Method and System for Detecting and Locating Leakage Points on Exterior Walls Based on Infrared Thermal Imaging

Abstract

The present invention provides a method and system for locating and detecting leakage points on the exterior wall based on infrared thermal imaging, which relates to the field of computer vision technology. First, a multi-period infrared thermal imaging data set of the exterior wall of the target building is collected, covering the surface temperature distribution matrix under different environmental temperatures. Then, dynamic temperature correction processing is performed on it to form a standardized temperature distribution set that eliminates the influence of environmental temperature fluctuations. After that, a set of thermal feature parameters including features such as temperature gradient distribution is extracted from it. Then, a pre-trained thermal anomaly detection model is called to perform leakage risk probability mapping on the set of thermal feature parameters to generate a leakage point probability distribution map. Finally, based on this leakage point probability distribution map and a preset dynamic determination threshold, region segmentation processing is performed to output a set of leakage point location coordinates and their corresponding leakage intensity level identifiers, realizing the precise location and intensity level identification of leakage points on the exterior wall.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology. Specifically, it relates to a method and system for locating and detecting exterior wall leakage points based on infrared thermal imaging. Background Art

[0002] In the field of exterior wall leakage detection of buildings, the existing technology mainly relies on simple single-point infrared thermal imaging detection technology. Although it can use temperature differences to discover leakage points to a certain extent, since it only collects infrared thermal imaging data at a single moment, it cannot fully consider the influence of environmental temperature fluctuations on the surface temperature measurement value. The change of environmental temperature will cause dynamic changes in the surface temperature distribution of the exterior wall, making the single-point detection data difficult to accurately reflect the true thermal anomaly characteristics of the leakage points, thus reducing the accuracy and reliability of leakage detection.

[0003] In addition, most of the existing infrared thermal imaging detection methods only focus on simple comparison of temperature values and lack in-depth excavation and analysis of thermal characteristic parameters. For example, multi-dimensional thermal characteristic parameters such as temperature gradient distribution characteristics, abnormal temperature difference region morphology characteristics, and heat conduction directionality characteristics are not comprehensively considered, resulting in the inability to comprehensively depict the thermal anomaly pattern caused by leakage points, and thus it is difficult to accurately evaluate the leakage risk.

[0004] Meanwhile, in terms of leakage point location and leakage intensity assessment, the existing technology lacks scientific and effective methods and means. Usually, only the location of the leakage point can be roughly determined, the set of leakage point location coordinates cannot be accurately output, and it is even more difficult to quantitatively classify the leakage intensity, which cannot meet the requirements of accurately locating and quantitatively evaluating exterior wall leakage points in actual projects. Summary of the Invention

[0005] In view of the problems mentioned above, in combination with the first aspect of the present invention, embodiments of the present invention provide a method for locating and detecting exterior wall leakage points based on infrared thermal imaging. The method includes:

[0006] Collect a multi-period infrared thermal imaging data set of the surface of the target building exterior wall, where the multi-period infrared thermal imaging data set includes surface temperature distribution matrices under different environmental temperature conditions;

[0007] Perform dynamic temperature correction processing on the multi-period infrared thermal imaging data set to generate a standardized temperature distribution set, and the standardized temperature distribution set eliminates the influence of environmental temperature fluctuations on the surface temperature measurement value;

[0008] Extract a set of thermal characteristic parameters from the standardized temperature distribution set, where the set of thermal characteristic parameters includes temperature gradient distribution characteristics, abnormal temperature difference region morphology characteristics, and heat conduction directionality characteristics;

[0009] Invoke the pre-trained thermal anomaly detection model to perform leakage risk probability mapping processing on the set of thermal feature parameters, and generate a leakage point probability distribution map of the surface of the target building exterior wall;

[0010] Based on the leakage point probability distribution map and a preset dynamic determination threshold, perform region segmentation processing, and output a set of leakage point positioning coordinates and corresponding leakage intensity level identifiers.

[0011] In another aspect, an embodiment of the present invention further provides an infrared thermal imaging-based exterior wall leakage point positioning and detection system, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0012] Based on the above aspects, the embodiment of the present invention realizes the precise positioning and quantitative evaluation of exterior wall leakage points. By collecting a multi-period infrared thermal imaging data set under different ambient temperature conditions and combining a dynamic temperature correction algorithm to generate a standardized temperature distribution set, the interference of ambient temperature fluctuations on thermal imaging data is eliminated, and the accuracy and stability of thermal feature parameter extraction are significantly improved. On this basis, the set of thermal feature parameters composed of the extracted temperature gradient distribution feature, abnormal temperature difference region morphology feature, and heat conduction directionality feature can comprehensively depict the thermal anomaly pattern caused by leakage points, providing a high-dimensional feature space for leakage risk probability mapping. By invoking the pre-trained thermal anomaly detection model for leakage risk probability mapping processing, not only the high-precision generation of the leakage point probability distribution map is realized, but also by introducing a dynamic determination threshold for region segmentation, the adaptability defect of the traditional static threshold method under different environmental conditions is effectively overcome, so that the output of the set of leakage point positioning coordinates has both spatial resolution and confidence. The finally output leakage intensity level identifier provides a basis for leakage repair priority ranking by quantifying the thermal anomaly feature intensity of leakage points, and significantly improves the intelligent level of exterior wall leakage detection. Description of the Drawings

[0013] Figure 1 is a schematic execution flowchart of an infrared thermal imaging-based exterior wall leakage point positioning and detection method provided by an embodiment of the present invention.

[0014] Figure 2 is a schematic diagram of exemplary hardware and software components of an infrared thermal imaging-based exterior wall leakage point positioning and detection system provided by an embodiment of the present invention. Detailed Embodiments

[0015] The present invention will be specifically described below in conjunction with the accompanying drawings of the specification. Figure 1It is a schematic flowchart of a method for locating and detecting leakage points on the exterior wall based on infrared thermal imaging provided by an embodiment of the present invention. The method for locating and detecting leakage points on the exterior wall based on infrared thermal imaging will be introduced in detail below.

[0016] Step S110: Collect a multi-period infrared thermal imaging data set of the surface of the exterior wall of the target building. The multi-period infrared thermal imaging data set includes surface temperature distribution matrices under different environmental temperature conditions.

[0017] In this embodiment, when collecting data, an infrared thermal imager can be selected. In order to comprehensively obtain the thermal imaging information of the exterior wall of the target building under different environmental temperatures, the collection work can be carried out at multiple different time periods. For example, different dates within a week can be selected, and the collection can be carried out at three time points in the morning, noon, and evening every day. Because the environmental temperature is relatively low in the morning, the temperature distribution of the exterior wall of the building is relatively uniform after cooling overnight; at noon, the sun shines directly, the surface temperature of the exterior wall rises, and the temperature differences of the walls facing different directions are obvious; in the evening, as the temperature drops, the heat dissipation situation of the exterior wall will also be different.

[0018] Suppose the exterior wall of the target building is a rectangular structure with a length of 60 meters and a height of 30 meters, and the resolution of the infrared thermal imager used is 1280×960 pixels. This means that each surface temperature distribution matrix obtained by each collection is a two-dimensional matrix with 1280 rows and 960 columns. Each element in this two-dimensional matrix represents the temperature value of the corresponding pixel point on the exterior wall surface, and the unit is degree Celsius. After multiple collections, the above surface temperature distribution matrices at different time periods are combined together to form a multi-period infrared thermal imaging data set. Each two-dimensional matrix in this multi-period infrared thermal imaging data set reflects the temperature distribution of the exterior wall surface under various environmental temperature conditions.

[0019] Step S120: Perform dynamic temperature correction processing on the multi-period infrared thermal imaging data set to generate a standardized temperature distribution set, and the standardized temperature distribution set eliminates the influence of environmental temperature fluctuations on the surface temperature measurement values.

[0020] In this embodiment, since environmental factors such as temperature, humidity, and wind speed will interfere with the measured values of the surface temperature of the exterior wall of the building, in order to obtain data that accurately reflects the true temperature of the exterior wall, it is necessary to perform dynamic temperature correction processing on the multi-period infrared thermal imaging data set. The specific steps are as follows:

[0021] Step S121: Obtain a historical meteorological data set of the area where the exterior wall of the target building is located. The historical meteorological data set includes real-time environmental temperature, real-time humidity, and real-time wind speed parameters corresponding to the collection of the multi-period infrared thermal imaging data.

[0022] To obtain accurate historical meteorological data, communication can be established with meteorological monitoring sources to acquire detailed and precise meteorological information of the area where the outer wall of the target building is located. While collecting infrared thermal imaging data at multiple time intervals, the real-time ambient temperature, real-time humidity, and real-time wind speed parameters corresponding to each collection moment are synchronously recorded.

[0023] For example, when collecting infrared thermal imaging data at 8:00 am on Monday, the recorded real-time ambient temperature is 12 degrees Celsius, the real-time humidity is 75%, and the real-time wind speed is 1.2 meters per second; when collecting data at 12:00 noon on Tuesday, the real-time ambient temperature is 26 degrees Celsius, the real-time humidity is 45%, and the real-time wind speed is 2.8 meters per second; when collecting data at 8:00 pm on Wednesday, the real-time ambient temperature is 16 degrees Celsius, the real-time humidity is 68%, and the real-time wind speed is 0.8 meters per second. Thus, the meteorological parameters corresponding to these different collection moments can be organized into a historical meteorological data set.

[0024] Step S122: Construct a temperature compensation function, which includes a linear compensation term based on the difference between the real-time ambient temperature and the reference ambient temperature, a temperature compensation term transformed based on the influence of the difference between the real-time humidity and the reference humidity, and a temperature compensation term transformed based on the influence of the difference between the real-time wind speed and the reference wind speed.

[0025] In this embodiment, when constructing the temperature compensation function, reasonable reference environmental parameters need to be determined first. For example, through a large number of experiments and analyses, a representative reference ambient temperature of 20 degrees Celsius, a reference humidity of 60%, and a reference wind speed of 2 meters per second are selected.

