.28] of the last level, and the detail coefficients of all levels are cD3=[-0.08,-0.09], cD2=[-0.15,0.05,0.11], and cD1=[-0.07,0.14,-0.07,0.21]. These coefficients are combined sequentially to generate a portion of the multi-scale temperature and humidity dynamic feature vector for detection point P23 at 09:00, specifically the temperature feature vector [32.15, 32.28, -0.08, -0.09, -0.15, 0.05, 0.11, -0.07, 0.14, -0.07, 0.21]. Similarly, the humidity sequence is processed in the same way to obtain the humidity feature vector.

[0084] Next, the energy of the multi-scale characteristic coefficient set is calculated. The energy of the approximation coefficients is calculated as the sum of the squares of all approximation coefficients at that level, and the energy of the detail coefficients is calculated as the sum of the squares of all detail coefficients at that level. For the temperature sequence above, the energy of its third-level approximation coefficients is... The energy of this approximation coefficient is defined as the temperature distribution parameter at that moment. Similarly, performing the same wavelet decomposition and energy calculation on the humidity sequence at detection point P23, if its third-level approximation coefficient is [64.21, 64.53], then its energy is... This value is defined as the humidity distribution parameter. The temperature and humidity distribution parameters of all detection points at each acquisition time point are combined with the corresponding time series index (e.g., timestamp in YYYY-MM-DDHH:MM:SS format) to finally construct the time series set of temperature and humidity characteristics of the radiation shielding wall.

[0085] The table below shows the time series of temperature and humidity characteristics of some detection points at three consecutive time points.

[0086] Table 1. Examples of time-series fragments showing the temperature and humidity characteristics of radiation shielding walls.

[0087]

[0088] As shown in Table 1, this time series set provides a structured data foundation for subsequent dynamic change analysis, which includes temperature distribution parameters, humidity distribution parameters, and time series index.

[0089] Please see Figure 1 and Figure 3 S2: By calling the time series set of temperature and humidity features of the radiation shielding wall, the continuous time series temperature and humidity parameter changes of each detection point are obtained by differential calculation, a two-dimensional rate of change vector is constructed, the vector is input into the sliding window extreme value detection algorithm for screening, and the temperature and humidity rate of change vector sequence of the radiation shielding wall detection points is output.

[0090] Call the time series set of temperature and humidity characteristics of the radiation shielding wall to extract the temperature distribution parameters and humidity distribution parameters at two consecutive time points;

[0091] The temperature distribution parameters at the next time point are calculated to obtain the temperature change difference. The humidity distribution parameters at the next time point are calculated to obtain the humidity change difference. The temperature change difference and humidity change difference are combined into a two-dimensional rate of change vector.

[0092] Set the sliding window size and extreme value judgment threshold, move the sliding window on the time series composed of two-dimensional rate of change vectors, judge whether the maximum value of the vector magnitude in the window exceeds the extreme value judgment threshold, filter out the two-dimensional rate of change vectors that exceed the threshold, and obtain the temperature and humidity rate of change vector sequence of the radiation shielding wall detection points.

[0093] The specific implementation of the sliding window extreme value detection algorithm is as follows: the sliding window size is set to an integer greater than 2, where the integer is the length of the time series segment for extreme value detection;

[0094] Set the sliding window's movement step size to a positive integer, where the positive integer is a parameter that controls the degree of overlap between two adjacent detections;

[0095] For the complete time series of each detection point consisting of a two-dimensional rate of change vector, starting from the initial moment, vector subsequences with a length equal to the sliding window size are successively extracted according to the moving step size;

[0096] Within each vector subsequence, calculate the Euclidean norm of all two-dimensional rate of change vectors, and identify the vector with the largest norm value as the local extremum vector of the current window.

[0097] The Euclidean norm of the local extreme value vector is compared with the preset extreme value judgment threshold, which is the upper limit of normal temperature and humidity fluctuations obtained based on historical data statistics.

[0098] If the Euclidean norm of the local extremum vector is greater than the extremum determination threshold, the vector is retained and a time-series change label is added to it. Finally, all the retained vectors together constitute the temperature and humidity change rate vector sequence of the radiation shielding wall detection points.