[0026] For the linear compensation term based on the difference between the real-time ambient temperature and the reference ambient temperature, the linear compensation coefficient k1 is determined to be 0.9 through a large number of experiments. The role of the linear compensation term is to linearly adjust the original surface temperature measurement value according to the difference between the real-time ambient temperature and the reference ambient temperature. Its calculation method is k1 multiplied by (real-time ambient temperature - reference ambient temperature). Since the units of both the real-time ambient temperature and the reference ambient temperature are degrees Celsius, the dimension of this linear compensation term is also degrees Celsius, which is consistent with the dimension of the original surface temperature measurement value.

[0027] Regarding the influence of the difference between the real-time humidity and the reference humidity on the temperature measurement value, the humidity difference cannot be directly added to the temperature. Instead, through experiments and data analysis, the correlation between the humidity difference and the temperature change needs to be found. For example, through research, it is found that the influence of the humidity difference on the temperature can be approximately represented by a quadratic function. Assume that there is a humidity-temperature conversion coefficient k2 of 0.03. The temperature compensation term converted by the humidity influence can be expressed as k2 multiplied by the square of (real-time humidity conversion parameter - reference humidity conversion parameter) and then multiplied by a unit conversion factor, which is determined through experiments and can make the final dimension of this compensation term be degrees Celsius. For example, when the real-time humidity is 70% and the reference humidity is 60%, first convert the real-time humidity and the reference humidity into positive integers from 0 to 100, that is, the real-time humidity conversion parameter is 70 and the reference humidity conversion parameter is 60. Then the square of (real-time humidity conversion parameter - reference humidity conversion parameter) is 100, multiplied by k2 to get 3, and then multiplied by the unit conversion factor (assumed to be 0.1), resulting in 0.3 degrees Celsius. Thus, the influence of the humidity difference is converted into a temperature compensation amount.

[0028] Regarding the influence of the difference between the real-time wind speed and the reference wind speed on the temperature measurement value, a conversion is also required. For example, through experiments, the wind speed-temperature conversion coefficient k3 is determined to be 0.15. The temperature compensation term converted by the wind speed influence can be expressed as k3 multiplied by (real-time wind speed - reference wind speed) and then multiplied by a unit conversion factor, which is also determined through experiments to ensure that the dimension of this compensation term is degrees Celsius. For example, when the real-time wind speed is 3 m / s and the reference wind speed is 2 m / s, (real-time wind speed - reference wind speed) is 1 m / s, multiplied by k3 to get 0.15, and then multiplied by the unit conversion factor (assumed to be 1), resulting in 0.15 degrees Celsius. The influence of the wind speed difference is converted into a temperature compensation amount.

[0029] Combined, the temperature compensation function can be expressed as: the compensated surface temperature value = the original surface temperature measurement value + k1 multiplied by (real-time ambient temperature - reference ambient temperature) + k2 multiplied by the square of (real-time humidity - reference humidity) multiplied by the unit conversion factor + k3 multiplied by (real-time wind speed - reference wind speed) multiplied by the unit conversion factor. Thus, it ensures that the dimensions of all terms in the temperature compensation function are unified as degrees Celsius, and can reasonably correct the original surface temperature measurement value.

[0030] Step S123: Input each surface temperature measurement value in the multi-period infrared thermal imaging data set into the temperature compensation function to obtain the compensated surface temperature value.

[0031] In this embodiment, for each surface temperature measurement value in the multi-period infrared thermal imaging data set, it is substituted into the temperature compensation function for calculation. For example, in the infrared thermal imaging data collected at 9:00 am on Thursday, the original surface temperature measurement value of a pixel is 14 degrees Celsius, the corresponding real-time ambient temperature is 13 degrees Celsius, the real-time humidity is 72%, and the real-time wind speed is 1.4 meters per second. Substitute these values into the temperature compensation function for calculation:

[0032] Linear compensation term = 0.9 × (13 - 20) = -6.3 degrees Celsius;

[0033] Temperature compensation term converted by humidity effect = 0.03 × (72 - 60)² × 0.1 = 0.432 degrees Celsius;

[0034] Temperature compensation term converted by wind speed effect = 0.15 × (1.4 - 2) × 1 = -0.09 degrees Celsius;

[0035] Compensated surface temperature value = 14 + (-6.3) + 0.432 + (-0.09) = 8.042 degrees Celsius.

[0036] Calculate the surface temperature measurement values of all pixel points in the multi-period infrared thermal imaging data set to obtain a set of compensated surface temperature values. Each value in this set of surface temperature values is the corrected surface temperature, effectively eliminating the interference of some environmental factors.

[0037] Step S124: Perform spatial interpolation processing on all the compensated surface temperature values to generate a standardized temperature distribution set consistent with the resolution of the multi-period infrared thermal imaging data set; wherein, the temperature value of each pixel point in the standardized temperature distribution set represents the true temperature of the exterior wall surface after eliminating environmental interference.

[0038] In this embodiment, after temperature compensation, there may be some pixel points with missing or inaccurate temperature values. In order to obtain complete and accurate temperature distribution information, it is necessary to perform spatial interpolation processing on all the compensated surface temperature values.

[0039] In this embodiment, the bicubic spline interpolation method is adopted. Bicubic spline interpolation is an interpolation method based on cubic polynomials, which can more accurately estimate the temperature values of unknown pixels according to the temperature values of known pixels. Specifically, for a pixel whose temperature value needs to be estimated, the temperature values of 16 known pixels within a certain range around it are selected as references. By establishing a cubic polynomial function, the values of this cubic polynomial function at these 16 known pixels are made equal to the known temperature values, and the first and second derivatives of the function are ensured to be continuous at these points. Then, the coordinates of the pixel to be estimated are substituted into this cubic polynomial function to calculate the estimated temperature value of this pixel.

[0040] For example, for an unknown pixel located at the coordinates (x, y), find the coordinates and corresponding temperature values of 16 known pixels around it, and construct a cubic polynomial function. Assume the temperature values of these 16 known pixels are T1, T2,..., T16 respectively. By calculating and solving the system of equations, determine the coefficients of the cubic polynomial function. Finally, substitute (x, y) into this function to obtain the temperature value of this unknown pixel.

[0041] After performing such bicubic spline interpolation processing on all the compensated surface temperature values, a standardized temperature distribution set consistent with the resolution of the multi - period infrared thermal imaging data set is generated. The temperature value of each pixel in this standardized temperature distribution set has undergone correction and interpolation processing, and can accurately represent the true temperature of the outer wall surface after eliminating environmental interference.

[0042] Step S130: Extract a set of thermal characteristic parameters from the standardized temperature distribution set. The set of thermal characteristic parameters includes temperature gradient distribution characteristics, abnormal temperature difference region morphological characteristics, and heat conduction directionality characteristics.

[0043] In order to accurately locate the leakage points on the outer wall, a set of thermal characteristic parameters that can reflect leakage characteristics needs to be extracted from the standardized temperature distribution set. The specific steps are as follows:

[0044] Step S131: Perform spatial gradient calculation on the standardized temperature distribution set to generate a horizontal temperature gradient distribution map and a vertical temperature gradient distribution map, and vectorially superimpose the horizontal temperature gradient distribution map and the vertical temperature gradient distribution map to obtain temperature gradient distribution characteristics.

[0045] Spatial gradient calculation can reflect the rate of change of temperature in space. For the standardized temperature distribution set, perform spatial gradient calculations in the horizontal and vertical directions respectively.

[0046] When calculating the horizontal temperature gradient, for each pixel in the standardized temperature distribution set, calculate the temperature difference between it and the adjacent pixel on the right, and then divide it by the distance between the pixels (assuming the horizontal distance between pixels is 1 unit) to obtain the horizontal temperature gradient value of the pixel. For example, for pixel (i, j), its horizontal temperature gradient value is (the temperature value of pixel (i, j + 1) in the standardized temperature distribution set - the temperature value of pixel (i, j)) divided by 1. By performing such calculations on all pixels in the standardized temperature distribution set, a horizontal temperature gradient distribution map is obtained.

[0047] The method for calculating the vertical temperature gradient is similar. For each pixel, calculate the temperature difference between it and the adjacent pixel below, and then divide it by the distance between the pixels (assuming the vertical distance between pixels is 1 unit) to obtain the vertical temperature gradient value of the pixel. For example, for pixel (i, j), its vertical temperature gradient value is (the temperature value of pixel (i + 1, j) in the standardized temperature distribution set - the temperature value of pixel (i, j)) divided by 1. After calculating all pixels, a vertical temperature gradient distribution map is obtained.

[0048] Perform vector superposition on the horizontal temperature gradient distribution map and the vertical temperature gradient distribution map, that is, use the horizontal temperature gradient value and the vertical temperature gradient value of each pixel as the two components of the vector to obtain the temperature gradient distribution feature. This temperature gradient distribution feature can more comprehensively reflect the spatial variation of temperature and is of great value for discovering areas with abnormal temperature changes on the exterior wall surface.

[0049] Step S132: Detect the connected regions in the standardized temperature distribution set where the continuous temperature difference exceeds the preset threshold, extract the geometric center coordinates, regional area, boundary curvature, and the distribution of internal temperature extreme points of each connected region, and generate the morphological features of the abnormal temperature difference region.

[0050] In the standardized temperature distribution set, the connected regions where the continuous temperature difference exceeds the preset threshold may be the regions where leakage points are located. The preset threshold is determined through a large number of experiments and actual case analyses. In this embodiment, the preset threshold is 2 degrees Celsius.

[0051] Use the connected region detection algorithm to find all connected regions in the standardized temperature distribution set where the continuous temperature difference exceeds 2 degrees Celsius. For each connected region, perform the following feature extraction:

[0052] Calculation of geometric center coordinates: Count the coordinates of all pixel points within the connected region. Add up the abscissas of all pixel points and divide by the number of pixel points to obtain the abscissa of the geometric center. Add up the ordinates of all pixel points and divide by the number of pixel points to obtain the ordinate of the geometric center. For example, if there are 100 pixel points in a connected region, with abscissas x1, x2, …, x100 and ordinates y1, y2, …, y100 respectively, then the abscissa of the geometric center is (x1 + x2 + … + x100) divided by 100, and the ordinate is (y1 + y2 + … + y100) divided by 100.