[0099] The temperature and humidity change rate vector sequence at the radiation shielding wall detection points includes a temperature change rate vector, a humidity change rate vector, and a time-series change label.

[0100] In the specific implementation, the time series set of temperature and humidity characteristics of the radiation shielding wall generated in step S1 is first called, as shown in Table 1. Taking detection point P23 as an example, the temperature distribution parameters and humidity distribution parameters at two consecutive time points, 09:10:00 and 09:00:00, are extracted. The temperature distribution parameter at 09:10:00 is 2083.81, and the humidity distribution parameter is 8268.54; the temperature distribution parameter at 09:00:00 is 2075.61, and the humidity distribution parameter is 8287.04. The difference between the parameters at the later time point and the earlier time point is calculated: the temperature change difference is... The difference in humidity is Combining these two differences, we can construct the two-dimensional rate of change vector of detection point P23 at 09:10:00. This calculation is repeated for all detection points at all consecutive time points to obtain a time series consisting of a two-dimensional rate of change vector.

[0101] Next, a sliding window extreme value detection was performed on the time series. The sliding window size was set to 5, and the step size to 1. The sliding window size was determined based on the following: in the retrospective analysis of historical leakage events, it was found that the abnormal temperature and humidity disturbances caused by gas leaks usually reached their peak within 30-50 minutes. With a sampling interval of 10 minutes, a window length of 5 time points (i.e., 50 minutes) was sufficient to capture this change process. Setting the step size to 1 allows for detailed extreme value judgment at each time point. The extreme value judgment threshold was set based on a two-dimensional rate of change vector dataset obtained from three months of continuous monitoring of a reference wall with confirmed airtightness. The Euclidean norm of all vectors in this dataset was calculated, and its probability distribution histogram was plotted. The 99.9% quantile of this distribution was taken as the threshold, which was calculated to be 15.0. This threshold represents the maximum amplitude of the combined temperature and humidity change caused by factors such as natural environmental fluctuations and equipment noise under normal operating conditions.

[0102] Taking the time period from 09:10:00 to 09:50:00 at detection point P23 as an example, its two-dimensional rate of change vector sequence is as follows: , , , , The first sliding window covers the five time points from 09:10 to 09:50. The Euclidean norm (i.e., modulus) of these five vectors is calculated respectively: Within this window, the vector with the largest norm is Its norm is 20.24. The norm of this local extremum vector is compared with the preset extremum threshold of 15.0. Because... Therefore, the vector is retained. The data is then labeled with a time-series change tag, "T0910_EXCEED". The window is moved forward one step, and the new window will cover the data from 09:20 to 10:00. The above calculation and judgment process is repeated. Finally, all the retained vectors and their labels together constitute the temperature and humidity change rate vector sequence of the radiation shielding wall detection points. This sequence includes the temperature change rate vector, the humidity change rate vector, and the time-series change tag.

[0103] Please see Figure 1 and Figure 4 S3: By calling the temperature and humidity change rate vector sequence of the radiation shielding wall detection points, Euclidean distance and vector angle are applied to the temperature and humidity change rate vectors of spatially adjacent detection points to determine the clustering based on the synchronization trend, and the detection results of suspected airtightness anomaly areas are output.

[0104] Call the vector sequence of temperature and humidity change rate at the radiation shielding wall detection points, and determine the set of spatially adjacent detection points for each detection point based on the spatial coordinates of the detection points;

[0105] Calculate the Euclidean distance and cosine similarity between the temperature and humidity change rate vector of the target detection point and the temperature and humidity change rate vectors of each detection point in the set of spatially adjacent detection points, and generate multidimensional associated feature parameters.

[0106] Set an upper threshold for distance and a lower threshold for similarity. Determine whether the Euclidean distance is less than the upper threshold for distance and whether the vector cosine similarity is greater than the lower threshold for similarity. Detection points that meet the conditions are identified as synchronously changing point pairs.