[0053] Calculation of region area: Determine the region area by counting the number of pixel points within the connected region. Assume that the actual area represented by each pixel point is 0.01 square meters. If there are 200 pixel points in a connected region, then the area of this region is 200 multiplied by 0.01 = 2 square meters.

[0054] Calculation of boundary curvature: For the boundary pixel points of the connected region, calculate the change rate of the angle between adjacent boundary pixel points. The boundary curvature can be obtained by calculating the vector angle between adjacent boundary pixel points and then statistically analyzing the changes in these angles.

[0055] Determination of the distribution of internal temperature extreme points: Find the maximum and minimum temperatures and their corresponding pixel point coordinates within the connected region. For example, within a connected region, after traversing the temperature values of all pixel points, it is found that the maximum temperature is 25 degrees Celsius, and the corresponding pixel point coordinates are (100, 200); the minimum temperature is 20 degrees Celsius, and the corresponding pixel point coordinates are (150, 250).

[0056] Combine these extracted features to generate the morphological features of the abnormal temperature difference region. These morphological features of the abnormal temperature difference region can reflect the morphological features of the abnormal temperature difference region in the standardized temperature distribution set, providing a basis for judging the location and scale of the leakage point.

[0057] Step S133: Analyze the temperature change directions of each pixel point in the standardized temperature distribution set with its adjacent pixel points, and statistically calculate the heat conduction rates in each direction to generate heat conduction directionality features.

[0058] For each pixel point in the standardized temperature distribution set, analyze its temperature change direction with adjacent pixel points. Adjacent pixel points include pixel points in the up, down, left, right, and four diagonal directions.

[0059] Taking a pixel point (i, j) as an example, calculate the temperature difference between it and its adjacent pixel points respectively. If the temperature difference with the adjacent pixel point on the right (i, j + 1) is positive, it indicates that the temperature conducts from the pixel point (i, j) to the right; if the temperature difference is negative, it indicates that the temperature conducts from the right to the pixel point (i, j).

[0060] Statistically analyze the temperature conduction rates in each direction. The temperature conduction rate can be calculated by dividing the temperature difference by the distance between pixel points. Assuming the distance between pixel points is 1 unit, for a pixel point with a temperature difference of 1 degree Celsius from its adjacent pixel point on the right, the temperature conduction rate in this direction is 1 degree Celsius per unit distance.

[0061] Perform such analysis and statistics on all pixel points in the standardized temperature distribution set to obtain the temperature conduction rate distribution in each direction. Then, based on these temperature conduction rates, calculate the temperature conduction probability in each direction. For example, in a certain area, the total temperature conduction rate in the upward direction is statistically obtained as 10 degrees Celsius per unit distance, the total temperature conduction rate in the downward direction is 8 degrees Celsius per unit distance, the total temperature conduction rate in the leftward direction is 6 degrees Celsius per unit distance, and the total temperature conduction rate in the rightward direction is 12 degrees Celsius per unit distance. Then the temperature conduction probability in the upward direction is 10 divided by (10 + 8 + 6 + 12) ≈ 0.278, the temperature conduction probability in the downward direction is 8 divided by (10 + 8 + 6 + 12) ≈ 0.222, the temperature conduction probability in the leftward direction is 6 divided by (10 + 8 + 6 + 12) ≈ 0.167, and the temperature conduction probability in the rightward direction is 12 divided by (10 + 8 + 6 + 12) ≈ 0.333.

[0062] Arrange the temperature conduction probabilities in each direction in a certain order to generate the heat conduction directionality feature. This heat conduction directionality feature can reflect the directionality law of heat conduction in the standardized temperature distribution set and plays an important role in judging the heat conduction anomaly caused by leakage points.

[0063] Step S134: Align and fuse the temperature gradient distribution feature, the abnormal temperature difference region morphology feature, and the heat conduction directionality feature according to spatial coordinates to obtain the heat feature parameter set.

[0064] After obtaining the temperature gradient distribution feature, the abnormal temperature difference region morphology feature, and the heat conduction directionality feature, they need to be fused to form a complete heat feature parameter set.

[0065] First, align these three features according to spatial coordinates. That is to say, ensure that the information at the same spatial position in each feature corresponds. For example, for the temperature gradient vector information of the pixel point (i, j) in the temperature gradient distribution feature, it should correspond to the relevant morphological features of the same coordinates (i, j) in the abnormal temperature difference region morphological feature and the heat conduction probability information of the same coordinates (i, j) in the heat conduction directionality feature.

[0066] Then, fuse these three features by means of weighted splicing. Different weights are assigned to the temperature gradient distribution feature, the abnormal temperature difference region morphological feature, and the heat conduction directionality feature respectively. Suppose the weight of the temperature gradient distribution feature is 0.4, the weight of the abnormal temperature difference region morphological feature is 0.3, and the weight of the heat conduction directionality feature is 0.3. For the feature information at each spatial position, splice the information of its corresponding three features according to the weights. For example, for the pixel point (i, j), multiply its temperature gradient vector information by 0.4, the abnormal temperature difference region morphological feature by 0.3, and the heat conduction directionality probability information by 0.3, and then splice these three weighted information together to form the feature information of this pixel point in the thermal feature parameter set. Such processing is carried out for all pixel points in the standardized temperature distribution set, and finally a complete thermal feature parameter set is obtained. This thermal feature parameter set integrates information such as temperature gradient, abnormal temperature difference region morphology, and heat conduction directionality, and can more comprehensively and accurately reflect the thermal characteristics of the exterior wall of the target building.

[0067] Step S140: Call the pre-trained thermal anomaly detection model to perform leakage risk probability mapping processing on the thermal feature parameter set, and generate a leakage point probability distribution map of the surface of the exterior wall of the target building.

[0068] The pre-trained thermal anomaly detection model can convert the information in the thermal feature parameter set into leakage risk probability, so as to generate a leakage point probability distribution map. The specific steps are as follows:

[0069] Step S141: Input the temperature gradient distribution feature into the first convolutional branch of the pre-trained thermal anomaly detection model to extract high-order abstract features of the spatial distribution of the temperature gradient.

[0070] The first convolutional branch of the pre-trained thermal anomaly detection model is specifically used to process the temperature gradient distribution feature. The convolutional branch consists of multiple convolutional layers, pooling layers, and activation function layers. When the temperature gradient distribution feature is input into the first convolutional branch, the convolutional layer will perform a convolution operation on the input feature. The convolution operation can be understood as sliding multiple different convolutional kernels on the temperature gradient distribution feature to extract local features. Each convolutional kernel has different weight parameters, and by multiplying with the input feature element by element and summing, the convolution result is obtained.

[0071] For example, assume that a 3×3 convolutional kernel performs a convolution operation with a 3×3 region in the temperature gradient distribution feature. The weight parameters of the convolutional kernel are w1, w2, …, w9 respectively, and the temperature gradient vector values of this 3×3 region are t1, t2, …, t9 respectively. Then the convolution result is w1×t1 + w2×t2 + … + w9×t9. Through the convolution operations of multiple convolutional kernels, different types of local features can be extracted.

[0072] The role of the pooling layer is to downsample the convolution result, reduce the data dimension, and at the same time retain important feature information. Common pooling methods include max pooling and average pooling. Taking max pooling as an example, within a 2×2 region, the maximum value among them is taken as the pooling result of this region. This can reduce the amount of data and improve the calculation efficiency.

[0073] The activation function layer is used to introduce non-linear factors and enhance the expression ability of the model. Commonly used activation functions include the ReLU function, whose role is to set the values less than 0 to 0 and keep the values greater than 0 unchanged. Through the alternating actions of multiple convolutional layers, pooling layers, and activation function layers, the first convolution branch can extract high-order abstract features of the spatial distribution of the temperature gradient. These high-order abstract features can more deeply reflect the distribution law and characteristics of the temperature gradient in space.

[0074] Step S142: Input the morphological features of the abnormal temperature difference region into the second convolution branch of the pre-trained thermal anomaly detection model to extract multi-scale geometric features of the region morphology.

[0075] The second convolution branch of the pre-trained thermal anomaly detection model is used to process the morphological features of the abnormal temperature difference region. Similar to the first convolution branch, the second convolution branch also includes multiple convolutional layers, pooling layers, and activation function layers.

[0076] When the morphological features of the abnormal temperature difference region are input into the second convolution branch, the convolutional layer will perform a convolution operation on the input features. Since the morphological features of the abnormal temperature difference region contain various information such as geometric center coordinates, region area, boundary curvature, and the distribution of internal temperature extreme points, the convolutional kernel will extract features for these different types of information. For example, for the region area information, the convolutional kernel can extract features of regions of different sizes; for the boundary curvature information, the convolutional kernel can extract features such as the degree of bending of the boundary.

[0077] The pooling layer will downsample the convolution result to reduce the data dimension. The activation function layer introduces non-linear factors to enhance the expression ability of the model. Through the actions of multiple convolutional layers, pooling layers, and activation function layers, the second convolution branch can extract multi-scale geometric features of the region morphology. Multi-scale geometric features mean that the model can capture the morphological features of the abnormal temperature difference region at different scales, such as the overall shape of the region at a large scale and the local detailed features at a small scale.

[0078] Step S143: Input the heat conduction directionality feature into the third convolutional branch of the pre-trained heat anomaly detection model to extract the temporal evolution feature of the heat conduction path.

[0079] The third convolutional branch of the pre-trained heat anomaly detection model is specifically designed to process the heat conduction directionality feature. This convolutional branch is also composed of a convolutional layer, a pooling layer, and an activation function layer.

[0080] When the heat conduction directionality feature is input into the third convolutional branch, the convolutional layer will perform a convolution operation on the input feature. The heat conduction directionality feature reflects the probability of heat conduction in different directions, and the convolutional kernel will extract features based on this probability information. For example, the convolutional kernel can extract features in the directions with higher heat conduction probability and the correlation features between different directions.

[0081] The pooling layer downsamples the convolution result to reduce the data dimension. The activation function layer introduces non-linearity to enhance the model's expressive power. Since heat conduction is a time-varying process, the third convolutional branch will consider this temporal evolution characteristic during the processing. Through the action of multiple convolutional layers, pooling layers, and activation function layers, the third convolutional branch can extract the temporal evolution feature of the heat conduction path. These temporal evolution features can reflect the variation law of heat conduction over time and play an important role in judging the heat conduction anomaly caused by the leakage point.

[0082] Step S144: Perform cross-channel attention fusion on the high-order abstract feature, the multi-scale geometric feature, and the temporal evolution feature to generate a fused feature tensor.

[0083] After extracting the high-order abstract feature, the multi-scale geometric feature, and the temporal evolution feature from the three convolutional branches respectively, it is necessary to fuse these features to generate more representative features. In this embodiment, a cross-channel attention fusion method is adopted.

[0084] The core idea of the cross-channel attention fusion method is to assign different weights to the features of different channels to highlight the important feature information. First, the high-order abstract feature, the multi-scale geometric feature, and the temporal evolution feature are concatenated in the channel dimension to form a new feature tensor. Then, through a global average pooling operation, the feature tensor is compressed in the spatial dimension to obtain the global feature information of each channel.

[0085] Next, the global feature information is input into a fully connected layer, and after passing through a Sigmoid activation function, the attention weight of each channel is obtained. The Sigmoid activation function maps the input value to between 0 and 1, and these values are the attention weights of each channel. The larger the attention weight, the more important the feature of that channel.

[0086] Finally, the original feature tensor is multiplied channel - by - channel with the attention weights to obtain the weighted feature tensor. In this way, the adaptive weighting of features in different channels is realized, highlighting the important feature information and generating the fused feature tensor. The fused feature tensor synthesizes high - order feature information in multiple aspects such as temperature gradient, the morphology of abnormal temperature difference regions, and the directionality of heat conduction, and can more accurately reflect the leakage risk situation of the outer wall of the target building.

[0087] Step S145: Map the fused feature tensor to the resolution of the original infrared thermal imaging data through a transposed convolution layer, and output the leakage risk probability value of each pixel point to generate the leakage point probability distribution map.

[0088] The role of the transposed convolution layer is to restore the resolution of the fused feature tensor to the resolution of the original infrared thermal imaging data. The transposed convolution operation can be regarded as the inverse process of the convolution operation. It expands the size of the feature map by performing upsampling and convolution operations on the input features.

[0089] When the fused feature tensor is input into the transposed convolution layer, the transposed convolution layer performs upsampling and convolution operations according to the preset transposed convolution kernel and stride. For example, assuming that the size of the transposed convolution kernel is 4×4 and the stride is 2, the transposed convolution layer will fill a certain number of 0s around each element of the fused feature tensor and then perform a convolution operation with the transposed convolution kernel to obtain a feature map with a larger size. Through the action of multiple transposed convolution layers, the resolution of the fused feature tensor is gradually restored to the resolution of the original infrared thermal imaging data.

[0090] During the transposed convolution process, each pixel point will obtain a corresponding feature value. These feature values are processed through a Softmax activation function to convert them into probability values between 0 and 1. These probability values represent the leakage risk probability of each pixel point. Arranging the leakage risk probabilities of each pixel point according to their positions in the original infrared thermal imaging data generates the leakage point probability distribution map of the outer wall surface of the target building. The leakage point probability distribution map intuitively shows the leakage risk situation of each position on the outer wall surface of the target building.

[0091] Step S150: Based on the leakage point probability distribution map and a preset dynamic determination threshold, perform region segmentation processing, and output a set of leakage point localization coordinates and the corresponding leakage intensity level identifiers.

[0092] In order to accurately locate the leakage points from the leakage point probability distribution map and determine their leakage intensity levels, region segmentation processing is required. The specific steps are as follows:

[0093] Step S151: Retrieve the basic determination threshold from the threshold configuration library according to the material type and environmental exposure level of the outer wall of the target building.

[0094] The threshold configuration library stores the basic judgment thresholds corresponding to different types of building exterior wall materials and environmental exposure levels. There are various types of building exterior wall materials, such as brick walls, concrete walls, glass curtain walls, etc. Different material types have different sensitivities to leakage, so the corresponding basic judgment thresholds are also different. The environmental exposure level takes into account the environmental conditions where the building exterior wall is located, such as whether it is close to a water source and whether it is frequently washed by rain.

[0095] For example, for a brick wall, under the general environmental exposure level, the basic judgment threshold may be 0.3; for a concrete wall, under the high environmental exposure level, the basic judgment threshold may be 0.4. According to the specific material type and environmental exposure level of the target building exterior wall, the corresponding basic judgment threshold is retrieved from the threshold configuration library. Suppose the target building exterior wall is a concrete wall and the environmental exposure level is medium, and the basic judgment threshold retrieved from the threshold configuration library is 0.35.

[0096] Step S152: Analyze the global statistical characteristics of the leakage point probability distribution map and calculate the mean and variance of the probability values.

[0097] Statistical analysis is performed on the leakage risk probability values of all pixel points in the leakage point probability distribution map. First, calculate the mean of the probability values. The calculation method of the mean is to add up the leakage risk probability values of all pixel points and then divide by the total number of pixel points. For example, if there are 10,000 pixel points in the leakage point probability distribution map, and their leakage risk probability values are p1, p2,..., p10000, then the mean of the probability values is (p1 + p2 +... + p10000) divided by 10000.

[0098] Next, calculate the variance of the probability values. The variance reflects the degree of dispersion of the probability values. The calculation method of the variance is to first calculate the square of the difference between the leakage risk probability value of each pixel point and the mean, then add up these squared values, and finally divide by the total number of pixel points. For example, for pixel point i, its leakage risk probability value is pi and the mean is μ, then the square of the difference between this pixel point and the mean is the square of (pi - μ). Add up the squares of (pi - μ) for all pixel points and divide by 10000 to obtain the variance of the probability values.

[0099] Suppose that after calculation, the mean of the probability values in the leakage point probability distribution map is 0.2 and the variance is 0.05.

[0100] Step S153: Dynamically offset and adjust the basic judgment threshold based on the mean and variance to generate a dynamic judgment threshold adapted to the current infrared thermal imaging data.

[0101] The purpose of dynamic offset adjustment is to adjust the basic determination threshold according to the actual situation of the leakage point probability distribution map, so as to improve the accuracy of leakage point positioning. The method of dynamic offset adjustment can be linearly adjusted according to the magnitudes of the mean and variance.

[0102] Assume that the dynamic offset adjustment coefficient k is 0.2, and the calculation formula for the dynamic determination threshold is: Dynamic determination threshold = Basic determination threshold + k × (Mean - Basic determination threshold) + 0.1 × Variance. Substitute the basic determination threshold of 0.35, the mean of 0.2, and the variance of 0.05 into the formula, and we get the dynamic determination threshold = 0.35 + 0.2 × (0.2 - 0.35) + 0.1 × 0.05 = 0.35 - 0.03 + 0.005 = 0.325. In this way, a dynamic determination threshold suitable for the current infrared thermal imaging data is generated.

[0103] Step S154: Perform threshold segmentation on the leakage point probability distribution map, and extract the connected regions whose probability values exceed the dynamic determination threshold.

[0104] Use the dynamic determination threshold to perform threshold segmentation on the leakage point probability distribution map. For each pixel point in the leakage point probability distribution map, if its leakage risk probability value exceeds the dynamic determination threshold of 0.325, then mark this pixel point as a possible leakage point; if its leakage risk probability value is less than or equal to the dynamic determination threshold, then mark this pixel point as a non-leakage point.

[0105] Then, use the connected region detection algorithm to find all the connected regions composed of pixel points marked as possible leakage points. A connected region refers to a region formed by connecting adjacent pixel points of possible leakage points. For example, in the leakage point probability distribution map, there are some adjacent pixel points whose leakage risk probability values all exceed the dynamic determination threshold, and these pixel points form a connected region. In this way, all the connected regions whose probability values exceed the dynamic determination threshold are extracted.

[0106] Step S155: Perform morphological optimization processing on each of the connected regions, remove the isolated pixel points caused by noise, and output the set of leakage point positioning coordinates and the corresponding leakage intensity level identifiers.

[0107] The purpose of morphological optimization processing is to remove the isolated pixel points in the connected region caused by factors such as noise, so that the connected region can more accurately reflect the actual position of the leakage point. Commonly used morphological optimization methods include erosion and dilation operations.

[0108] The erosion operation is to contract the boundary of the connected region inward to remove some small and isolated pixel points. The specific method is to slide a structuring element over the connected region. If the overlapping part between the structuring element and the connected region does not exactly match the shape of the structuring element, then the pixel point is removed from the connected region. For example, using a 3×3 structuring element, if at a certain position, the overlapping part between the structuring element and the connected region has only one pixel point, then that pixel point will be removed.

[0109] The dilation operation is to expand the boundary of the connected region outward to fill some internal small holes. The specific method is to slide a structuring element over the connected region. If there is an overlapping part between the structuring element and the connected region, then all pixel points covered by the structuring element are added to the connected region.

[0110] Perform the erosion and dilation operations on each connected region in turn to remove the isolated pixel points caused by noise and obtain the optimized connected region. Then, extract the geometric center coordinates of the optimized connected region, and these coordinates form the set of leakage point positioning coordinates.

[0111] Next, determine the leakage intensity level identifier corresponding to each leakage point positioning coordinate. The specific process is as follows:

[0112] Step S1551: Calculate the mean value of the probability values of the region corresponding to each of the leakage point positioning coordinates as the leakage intensity reference value.

[0113] For the connected region corresponding to each leakage point positioning coordinate, calculate the mean value of the leakage risk probability values of all pixel points in the region. For example, if there are 50 pixel points in a connected region, and their leakage risk probability values are p1, p2,..., p50 respectively, then the mean value of the probability values of this region is (p1 + p2 +... + p50) divided by 50, and this mean value of the probability values is the leakage intensity reference value of the region corresponding to the leakage point positioning coordinate.

[0114] Step S1552: Statistically calculate the maximum temperature gradient value, the abnormal region area, and the heat conduction direction consistency index of the region in the set of thermal characteristic parameters.

[0115] In the set of thermal characteristic parameters, find the relevant information of the region corresponding to each leakage point positioning coordinate. The maximum temperature gradient value refers to the maximum value of the temperature gradient in the region, which reflects the severity of the temperature change in the region. The abnormal region area refers to the area of the connected region, which reflects the size of the leakage point.