[0107] The determination of synchronous trends is achieved by calculating the synchronous trend discriminant index. Finish;

[0108] Synchronous trend discriminant indicator The calculation formula is: ;

[0109] in, For target detection points Detection points adjacent to space Indicators for judging synchronous trends between them;

[0110] These are the Euclidean distance weighting coefficients. Let be the weighting coefficient of the cosine value of the angle between the vectors, and ;

[0111] For testing points temperature and humidity change rate vector With the testing point temperature and humidity change rate vector The Euclidean distance between them;

[0112] The preset maximum normalized distance;

[0113] Calculate the synchronous trend discriminant index Then, it is compared with the preset synchronization trend judgment threshold. If If the value is greater than the synchronization trend determination threshold, then the two detection points are determined to have a synchronization trend.

[0114] Density clustering algorithm is applied to aggregate analysis of all synchronous change point pairs. The set of detection points that are spatially connected and have the same synchronous change trend is marked as the rate of change synchronization group, and the detection results of suspected airtightness anomaly areas are generated.

[0115] The detection results for suspected airtightness anomalies include the spatial coordinates of the detection point, the rate of change synchronization group, and the anomaly cluster label.

[0116] In the specific implementation, the temperature and humidity change rate vector sequence of the radiation shielding wall detection points output in step S2 is called. Assuming that at time 09:10:00, detection point P23 and its spatially adjacent detection points P22, P24, P13, and P33 are all selected as detection points with significant changes, their temperature and humidity change rate vectors are as follows: , , , , Taking the synchronization trend discrimination between target detection point P23 and its spatially adjacent detection point P22 as an example, the following formula is used for calculation: .

[0117] The innovation of this formula lies in constructing a comprehensive index that simultaneously measures both the similarity of change "intensity" and the consistency of change "direction" by weighted fusion of normalized Euclidean distance and vector cosine similarity. The first term... The distance between vectors is converted into a similarity metric between 0 and 1; the closer the distance, the higher the similarity. (Second term) Directly calculating the cosine of the angle between the vectors measures the consistency of the direction of temperature and humidity changes. Weighting coefficients. and The introduction of this technology allows for adjustments to the emphasis on intensity and direction based on the characteristics of the actual physical process (e.g., whether the leak is a diffusion process or a directional jetting process), thereby improving the flexibility and accuracy of the judgment.

[0118] Before calculation, the parameters in the formula need to be determined. Let the target detection point be... Spatial adjacent detection points are Weighting coefficients and The optimal value was determined through grid search optimization on a validation dataset containing 50 known leaked samples and 200 non-leaked samples. The search objective was to maximize... The discriminant strength between leaked and non-leaked sample pairs. Experimental results show that when When the model achieves its highest F1 score on the validation set, then... Therefore, set... ,satisfy . This is the preset maximum normalized distance. Analysis of historical data shows that, under extreme conditions (such as direct air conditioning in summer and proximity to heating in winter), the norm of the temperature and humidity change rate vector can reach a maximum of 40. The Euclidean distance between two opposing such vectors is 80. A safety margin is defined. . and This is the vector obtained from S2: and .

[0119] The calculation process is as follows: 1. Calculate the Euclidean distance between the two vectors. 2. Calculate the norm of two vectors: (Calculations from S2) 3. Calculate the dot product of two vectors. 4. Substitute the values ​​into the formula

[0120] The synchronous trend determination threshold is calculated based on all neighboring point pairs in the validation dataset. The values ​​were set for analysis. By plotting the ROC curve, the point corresponding to the maximum Youden index was selected. The value was used as a threshold and set to 0.80. The calculated value... Compared to the synchronization trend determination threshold of 0.80. Because Therefore, P23 and P22 are determined to have a synchronous trend, forming a synchronous change point pair. This process is repeated for P23 and its other neighboring points (P24, P13, P33). For example, calculate the relationship between P23 and P24. The values ​​will be much lower than 0.80 because their vectors differ significantly in both magnitude and direction. And the values ​​of P23 and P13... Value will be with Similar to each other, both exceeding 0.80. After identifying all anomalous adjacent point pairs, a density-based spatial clustering algorithm (DBSCAN) is applied. Each detection point is considered a spatial sample point; if two points form a synchronous change pair, their "distance" is considered sufficiently close. This algorithm aggregates spatially connected detection points that meet the synchronous trend discrimination criteria (e.g., P22, P23, P13) into a rate-of-change synchronization group, labeled as "suspected anomalous cluster A". The result includes the spatial coordinates of the detection points, the rate-of-change synchronization group, and the anomalous cluster label.