[0116] The heat conduction direction consistency index is an index for measuring the consistency of the heat conduction direction in this area. The calculation method is to count the proportion of pixel points with the same heat conduction direction in this area. For example, in an area, there are 80 pixel points, and among them, 60 pixel points have the same heat conduction direction, then the heat conduction direction consistency index of this area is 60 divided by 80 = 0.75.

[0117] Step S1553: Input the leakage intensity reference value, the maximum temperature gradient value, the abnormal area, and the heat conduction direction consistency index into a pre-trained intensity classification model, and output a discrete leakage intensity level.

[0118] The pre-trained intensity classification model uses the gradient boosting decision tree algorithm. In the training process of this model, the intensity labels of historical leakage points and the corresponding multi-dimensional feature vectors are used as training data.

[0119] When inputting the leakage intensity reference value, the maximum temperature gradient value, the abnormal area, and the heat conduction direction consistency index into the pre-trained intensity classification model, the intensity classification model constructs and predicts a decision tree based on these input features. The decision tree makes branch judgments according to the values of different features, and finally outputs a discrete leakage intensity level. For example, the leakage intensity level can be divided into three levels: mild leakage, moderate leakage, and severe leakage. The intensity classification model determines which level the leakage point belongs to according to the input features and outputs the corresponding level identifier.

[0120] So far, all steps of the method for locating and detecting external wall leakage points based on infrared thermal imaging have been completed, and the set of leakage point location coordinates and the corresponding leakage intensity level identifiers have been output.

[0121] In a possible implementation manner, the method further includes:

[0122] Step S210: The training process of the pre-trained thermal anomaly detection model includes the following steps:

[0123] Step S211: Obtain a sample leakage point annotation data set, and the sample leakage point annotation data set contains infrared thermal imaging samples of multiple building facades and their corresponding true leakage point coordinates and leakage intensity labels.

[0124] To train the thermal anomaly detection model, a large amount of sample data needs to be collected. By collecting infrared thermal imaging data of multiple different building facades, infrared thermal imaging samples of multiple building facades are obtained. At the same time, manual detection or other reliable detection methods are used to determine the coordinates and leakage intensity levels of the true leakage points in each infrared thermal imaging sample, and this information is used as labels and marked on the corresponding infrared thermal imaging samples.

[0125] For example, 1000 infrared thermal imaging samples of different building facades are collected. For each sample, through the detection by professionals, the coordinates of the real leakage points therein are determined, such as (x1, y1), (x2, y2), …, and according to the severity of the leakage, each leakage point is labeled with a leakage intensity level, such as mild leakage, moderate leakage or severe leakage. These infrared thermal imaging samples and their corresponding real leakage point coordinates and leakage intensity labels are sorted into a sample leakage point annotation dataset.

[0126] Step S212: Perform the dynamic temperature correction process and the extraction process of the set of thermal characteristic parameters on each of the infrared thermal imaging samples to obtain a set of sample thermal characteristic parameters.

[0127] In this embodiment, each infrared thermal imaging sample in the sample leakage point annotation dataset is processed according to the method described in the previous steps S120 to S134. First, perform the dynamic temperature correction process, obtain the historical meteorological data corresponding to the sample collection, construct a temperature compensation function, input the surface temperature measurement value in the infrared thermal imaging sample into the temperature compensation function to obtain the compensated surface temperature value, and then perform spatial interpolation processing on the compensated surface temperature value to generate a standardized temperature distribution set.

[0128] Next, perform the extraction process of the set of thermal characteristic parameters. Calculate the spatial gradient of the standardized temperature distribution set to generate a horizontal temperature gradient distribution map and a vertical temperature gradient distribution map, perform vector superposition on the two to obtain the temperature gradient distribution characteristics. Detect the connected regions in the standardized temperature distribution set where the continuous temperature difference exceeds a preset threshold (such as 2 degrees Celsius), extract the geometric center coordinates, region area, boundary curvature and internal temperature extreme point distribution of each connected region to generate the morphological characteristics of the abnormal temperature difference region. Analyze the temperature change direction of each pixel point in the standardized temperature distribution set with its adjacent pixel points, and statistically calculate the temperature conduction rate in each direction to generate the thermal conduction directionality characteristics. Finally, align and fuse the temperature gradient distribution characteristics, the morphological characteristics of the abnormal temperature difference region and the thermal conduction directionality characteristics according to the spatial coordinates to obtain a set of sample thermal characteristic parameters.

[0129] For example, for a specific infrared thermal imaging sample, the corresponding historical meteorological data shows that the real-time ambient temperature at the time of collection is 18 degrees Celsius, the real-time humidity is 62%, and the real-time wind speed is 1.6 meters per second. The reference ambient temperature is set at 20 degrees Celsius, the reference humidity is 60%, and the reference wind speed is 2 meters per second. The linear compensation coefficient k1 is 0.9, the humidity-temperature conversion coefficient k2 is 0.03, the wind speed-temperature conversion coefficient k3 is 0.15, the unit conversion factor for humidity influence conversion is 0.1, and the unit conversion factor for wind speed influence conversion is 1.

[0130] For a certain pixel point in the sample, the original surface temperature measurement value is 16 degrees Celsius. Substitute it into the temperature compensation function for calculation:

[0131] Linear compensation term = 0.9×(18 - 20) = -1.8 degrees Celsius;

[0132] Temperature compensation term converted by humidity effect = 0.03×(62 - 60)²×0.1 = 0.012 degrees Celsius;

[0133] Temperature compensation term converted by wind speed effect = 0.15×(1.6 - 2)×1 = -0.06 degrees Celsius;

[0134] Compensated surface temperature value = 16 + (-1.8) + 0.012 + (-0.06) = 14.152 degrees Celsius.

[0135] Perform such calculations and subsequent spatial interpolation processing and thermal feature extraction processing on all pixel points in the sample. Finally, obtain the sample thermal feature parameter set of the sample. Perform the same operation on all infrared thermal imaging samples in the sample leakage point annotation dataset to obtain a complete sample thermal feature parameter set.

[0136] Step S213: Construct an initial thermal anomaly detection model, and the initial thermal anomaly detection model includes three parallel convolutional branches, a cross-channel attention fusion module, and a deconvolution layer.

[0137] The construction of the initial thermal anomaly detection model requires reasonable design of each module to effectively process and analyze the sample thermal feature parameters.

[0138] First are the three parallel convolutional branches. The first convolutional branch is used to process the temperature gradient distribution feature, which consists of multiple convolutional layers, pooling layers, and activation function layers. Parameters such as the convolutional kernel size, number, and stride of the convolutional layer need to be carefully selected. For example, the first convolutional layer can use a 3×3 convolutional kernel, with a number of 16 and a stride of 1. The weight parameters of the convolutional kernel are randomly assigned during model initialization and are adjusted through training later. The pooling layer can use max pooling, with a pooling window size of 2×2 and a stride of 2 to reduce the data dimension. The activation function uses the ReLU function to enhance the nonlinear expression ability of the model.

[0139] The second convolutional branch is used to process the morphological feature of the abnormal temperature difference region. Its structure is similar to that of the first convolutional branch, but the parameters of the convolutional kernel and the number of convolutional layers can be adjusted according to the characteristics of this feature. For example, the first convolutional layer can use a 5×5 convolutional kernel, with a number of 24 and a stride of 1.

[0140] The third convolutional branch is used to process the directional characteristics of heat conduction, and it is also composed of a convolutional layer, a pooling layer, and an activation function layer. Parameters such as the size, number, and stride of its convolutional kernel also need to be designed according to the characteristics of this feature. For example, the first convolutional layer uses a 4×4 convolutional kernel, with 20 in number and a stride of 1.

[0141] The cross-channel attention fusion module is located after the three convolutional branches, and its function is to fuse the high-order abstract features, multi-scale geometric features, and temporal evolution features output by the three convolutional branches. This module first concatenates these three features in the channel dimension, and then compresses the feature tensor in the spatial dimension through global average pooling operation to obtain the global feature information of each channel. Then, the global feature information is input into a fully connected layer, and the attention weights of each channel are obtained through the Sigmoid activation function. Finally, the original feature tensor is multiplied by the attention weights channel by channel to achieve adaptive weighted fusion of the features.

[0142] The deconvolution layer is located after the cross-channel attention fusion module, and its function is to restore the resolution of the fused feature tensor to the resolution of the original infrared thermal imaging data. Parameters such as the size, number, and stride of the deconvolution kernel of the deconvolution layer need to be designed according to the resolution of the original infrared thermal imaging data and the size of the fused feature tensor. For example, the size of the deconvolution kernel can be 4×4, the number is the same as the number of channels of the fused feature tensor, and the stride is 2. Through the operations of multiple deconvolution layers, the size of the feature map is gradually enlarged, and finally an output with the same resolution as the original infrared thermal imaging data is obtained.

[0143] Step S214: Input the sample thermal feature parameter set into the initial thermal anomaly detection model, output the predicted leakage probability distribution map, and calculate the focal loss function between the predicted leakage probability distribution map and the true leakage point coordinates.

[0144] When the sample thermal feature parameter set is input into the initial thermal anomaly detection model, the temperature gradient distribution feature enters the first convolutional branch, and through the convolutional operation of the convolutional layer, local features are extracted. For example, 16 3×3 convolutional kernels of the first convolutional layer slide on the temperature gradient distribution feature, multiply element by element with the input feature and sum, obtaining 16 convolutional result feature maps. These feature maps then undergo downsampling by the pooling layer and non-linear transformation by the activation function layer to obtain a more advanced feature representation.

[0145] The abnormal temperature difference region morphology feature enters the second convolutional branch, and through operations such as convolution, pooling, and activation, multi-scale geometric features of the region morphology are extracted. The heat conduction directional feature enters the third convolutional branch to extract the temporal evolution feature of the heat conduction path.

[0146] The features output by the three convolution branches are fused in the cross-channel attention fusion module to obtain a fused feature tensor. The fused feature tensor then undergoes an upsampling operation through a deconvolution layer to output a predicted leakage probability distribution map. Each pixel point in the predicted leakage probability distribution map has a corresponding leakage risk probability value.