[0121] Please see Figure 1 and Figure 5 S4: Based on the detection results of suspected airtightness anomalies, the continuous time-series temperature and humidity change rate vector of the detection points in the area is collected, and the multinomial trend line fitting is called to model the vector trend line. The slope of the trend line and the fitting residual are input into the trend line consistency discrimination model, and the results of the location and dynamic tracking of the airtightness anomaly of the radiation protection wall are output.

[0122] Based on the detection results of suspected airtightness anomalies, extract the vector sequence of temperature and humidity change rates of all detection points in the synchronous group within the continuous time period of the radiation shield wall detection points.

[0123] For the temperature and humidity change rate vector sequence of each detection point, the least squares method is used to perform polynomial fitting, a unique polynomial trend line is established for the temperature and humidity change trend of each detection point, and the slope of the trend line and the fitting residual are calculated.

[0124] Based on the preset baseline slope and residual threshold, the deviation of the trend line slope and the dispersion of the fitted residual are analyzed. When both the slope deviation and the residual dispersion exceed the preset threshold, the detection point is determined to be an airtightness anomaly.

[0125] The trend line consistency discrimination model calculates the trend line deviation index. accomplish;

[0126] Trendline Deviation Indicator The calculation formula is: ;

[0127] in, For the first The trend line deviation index of the detection points in the suspected airtightness anomaly area;

[0128] This is the weighting factor for the relative slope deviation of the trend line. This is the weighting factor for the root mean square error term of the normalized fitting residuals;

[0129] For the first The slope of the polynomial trend line fitted to the detection points of the suspected airtightness anomaly area;

[0130] Let be the slope of the baseline trend line for temperature and humidity changes in the radiation shielding wall under normal airtight conditions, and its absolute value. Not zero;

[0131] For the first Each detection point at the time point The difference between the actual temperature and humidity change rate vector and the fitted value of the polynomial trend line, i.e., the fitting residual;

[0132] This represents the total number of continuous time-series temperature and humidity rate vectors used for fitting;

[0133] is the root mean square error of the baseline fitting residual under normal airtightness conditions, and Not zero;

[0134] Calculate the trendline deviation index Then, it is compared with a preset trendline consistency threshold. If... If the value is greater than the trend line consistency threshold, then the detection point is finally confirmed as an airtightness anomaly detection point;

[0135] Integrate the location information of all airtightness anomalies and record their location changes over a continuous period of time to generate the location and dynamic tracking results of airtightness anomalies in the radiation shielding wall.

[0136] The results of the location and dynamic tracking of airtightness anomalies in the radiation shielding wall include the location of the airtightness anomaly detection point, dynamic trajectory parameters, and trend line fitting parameters.

[0137] In practice, based on the detection results of suspected airtightness anomalies output by S3, "suspected anomaly cluster A" is identified. This cluster includes detection points P13, P22, and P23. These three detection points are then analyzed at 09:10:00 to 11:00:00 (a total of 12 10-minute time points). The temperature and humidity change rate vector sequence within the detection point P23. Taking the detection point P23 as an example, the Euclidean norm sequence of its 12 continuous vectors is: [20.24,19.36,6.12,3.54,3.12,2.80,2.55,2.40,2.31,2.25,2.20,2.18].

[0138] For the norm sequence of P23, a first-order polynomial (linear) fitting was performed using the least squares method to establish its temperature and humidity variation trend line. The fitting result is as follows: ,in Indexed by time points (1 to 12). This represents the norm value for the prediction. The slope of the trend line. (in The value for the P23 test point is -1.52. Simultaneously, the difference between the actual norm value and the trend line predicted value at each time point is calculated, i.e., the fitting residual. For example, at the first point in time ( The predicted value is The actual value is 20.24, then the residual... Calculate the residuals for all 12 points to obtain the residual sequence.