[0147] Next, calculate the focal loss function between the predicted leakage probability distribution map and the true leakage point coordinates. The specific steps are as follows:

[0148] Step S2141: Perform Gaussian kernel diffusion processing on the true leakage point coordinates to generate a continuous leakage probability density map.

[0149] The purpose of Gaussian kernel diffusion processing is to convert the true leakage point coordinates from discrete point information into a continuous probability density map. First, the parameters of the Gaussian kernel need to be determined, such as the standard deviation of the Gaussian kernel. Assume the standard deviation is 5 pixel units.

[0150] For each true leakage point coordinate, centered on this coordinate, calculate the probability values of the surrounding pixel points according to the formula of the Gaussian function. The form of the Gaussian function is: in a two-dimensional plane, for a point centered at (x0, y0), the probability value of its surrounding point (x, y) is proportional to exp(-((x - x0)² + (y - y0)²) / (2 × standard deviation²)). By performing such processing on all true leakage point coordinates and accumulating the probability values around each true leakage point, a continuous leakage probability density map is obtained.

[0151] For example, there is a true leakage point coordinate of (100, 200). For the pixel point with coordinates (101, 201), calculate the square of the distance between it and (100, 200) as (101 - 100)² + (201 - 200)² = 2, and substitute it into the Gaussian function to calculate the probability value of this pixel point. Perform such calculations and accumulations for all true leakage points and their surrounding pixel points, and finally generate a continuous leakage probability density map.

[0152] Step S2142: Calculate the per-pixel cross-entropy loss between the predicted leakage probability distribution map and the continuous leakage probability density map.

[0153] The per-pixel cross-entropy loss is used to measure the difference between the predicted leakage probability distribution map and the continuous leakage probability density map. For each corresponding pixel point in the predicted leakage probability distribution map and the continuous leakage probability density map, calculate its cross-entropy loss.

[0154] Assume that the predicted probability value of the pixel point (i, j) in the predicted leakage probability distribution map is pij, and the true probability value of the corresponding pixel point (i, j) in the continuous leakage probability density map is qij. The calculation formula for the cross-entropy loss is -qij×log(pij) - (1 - qij)×log(1 - pij). Such calculations are performed for all pixel points in the predicted leakage probability distribution map and the continuous leakage probability density map, and then the cross-entropy losses of all pixel points are added together to obtain the total of the per-pixel cross-entropy losses.

[0155] For example, for the pixel point (10, 20), the predicted probability value p10,20 is 0.6, and the true probability value q10,20 is 0.8. The cross-entropy loss of this pixel point is -0.8×log(0.6) - (1 - 0.8)×log(1 - 0.6). Similar calculations and summations are performed for all pixel points.

[0156] Step S2143: Dynamically adjust the loss weight according to the degree of difference between the probability value in the predicted leakage probability distribution map and the true probability value in the continuous leakage probability density map, where higher weights are assigned to difficult-to-classify samples.

[0157] In this embodiment, difficult-to-classify samples refer to samples with a large difference (such as greater than a set difference value) between the predicted probability value and the true probability value. In order to make the model pay more attention to these difficult-to-classify samples, it is necessary to dynamically adjust the loss weight.

[0158] First, calculate the degree of difference between the predicted probability value and the true probability value of each pixel point. For example, it can be represented by the absolute value of the difference between the two. Then, determine the loss weight according to the degree of difference. A function can be designed to achieve this dynamic adjustment. For example, when the degree of difference is less than a certain threshold, the loss weight is 1; when the degree of difference is greater than this threshold, the loss weight increases as the degree of difference increases.

[0159] Assume that the degree of difference is represented by |pij - qij|, and the threshold is 0.2. For the pixel point (i, j), if |pij - qij| < 0.2, the loss weight wij is 1; if |pij - qij| ≥ 0.2, the loss weight wij = 1 + k×(|pij - qij| - 0.2), where k is a constant, such as k = 2.

[0160] Step S2144: Introduce a regularization term to constrain the low-probability response intensity in the non-leakage area of the predicted leakage probability distribution map.

[0161] In this embodiment, the low-probability response intensity in the non-leakage area needs to meet a preset condition, that is, the leakage risk probability value of each pixel point or area in the non-leakage area is lower than a preset low-probability threshold, and the low-probability values show consistency in spatial distribution.

[0162] The introduction of the regularization term is to prevent the model from overfitting and at the same time constrain the low-probability response intensity in the non-leakage area. The L2 regularization term can be adopted, and its calculation method is to multiply the sum of the squares of all trainable parameters in the thermal anomaly detection model by a regularization coefficient λ.

[0163] Suppose the trainable parameters in the model are w1, w2, …, wn, and the L2 regularization term is λ×(w1² + w2² + … + wn²). The regularization coefficient λ needs to be adjusted according to the actual situation. For example, λ can be taken as 0.001.

[0164] Step S2145: Weightedly sum the per-pixel cross-entropy loss, the dynamically adjusted loss weight, and the regularization term to obtain the final focal loss function.

[0165] Multiply the per-pixel cross-entropy loss by the dynamically adjusted loss weight, and then add the regularization term to obtain the final focal loss function. That is, focal loss function = per-pixel cross-entropy loss × dynamically adjusted loss weight + regularization term.

[0166] For example, if the sum of the per-pixel cross-entropy losses is 100, the dynamically adjusted loss weight is 1.2 after calculation and accumulation for each pixel point, and the regularization term is 0.5, then the value of the focal loss function is 100×1.2 + 0.5 = 120.5.

[0167] Step S215: Iteratively optimize the focal loss function through the backpropagation algorithm until the model converges to obtain the pre-trained thermal anomaly detection model.

[0168] The backpropagation algorithm calculates the gradient of the focal loss function with respect to each trainable parameter in the model, and then updates the parameters according to the gradient to reduce the value of the focal loss function.

[0169] First, according to the focal loss function, use the chain rule to calculate the gradient of the focal loss function with respect to the convolutional kernel weights of each convolutional layer, the weights of the fully connected layers, and other trainable parameters in the model. For example, for the convolutional kernel weights of the convolutional layer, calculate the partial derivative of the focal loss function with respect to each element.

[0170] Then, according to the calculated gradient, use an optimization algorithm (such as the stochastic gradient descent algorithm) to update the trainable parameters. The update formula of the stochastic gradient descent algorithm is: new parameter value = old parameter value - learning rate × gradient. The learning rate is a hyperparameter that needs to be adjusted. For example, the learning rate can be initialized as 0.001.

[0171] In each iteration, the set of sample thermal feature parameters is input into the model, the focal loss function is calculated, the gradient is calculated through backpropagation, and then the parameters are updated. As the number of iterations increases, the value of the focal loss function will gradually decrease. When the change in the value of the focal loss function is less than a certain threshold (such as 0.001) in consecutive multiple iterations, the model is considered to have converged. At this time, the obtained model is the pre-trained thermal anomaly detection model.

[0172] Step S310: The online optimization process of the thermal anomaly detection model includes the following steps:

[0173] Step S3101: Receive the feedback information of the user on the set of leakage point location coordinates, and the feedback information includes false alarm location marks and missed alarm location marks.

[0174] In practical applications, there may be false alarms and missed alarms in the set of leakage point location coordinates output by the thermal anomaly detection model. The user can provide feedback on the set of leakage point location coordinates according to the actual inspection results.

[0175] For example, after the user conducts a field inspection on the exterior wall of a building and finds that a certain location marked as a leakage point by the model actually has no leakage, the user can mark this location as a false alarm location; conversely, if the thermal anomaly detection model does not mark a location where there is actually a leakage, the user can mark this location as a missed alarm location, and submit these false alarm location marks and missed alarm location marks as feedback information.

[0176] Step S3102: Extract the set of thermal feature parameters corresponding to the false alarm location marks as negative samples, and the set of thermal feature parameters corresponding to the missed alarm location marks as positive samples.

[0177] For the area corresponding to the false alarm location mark, according to the methods in the previous steps S120 to S134, perform dynamic temperature correction processing and extraction processing of the thermal feature parameter set on the infrared thermal imaging data corresponding to this area. The obtained set of thermal feature parameters is used as negative samples because there is actually no leakage in these areas.

[0178] For example, for the infrared thermal imaging data of the area corresponding to the false alarm location mark, the real-time ambient temperature during data collection is 22 degrees Celsius, the real-time humidity is 55%, and the real-time wind speed is 2.2 meters per second. After dynamic temperature correction processing and thermal feature extraction processing, the temperature gradient distribution feature, abnormal temperature difference area morphology feature, and thermal conduction directionality feature of this area are obtained. After aligning these features according to spatial coordinates and fusing them, the set of thermal feature parameters of the negative samples is obtained.

[0179] For the regions corresponding to the missed detection location markers, dynamic temperature correction processing and extraction processing of the thermal feature parameter set are also performed. The obtained thermal feature parameter set is used as positive samples because there are actually leaks in these regions.

[0180] Step S3103: Add the newly added positive and negative samples to the sample leakage point annotation data set, triggering the derivative training process of the pre-trained thermal anomaly detection model.

[0181] Add the newly added positive and negative samples to the original sample leakage point annotation data set to expand the data set. Then trigger the derivative training process of the pre-trained thermal anomaly detection model to further optimize the performance of the model.

[0182] The derivative training process is similar to the initial training process but is different in some aspects. The specific steps are as follows:

[0183] Step S3103-1: Perform rotation, translation, and scale transformation on the newly added positive and negative samples to generate an enhanced sample set.

[0184] To increase the diversity of the samples, data augmentation operations are performed on the newly added positive and negative samples. The rotation operation can rotate the sample around a certain center point by a certain angle, such as 90 degrees, 180 degrees, or 270 degrees. The translation operation can move the sample a certain distance in the horizontal and vertical directions. The scale transformation operation can enlarge or reduce the sample.

[0185] For example, for the thermal feature parameter set of a positive sample, rotate it 90 degrees around the center point to obtain a new sample; translate it 5 pixel units to the right in the horizontal direction to obtain another new sample; enlarge it by 1.2 times to obtain another new sample. Perform such rotation, translation, and scale transformation operations on all newly added positive and negative samples to generate an enhanced sample set.