[0139] Next, the trendline consistency discrimination model is invoked to calculate the trendline deviation index. Its formula is: ;

[0140] The innovation of this formula lies in its approach: instead of viewing trends or noise in isolation, it weights and combines the "anomaly" of the trend (deviation from the baseline trend) with the "uncertainty" of the trend (the degree of dispersion of data points around the trend). This dual consideration makes the discrimination more robust: a pattern with an anomalous trend slope but highly dispersed data points (possibly caused by random disturbances) can be effectively distinguished from a pattern with an anomalous slope and data points closely following the trend (more likely caused by a continuous physical process such as leakage).

[0141] calculate Before proceeding, the following parameters need to be determined: and The weighting factor is determined by receiver operating characteristic (ROC) analysis on a test dataset containing true leakage (positive samples) and false positives (negative samples) caused by transient interference. The goal is to maximize the area under the curve (AUC) that distinguishes between the two classes. Experimental validation shows that when... When the time is right, the judgment effect is optimal, that is, it focuses more on the abnormality of the trend itself. and These are the baseline parameters. These values ​​were obtained through a 6-month long-term monitoring of a standard reference wall (confirmed to be leak-free). The trend line slope and root mean square error (RMSE) of the fitted residuals were calculated for all monitoring points on the wall using the procedure described above, resulting in a large dataset. Take the average of all slopes. Take the average of all RMSEs. The table below shows some of the experimental data used to calculate the baseline value.

[0142] Table 2. Sample of experimental data for baseline parameter calculation.

[0143]

[0144] As shown in Table 2, after statistical analysis of thousands of such samples, the final determination was made. (Under normal circumstances, the trend of temperature and humidity changes is close to horizontal), and its absolute value is not zero; (The level of data fluctuation under normal circumstances), its value is not zero.

[0145] The calculation process for P23 is as follows: 1. , 2. Calculate the root mean square error (RMSE) of the fitting residuals: Assuming the sum of squares of all 12 residuals is calculated to be 210.25, then RMSE = 3. , 4. Substitute all values ​​into formula:

[0146] The trendline consistency threshold is calculated based on all... The value is set. The positive and negative samples... The value distributions were compared, and a threshold that could distinguish the two with a 95% confidence level was selected; this threshold was set to 3.0. The calculated... Compare with the trendline consistency threshold of 3.0. Because Therefore, detection point P23 was ultimately confirmed as the airtightness anomaly detection point. The same calculation was performed on P13 and P22 in cluster A, if their... If the value also exceeds the threshold, it is also identified as an anomaly. The location information of all identified airtightness anomalies (P13, P22, P23) over a continuous period is integrated, and their anomaly status at different times is recorded to form a dynamic trajectory. The final output of the radiation shielding wall airtightness anomaly location and dynamic tracking results includes the set of location coordinates of {P13, P22, P23}, the dynamic trajectory parameters of their status changes over time, and their respective trend line fitting parameters (such as slope -1.52, RMSE 4.185, etc.).

[0147] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. A method for detecting the airtightness of radiation shielding walls based on image fusion, characterized in that, Includes the following steps: S1: Obtain multi-band images of the entire radiation shielding wall using a multispectral imaging device, collect temperature and humidity values ​​at each detection point, input the parameters into a wavelet transform algorithm for feature extraction, and construct a time series set of temperature and humidity features of the radiation shielding wall. S2: By calling the time series set of temperature and humidity features of the radiation shielding wall, the continuous time series temperature and humidity parameter changes of each detection point are obtained by differential calculation, a two-dimensional rate of change vector is constructed, the vector is input into the sliding window extreme value detection algorithm for screening, and the temperature and humidity rate of change vector sequence of the radiation shielding wall detection points is output. S3: By calling the temperature and humidity change rate vector sequence of the radiation shielding wall detection points, Euclidean distance and vector angle are applied to the temperature and humidity change rate vectors of spatially adjacent detection points to determine the clustering based on the synchronization trend, and the detection results of suspected airtightness anomalies are output. S4: Based on the detection results of the suspected airtightness anomaly area, collect the continuous time-series temperature and humidity change rate vector of the detection points in the area, call the polynomial trend line fitting to model the vector trend line, input the trend line slope and fitting residual into the trend line consistency discrimination model, and output the location and dynamic tracking results of the airtightness anomaly of the radiation protection wall.

2. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 1, characterized in that, The time series set of temperature and humidity characteristics of the radiation shielding wall includes temperature distribution parameters, humidity distribution parameters, and time series index. The temperature and humidity change rate vector sequence of the radiation shielding wall detection points includes temperature change rate vector, humidity change rate vector, and time series change label. The detection results of the suspected airtightness anomaly area include the spatial coordinates of the detection point, the change rate synchronization group, and the anomaly clustering label. The airtightness anomaly location and dynamic tracking results of the radiation shielding wall include the location of the airtightness anomaly detection point, dynamic trajectory parameters, and trend line fitting parameters.

3. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 2, characterized in that, Step S1 includes: The infrared and visible light images of the entire radiation shielding wall are continuously acquired by the multispectral imaging device within a preset time interval, and the real-time temperature and humidity values ​​of all preset detection points on the surface of the radiation shielding wall are monitored simultaneously to generate the original multimodal dataset. For the real-time temperature and humidity value sequences in the original multimodal dataset, the multi-scale wavelet decomposition algorithm is applied to calculate the approximation coefficients and detail coefficients of each sequence at different decomposition levels, and a multi-scale feature coefficient set is obtained. Calculate the energy of the approximation coefficients and the energy of the detail coefficients at each level in the multi-scale feature coefficient set. Use the energy of the approximation coefficients as the temperature distribution parameter and the humidity distribution parameter, and combine them with the time series information of the multi-band images to jointly establish the time series set of temperature and humidity features of the radiation shielding wall.

4. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 2, characterized in that, Step S2 includes: The temperature and humidity distribution parameters at two consecutive time points are extracted by calling the time series set of temperature and humidity characteristics of the radiation shielding wall. The temperature distribution parameter at the next time point is calculated to be different from the temperature distribution parameter at the previous time point to obtain the temperature change difference. The humidity distribution parameter at the next time point is calculated to be different from the humidity distribution parameter at the previous time point to obtain the humidity change difference. The temperature change difference and the humidity change difference are combined into the two-dimensional rate of change vector. Set the sliding window size and the extreme value judgment threshold, move the sliding window on the time series composed of the two-dimensional rate of change vector, determine whether the maximum value of the vector magnitude in the window exceeds the extreme value judgment threshold, filter out the two-dimensional rate of change vectors that exceed the threshold, and obtain the temperature and humidity rate of change vector sequence of the radiation protection wall detection point.

5. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 2, characterized in that, Step S3 includes: The temperature and humidity change rate vector sequence of the radiation shielding wall detection points is invoked, and the set of spatially adjacent detection points for each detection point is determined based on the spatial coordinates of the detection points. Calculate the Euclidean distance and cosine similarity between the temperature and humidity change rate vector of the target detection point and the temperature and humidity change rate vectors of each detection point in the set of spatially adjacent detection points, and generate multidimensional associated feature parameters. Set an upper limit threshold for distance and a lower limit threshold for similarity. Determine whether the Euclidean distance is less than the upper limit threshold for distance and whether the vector cosine similarity is greater than the lower limit threshold for similarity. Detection points that meet the conditions are identified as synchronously changing point pairs. Density clustering algorithm is applied to aggregate and analyze all the synchronous change point pairs. The set of detection points that are spatially connected and have the same synchronous change trend is marked as the rate of change synchronization group, and the detection results of the suspected airtightness anomaly area are generated.

6. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 2, characterized in that, The S4 step includes: Based on the detection results of the suspected airtightness anomaly area, extract the vector sequence of temperature and humidity change rate of all detection points in the synchronous change rate group within the continuous time period of the radiation shield wall detection point; For the temperature and humidity change rate vector sequence at each detection point, a polynomial fitting is performed using the least squares method to establish a unique polynomial trend line for the temperature and humidity change trend at each detection point, and the slope of the trend line and the fitting residual are calculated. Based on the preset baseline slope and residual threshold, the deviation of the trend line slope and the dispersion of the fitted residual are analyzed. When both the slope deviation and the residual dispersion exceed the preset threshold, the detection point is determined to be an airtightness anomaly. The location information of all the airtightness anomalies is integrated, and their location changes over a continuous period of time are recorded to generate the location and dynamic tracking results of the airtightness anomalies of the radiation shielding wall.

7. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 5, characterized in that, The determination of the synchronization trend is achieved by calculating the synchronization trend discriminant index. Finish; The synchronous trend discrimination indicator The calculation formula is: ； in, For target detection points Detection points adjacent to space Indicators for judging synchronous trends between them; These are the Euclidean distance weighting coefficients. Let be the weighting coefficient of the cosine value of the angle between the vectors, and ; For testing points temperature and humidity change rate vector With the testing point temperature and humidity change rate vector The Euclidean distance between them; The preset maximum normalized distance; Calculate the synchronous trend discriminant index Then, it is compared with a preset synchronization trend determination threshold. If the... If the value is greater than the synchronization trend determination threshold, then the two detection points are determined to have a synchronization trend.

8. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 6, characterized in that, The trendline consistency discrimination model calculates the trendline deviation index. accomplish; The trendline deviation index The calculation formula is: ； in, For the first The trend line deviation index of the detection points in the suspected airtightness anomaly area; This is the weighting factor for the relative slope deviation of the trend line. This is the weighting factor for the root mean square error term of the normalized fitting residuals; For the first The slope of the polynomial trend line fitting for each suspected airtightness anomaly detection point; Let be the slope of the baseline trend line for temperature and humidity changes in the radiation shielding wall under normal airtight conditions, and its absolute value. Not zero; For the first Each detection point at the time point The difference between the actual temperature and humidity change rate vector and the fitted value of the polynomial trend line, i.e., the fitting residual; This represents the total number of continuous time-series temperature and humidity rate vectors used for fitting; is the root mean square error of the baseline fitting residual under normal airtightness conditions, and Not zero; Calculate the trendline deviation index Then, it is compared with a preset trendline consistency threshold. If the... If the value is greater than the trend line consistency threshold, then the detection point is finally confirmed as the airtightness anomaly detection point.

9. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 2, characterized in that, The specific implementation of the sliding window extreme value detection algorithm is as follows: The sliding window size is set to an integer greater than 2, where the integer is the length of the time series segment for extreme value detection; The movement step size of the sliding window is set to a positive integer, which is a parameter that controls the degree of overlap between two adjacent detections; For each detection point, the complete time series is composed of the two-dimensional rate of change vector. Starting from the initial moment, vector subsequences with a length equal to the sliding window size are successively extracted according to the moving step size. Within each of the vector subsequences, the Euclidean norm of all two-dimensional rate of change vectors is calculated, and the vector with the largest norm value is identified as the local extremum vector of the current window. The Euclidean norm of the local extreme value vector is compared with the preset extreme value determination threshold, which is the upper limit of normal temperature and humidity fluctuations obtained based on historical data statistics. If the Euclidean norm of the local extremum vector is greater than the extremum determination threshold, the vector is retained and the time-series change label is added to it. Finally, all the retained vectors together constitute the temperature and humidity change rate vector sequence of the radiation protection wall detection points.

10. The method for detecting the airtightness of a radiation shielding wall based on image fusion according to claim 3, characterized in that, The application of the multi-scale wavelet decomposition algorithm to calculate the approximation coefficients and detail coefficients of each sequence at different decomposition levels is as follows: The Daubechies wavelet basis function is selected to perform multi-level decomposition on the real-time temperature value sequence and the real-time humidity value sequence at each detection point. The number of decomposition levels is adaptively determined or preset according to the non-stationary characteristics of the signal. At each decomposition level, the signal is decomposed into the approximation coefficients representing the low-frequency trend of the signal and the detail coefficients representing the high-frequency abrupt changes of the signal. Extract the approximation coefficients from the last level of decomposition; Extract the detail coefficients from all decomposition levels; The extracted approximation coefficients and the detail coefficients of all levels are combined to form a feature vector, generating a multi-scale temperature and humidity dynamic feature vector for the detection point.

Images