[0186] Step S3103-2: Extract the standardized temperature distribution set and thermal feature parameter set of the enhanced sample set and the original samples in the sample leakage point annotation data set to generate a multi-dimensional feature vector set.

[0187] For the enhanced sample set and the original samples in the sample leakage point annotation data set, perform dynamic temperature correction processing according to the previous method to obtain the standardized temperature distribution set. Then extract the thermal feature parameter set from the standardized temperature distribution set, including temperature gradient distribution characteristics, abnormal temperature difference region morphology characteristics, and heat conduction directionality characteristics.

[0188] Further process these sets of thermal feature parameters, for example, perform operations such as splicing or dimensionality reduction on them to generate a set of multi-dimensional feature vectors. Each vector in the set of multi-dimensional feature vectors contains rich feature information of the samples and is used for subsequent model training.

[0189] Step S3103-3: Calculate the Euclidean distance between the multi-dimensional feature vectors of the enhanced sample set and the multi-dimensional feature vectors of the original samples to generate a difference degree distribution matrix.

[0190] The Euclidean distance is a commonly used method to measure the distance between two vectors. For each multi-dimensional feature vector in the enhanced sample set and each multi-dimensional feature vector in the original samples, calculate the Euclidean distance between them.

[0191] Suppose there are m multi-dimensional feature vectors in the enhanced sample set and n multi-dimensional feature vectors in the original samples. For the i-th multi-dimensional feature vector in the enhanced sample set and the j-th multi-dimensional feature vector in the original samples, the calculation method of the Euclidean distance is to add the squares of the differences of the corresponding elements of the two vectors and then take the square root.

[0192] For example, the i-th multi-dimensional feature vector in the enhanced sample set is (a1, a2,..., ak), and the j-th multi-dimensional feature vector in the original samples is (b1, b2,..., bk). The Euclidean distance between them is √((a1 - b1)² + (a2 - b2)² +... + (ak - bk)²). After calculating the Euclidean distances between all the multi-dimensional feature vectors of the enhanced sample set and the multi-dimensional feature vectors of the original samples, a matrix with m rows and n columns is formed, and this matrix is the difference degree distribution matrix. Each element in the matrix represents the difference degree between the corresponding enhanced sample and the original sample.

[0193] Step S3103-4: Determine a filtering threshold according to the cumulative distribution function of the difference degree distribution matrix, and remove the samples in the enhanced sample set whose difference degree exceeds the filtering threshold to generate a filtered enhanced sample set.

[0194] First, calculate the cumulative distribution function of the difference degree distribution matrix. The cumulative distribution function describes the proportion of samples with a difference degree less than or equal to a certain value in the total samples.

[0195] Sort all the elements in the difference degree distribution matrix, and successively accumulate the sample quantities corresponding to each element from small to large to obtain the cumulative quantity. Then divide the cumulative quantity by the total sample quantity to obtain the cumulative distribution function value.

[0196] For example, there are 100 elements in the difference degree distribution matrix. After sorting, the sample quantity corresponding to the first element is 1, the cumulative quantity is 1, and the cumulative distribution function value is 1 / 100 = 0.01; the sample quantity corresponding to the second element is 1, the cumulative quantity is 2, and the cumulative distribution function value is 2 / 100 = 0.02, and so on.

[0197] Determine the filtering threshold according to the cumulative distribution function. A suitable cumulative distribution function value can be selected, such as 0.9, and the corresponding difference degree value is the filtering threshold. For each sample in the enhanced sample set, if its difference degree from the original sample exceeds the filtering threshold, then remove this sample from the enhanced sample set, and finally obtain the filtered enhanced sample set.

[0198] Step S3103-5: Based on the model training iteration times of the filtered enhanced sample set, statistically calculate the gradient update amplitude of each model parameter, and generate a parameter importance weight matrix.

[0199] During the process of using the filtered enhanced sample set for model training, record the gradient update amplitude of each model parameter in each iteration. Model parameters include the convolution kernel weights of the convolutional layer, the weights of the fully connected layer, etc.

[0200] For each model parameter, statistically calculate the sum of its gradient update amplitudes in all training iteration times. For example, in 100 iterations, the gradient update amplitudes of a certain convolution kernel weight are g1, g2,..., g100 respectively. Add these gradient update amplitudes to obtain the sum of the gradient update amplitudes of this convolution kernel weight.

[0201] Determine its importance weight according to the sum of the gradient update amplitudes of each model parameter. The sum of the gradient update amplitudes can be normalized so that its value range is between 0 and 1 to obtain the parameter importance weight. Arrange the importance weights of all model parameters according to their positions in the model to form a parameter importance weight matrix.

[0202] Step S3103-6: During the backpropagation optimization process, perform weighted attenuation on the gradients of the cross-channel attention fusion module and the transposed convolution layer parameters according to the parameter importance weight matrix.

[0203] During the backpropagation optimization process, after calculating the gradients of the cross-channel attention fusion module and the transposed convolution layer parameters, perform weighted attenuation on these gradients according to the parameter importance weight matrix.

[0204] For each parameter in the cross-channel attention fusion module and the transposed convolution layer, multiply its gradient by the corresponding parameter importance weight. For example, if the gradient of a certain parameter is g and its corresponding parameter importance weight is w, then the gradient after weighted attenuation is g×w.

[0205] The purpose of this is to make the model pay more attention to the parameters with larger gradient updates and greater impact on the model performance during the training process, while suppressing the updates of the parameters with smaller gradient updates and less importance, so as to improve the training efficiency and performance of the model.

[0206] Step S3103-7: Update the parameter gradients after weighted decay to the pre-trained thermal anomaly detection model to generate an optimized thermal anomaly detection model.

[0207] Apply the parameter gradients after weighted decay to the pre-trained thermal anomaly detection model using an optimization algorithm (such as the stochastic gradient descent algorithm) to update the parameters of the model.

[0208] For example, for a certain parameter, its old parameter value is p, the gradient after weighted decay is g×w, and the learning rate is η, then the new parameter value is p - η×(g×w). Such update operations are performed on all parameters in the cross-channel attention fusion module and the deconvolution layer to obtain an optimized thermal anomaly detection model.

[0209] Step S3103-8: Verify the leakage point localization accuracy of the optimized thermal anomaly detection model on the sample leakage point annotation dataset. If the leakage point localization accuracy is lower than the preset threshold, readjust the filtering threshold and repeat the derivative training.

[0210] Verify the optimized thermal anomaly detection model using the sample leakage point annotation dataset. Input the infrared thermal imaging samples in the sample leakage point annotation dataset into the optimized thermal anomaly detection model to obtain a set of predicted leakage point localization coordinates.

[0211] Compare the set of predicted leakage point localization coordinates with the true leakage point coordinates in the sample leakage point annotation dataset to calculate the leakage point localization accuracy. The calculation method of the leakage point localization accuracy is the number of correctly predicted leakage points divided by the total number of true leakage points.

[0212] For example, there are 100 true leakage points in the sample leakage point annotation dataset, and the number of correctly predicted leakage points by the optimized thermal anomaly detection model is 80, then the leakage point localization accuracy is 80 / 100 = 0.8.

[0213] The preset threshold is set according to actual requirements and model performance requirements. For example, the preset threshold is 0.85. If the verified leakage point localization accuracy is lower than the preset threshold, readjust the filtering threshold. The filtering threshold can be increased or decreased, and then repeat the steps of derivative training, including data augmentation for new samples, feature extraction, calculation of difference degree, sample filtering, parameter update, etc., until the leakage point localization accuracy reaches or exceeds the preset threshold.

[0214] Step S3104: Deploy the optimized pre-trained thermal anomaly detection model to an online detection system to continuously improve the model performance.

[0215] When the leakage point location accuracy of the optimized thermal anomaly detection model on the sample leakage point annotation data set reaches or exceeds a preset threshold, deploy the model to the online detection system.

[0216] The online detection system can receive new infrared thermal imaging data of building facades in real time, process these data using the optimized thermal anomaly detection model, and output a set of leakage point location coordinates and corresponding leakage intensity level identifiers.

[0217] As the online detection system continuously receives new data, users can continue to provide feedback on the detection results, repeat the online optimization process of the thermal anomaly detection model, continuously add new positive and negative samples for derivative training, further optimize the model performance, continuously improve the model performance, so as to more accurately locate the leakage points of the building facade and determine their leakage intensity levels.

[0218] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of an infrared thermal imaging-based exterior wall leakage point location detection system 100 that can implement the inventive concept provided by some embodiments of the present invention. For example, a processor 120 can be used on the infrared thermal imaging-based exterior wall leakage point location detection system 100 and is used to execute the functions in the present invention.

[0219] The infrared thermal imaging-based exterior wall leakage point location detection system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the infrared thermal imaging-based exterior wall leakage point location detection method of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0220] For example, the infrared thermal imaging-based exterior wall leakage point location detection system 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROMs, or RAMs, or any combination thereof. Exemplarily, the infrared thermal imaging-based exterior wall leakage point location detection system 100 can also include program instructions stored in ROMs, RAMs, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The infrared thermal imaging-based exterior wall leakage point location detection system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0221] For ease of explanation, only one processor is described in the exterior wall leakage point location detection system 100 based on infrared thermal imaging. However, it should be noted that the exterior wall leakage point location detection system 100 based on infrared thermal imaging in the present invention may also include multiple processors. Therefore, the steps performed by one processor described in the present invention may also be jointly performed or separately performed by multiple processors. For example, if the processor of the exterior wall leakage point location detection system 100 based on infrared thermal imaging performs step A and step B, it should be understood that step A and step B may also be jointly performed by two different processors or separately performed in one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0222] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned exterior wall leakage point location detection method based on infrared thermal imaging is implemented.

[0223] It should be noted that, in order to simplify the presentation of the disclosure of the present invention and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A method for locating and detecting leakage points on the exterior wall based on infrared thermal imaging, characterized in that The method includes: Collecting a multi - period infrared thermal imaging data set of the outer wall surface of the target building, where the multi - period infrared thermal imaging data set contains surface temperature distribution matrices under different ambient temperature conditions; Performing dynamic temperature correction processing on the multi - period infrared thermal imaging data set to generate a standardized temperature distribution set, which eliminates the influence of ambient temperature fluctuations on the surface temperature measurement values; Extracting a set of thermal characteristic parameters from the standardized temperature distribution set, where the set of thermal characteristic parameters includes temperature gradient distribution characteristics, abnormal temperature difference region morphological characteristics, and heat conduction directionality characteristics; Invoking a pre - trained thermal anomaly detection model to perform leakage risk probability mapping processing on the set of thermal characteristic parameters to generate a leakage point probability distribution map of the outer wall surface of the target building; Performing region segmentation processing based on the leakage point probability distribution map and a preset dynamic decision threshold, and outputting a set of leakage point location coordinates and corresponding leakage intensity level identifiers.

2. The method for detecting and locating the leakage points of the exterior wall based on infrared thermal imaging according to claim 1, wherein, The performing dynamic temperature correction processing on the multi - period infrared thermal imaging data set to generate a standardized temperature distribution set includes: Obtaining a historical meteorological data set of the area where the outer wall of the target building is located, where the historical meteorological data set contains real - time ambient temperature, real - time humidity, and real - time wind speed parameters corresponding to the collection of the multi - period infrared thermal imaging data; Constructing a temperature compensation function, which includes a linear compensation term based on the difference between the real - time ambient temperature and the reference ambient temperature, a non - linear compensation term based on the difference between the real - time humidity and the reference humidity, and an attenuation compensation term based on the difference between the real - time wind speed and the reference wind speed; Inputting each surface temperature measurement value in the multi - period infrared thermal imaging data set into the temperature compensation function to obtain a compensated surface temperature value; Performing spatial interpolation processing on all the compensated surface temperature values to generate a standardized temperature distribution set with the same resolution as the multi - period infrared thermal imaging data set; Wherein, the temperature value of each pixel point in the standardized temperature distribution set represents the true temperature of the outer wall surface after eliminating environmental interference.

3. The method for detecting and locating the leakage point of the exterior wall based on infrared thermal imaging according to claim 1, characterized in that The extracting a set of thermal characteristic parameters from the standardized temperature distribution set includes: Performing spatial gradient calculation on the standardized temperature distribution set to generate a horizontal temperature gradient distribution map and a vertical temperature gradient distribution map, and vectorially superimposing the horizontal temperature gradient distribution map and the vertical temperature gradient distribution map to obtain temperature gradient distribution characteristics; Detecting connected regions in the standardized temperature distribution set where the continuous temperature difference exceeds a preset threshold, and extracting the geometric center coordinates, region area, boundary curvature, and internal temperature extreme point distribution of each connected region to generate abnormal temperature difference region morphological characteristics; Analyzing the temperature change direction of each pixel point in the standardized temperature distribution set with its adjacent pixel points, and statistically calculating the temperature conduction rate in each direction to generate heat conduction directionality characteristics; Aligning and fusing the temperature gradient distribution characteristics, the abnormal temperature difference region morphological characteristics, and the heat conduction directionality characteristics according to spatial coordinates to obtain the set of thermal characteristic parameters.

4. The method for detecting and locating external wall leakage points based on infrared thermal imaging according to claim 3, wherein Invoking the pre-trained thermal anomaly detection model to perform leakage risk probability mapping processing on the set of thermal feature parameters to generate the leakage point probability distribution map of the target building exterior wall, including: Inputting the temperature gradient distribution feature into the first convolutional branch of the pre-trained thermal anomaly detection model to extract high-order abstract features of the spatial distribution of the temperature gradient; Inputting the abnormal temperature difference region morphology feature into the second convolutional branch of the pre-trained thermal anomaly detection model to extract multi-scale geometric features of the region morphology; Inputting the thermal conduction directionality feature into the third convolutional branch of the pre-trained thermal anomaly detection model to extract the temporal evolution features of the thermal conduction path; Performing cross-channel attention fusion on the high-order abstract features, the multi-scale geometric features, and the temporal evolution features to generate a fused feature tensor; Mapping the fused feature tensor to the resolution of the original infrared thermal imaging data through a deconvolution layer, and outputting the leakage risk probability value of each pixel point to generate the leakage point probability distribution map.

5. The method for locating and detecting exterior wall leakage points based on infrared thermal imaging according to claim 1, wherein, The training process of the pre-trained thermal anomaly detection model includes: Obtaining a sample leakage point annotation data set, which contains infrared thermal imaging samples of multiple building exterior walls and their corresponding true leakage point coordinates and leakage intensity labels; Performing the dynamic temperature correction processing and the extraction processing of the set of thermal feature parameters on each of the infrared thermal imaging samples to obtain a sample set of thermal feature parameters; Constructing an initial thermal anomaly detection model, which includes three parallel convolutional branches, a cross-channel attention fusion module, and a deconvolution layer; Inputting the sample set of thermal feature parameters into the initial thermal anomaly detection model, outputting a predicted leakage probability distribution map, and calculating the focal loss function between the predicted leakage probability distribution map and the true leakage point coordinates; Iteratively optimizing the focal loss function through the backpropagation algorithm until the model converges to obtain the pre-trained thermal anomaly detection model.

6. The method for detecting and locating the leakage point of the exterior wall based on infrared thermal imaging according to claim 5, wherein, The calculation of the focal loss function between the predicted leakage probability distribution map and the true leakage point coordinates includes: Performing Gaussian kernel diffusion processing on the true leakage point coordinates to generate a continuous leakage probability density map; Calculating the per-pixel cross-entropy loss between the predicted leakage probability distribution map and the continuous leakage probability density map; Dynamically adjusting the loss weight according to the difference degree between the probability value in the predicted leakage probability distribution map and the true probability value in the continuous leakage probability density map, where higher weights are assigned to difficult-to-classify samples; Introducing a regularization term to constrain the low-probability response intensity in the non-leakage region of the predicted leakage probability distribution map; Performing weighted summation of the per-pixel cross-entropy loss, the dynamically adjusted loss weight, and the regularization term to obtain the final focal loss function.

7. The method for locating and detecting the leakage points of the exterior wall based on infrared thermal imaging according to claim 1, wherein The region segmentation processing based on the leakage point probability distribution map and a preset dynamic determination threshold, and outputting a set of leakage point localization coordinates and corresponding leakage intensity level identifiers, including: Retrieving a basic determination threshold from a threshold configuration library according to the material type and environmental exposure level of the target building exterior wall; Analyze the global statistical characteristics of the probability distribution map of the leakage points, and calculate the mean and variance of the probability values; Based on the mean and variance, dynamically offset and adjust the basic determination threshold to generate a dynamic determination threshold adapted to the current infrared thermal imaging data; Perform threshold segmentation on the probability distribution map of the leakage points, and extract the connected regions where the probability values exceed the dynamic determination threshold; Perform morphological optimization processing on each of the connected regions to remove isolated pixel points caused by noise, and output the set of leakage point location coordinates and the corresponding leakage intensity level identifiers; 8. The method for locating and detecting exterior wall leakage points based on infrared thermal imaging according to claim 7, wherein The generation process of the leakage intensity level identifier includes: Calculate the mean of the probability values of the regions corresponding to each of the leakage point location coordinates as the leakage intensity reference value; Statistically calculate the maximum temperature gradient value, abnormal region area, and thermal conduction direction consistency index in the set of thermal characteristic parameters of this region; Input the leakage intensity reference value, the maximum temperature gradient value, the abnormal region area, and the thermal conduction direction consistency index into a pre-trained intensity classification model, and output discrete leakage intensity levels; Among them, the pre-trained intensity classification model uses the gradient boosting decision tree algorithm, and its training data includes the intensity labels of historical leakage points and the corresponding multi-dimensional feature vectors.

9. The method for locating and detecting external wall leakage points based on infrared thermal imaging according to claim 5, wherein The method also includes the online optimization process of the pre-trained thermal anomaly detection model: Receive the feedback information of the user on the set of leakage point location coordinates, and the feedback information includes misdetection location marks and missed detection location marks; Extract the set of thermal characteristic parameters corresponding to the misdetection location marks as negative samples, and the set of thermal characteristic parameters corresponding to the missed detection location marks as positive samples; Add the newly added positive and negative samples to the sample leakage point annotation data set, and trigger the derivative training process of the pre-trained thermal anomaly detection model; Freeze the parameters of some convolutional layers during the derivative training process, and only optimize the parameters of the cross-channel attention fusion module and the deconvolution layer; Deploy the optimized pre-trained thermal anomaly detection model to the online detection system to continuously improve the model performance; Among them, the specific steps of the derivative training process include: Perform rotation, translation, and scale transformation on the newly added positive and negative samples to generate an enhanced sample set; Extract the standardized temperature distribution set and the set of thermal characteristic parameters of the enhanced sample set and the original samples in the sample leakage point annotation data set to generate a multi-dimensional feature vector set; Calculate the Euclidean distance between the multi-dimensional feature vectors of the enhanced sample set and the multi-dimensional feature vectors of the original samples to generate a difference degree distribution matrix; Determine the filtering threshold according to the cumulative distribution function of the difference degree distribution matrix, and remove the samples in the enhanced sample set whose difference degree exceeds the filtering threshold to generate a filtered enhanced sample set; Based on the number of model training iterations of the filtered enhanced sample set, statistically calculate the gradient update amplitude of each model parameter to generate a parameter importance weight matrix; During the backpropagation optimization process, perform weighted attenuation on the gradients of the cross-channel attention fusion module and the deconvolution layer parameters according to the parameter importance weight matrix; Update the parameter gradient after weighted attenuation to the pre-trained thermal anomaly detection model to generate an optimized thermal anomaly detection model; Verify the leakage point localization accuracy of the optimized thermal anomaly detection model on the sample leakage point annotation dataset. If the leakage point localization accuracy is lower than the preset threshold, readjust the filtering threshold and repeat the derivative training.

10. An exterior wall leakage point positioning and detection system based on infrared thermal imaging, characterized in that, It includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the infrared thermal imaging-based exterior wall leakage point localization and detection method according to any one of claims 1-9 above.

Images