A method for detecting a hot infrared building water penetration point of a drone for digital twinning

Abstract

This invention discloses a method for detecting building seepage points using UAV thermal infrared imaging for digital twin purposes, relating to the field of image processing technology. The method includes the following steps: First, acquiring thermal infrared images of the building from a UAV; based on the temperature difference between the seepage point and the background, extracting thermal anomaly regions through adaptive threshold segmentation and morphological processing; and generating candidate seepage regions using connected component analysis. Then, using these candidate regions as regions of interest, a deep learning object detection model identifies the location and category of the seepage points. Finally, the detection results are converted into structured state parameters and input into the digital twin model to update the seepage state variables of the corresponding building components. This invention improves the accuracy and robustness of seepage point identification by generating candidate seepage regions to constrain deep learning object detection, and by structurally updating the detection results to the digital twin model, it achieves intelligent and automated closed-loop perception of building seepage status.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method for detecting building water seepage points using unmanned aerial vehicles (UAVs) with thermal infrared technology for digital twins. Background Technology

[0002] Building seepage, especially in roofs and facades, is a common problem affecting structural safety and lifespan. Detection technology has evolved from manual inspection to sensor-assisted methods and then to drone remote sensing. Traditional manual inspection relies on experienced engineers visually inspecting, tapping, or using simple handheld devices. This method is inefficient, has limited coverage, and poses safety hazards due to working at heights. The results are often unstructured textual descriptions or on-site photographs, lacking standardized quantitative criteria. Handheld infrared thermal imager inspection utilizes surface temperature anomalies caused by water evaporation and cooling or differences in heat capacity in the seepage area to aid in judgment. However, this is still a discrete, manual inspection method with a small field of view, inconsistent shooting distances and angles, and inconsistent data quality. Furthermore, the interpretation of thermal infrared images is highly dependent on the experience of professionals, making large-scale, standardized automated analysis difficult. In recent years, visible light detection methods based on drones have acquired images of building surfaces using high-definition cameras and identified leaks using visual features such as water stains and discoloration. While this method improves acquisition efficiency, visible light can only reflect surface optical signals and cannot directly perceive the physical presence of moisture. It is also susceptible to interference from lighting, shadows, dirt, and material color differences, posing a fundamental risk of missed detection for hidden leaks or leaks in areas similar in color to the background. Furthermore, some studies have attempted to combine drone thermal infrared images with 3D building models (such as BIM), but most only involve simple overlay of images and models, lacking in-depth processing of thermal infrared images and intelligent identification mechanisms for leak targets. This makes it difficult to form a closed-loop detection process from data acquisition and feature extraction to model state updates. Therefore, how to provide a method that can automatically and stably identify leak points in thermal infrared images and integrate the detection results as structured state parameters into a digital twin model has become a pressing technical problem in this field. Summary of the Invention

[0003] To automatically and reliably identify water seepage points in thermal infrared images and integrate the detection results as structured state parameters into a digital twin model, this invention proposes a UAV-based thermal infrared building water seepage point detection method for digital twins, comprising the following steps: S1: Acquire thermal infrared images of buildings collected by a thermal infrared imaging device mounted on a drone; S2: Based on the temperature difference between the seepage area and the background area, adaptive threshold segmentation and morphological processing are performed on the thermal infrared image to extract the thermal anomaly area. S3: Perform connected component analysis on the extracted thermal anomaly region to generate at least one candidate region for water seepage to characterize the range of areas where water seepage may exist in the thermal infrared image. S4: Use a pre-trained deep learning object detection model to identify seepage points in the seepage candidate region and output the location and category information of the seepage points in the thermal infrared image; S5: Convert location and category information into structured state parameters and input them into the digital twin model to update the seepage state variables of the corresponding building components in the digital twin model.

[0004] This invention generates seepage candidate regions by adaptive segmentation and morphological processing of thermal infrared images, and performs accurate identification by combining deep learning target detection. Finally, the structured detection results are directly updated to update the seepage state variables of the digital twin model, thereby achieving high-precision, low-background-interference automated seepage detection and twin model closed-loop update.

[0005] Furthermore, in step S2, the adaptive threshold segmentation employs the Otsu algorithm, specifically as follows: Based on the criterion of maximizing the inter-class variance, the inter-class variance of each gray value is calculated in the gray-level range of the thermal infrared image as a candidate threshold. The gray value corresponding to the maximum value of the inter-class variance is determined as the segmentation threshold. The thermal infrared image is divided into a binary image of thermal anomaly region and background region using this segmentation threshold.

[0006] Furthermore, in step S2, the morphological processing specifically includes: Closing operations are used to fill the voids inside the thermal anomaly region, opening operations are used to remove isolated noise points, and erosion operations are used to separate adjacent thermal anomaly regions.

[0007] Furthermore, in step S3, the connection component analysis of the extracted thermal anomaly region specifically involves: Connected component labeling is performed on the binary image corresponding to the extracted thermal anomaly region. Adjacent pixels with the same pixel value in the binary image are aggregated into independent connected components. The pixel area occupied by each connected component in the binary image is calculated. Based on the ratio of the pixel area of ​​each connected component to the pixel area of ​​the entire thermal infrared image, connected components whose area ratio exceeds a preset threshold range are removed. The image regions corresponding to the remaining connected components are used as the water seepage candidate regions.

[0008] Furthermore, in step S4, the deep learning object detection model is a YOLO series object detection network.

[0009] Furthermore, the training process of the deep learning object detection model includes: The candidate regions for seepage generated in step S3 are used to filter or label thermal infrared images, a training dataset containing seepage target samples is constructed, and the deep learning target detection model is iteratively trained based on the training dataset.

[0010] Furthermore, in step S5, the structured state parameters include the spatial coordinates of the seepage point in the digital twin model, the area of ​​the seepage region, and the identification confidence level.

[0011] Furthermore, step S5 is followed by the following step: S6: Based on the updated seepage state variables, analyze the spatial distribution and changing trend of seepage state and compare it with historical state, assess the seepage risk level, and provide early warning information to operation and maintenance personnel based on the assessment results.

[0012] Furthermore, in step S1, when the UAV collects thermal infrared images, the shooting distance is preset according to the requirements of operational efficiency and spatial resolution of thermal infrared images; the collection period is selected after the surface temperature of the building roof and the ambient temperature have reached thermal equilibrium and the solar altitude angle is lower than a preset threshold; and multiple collections are performed at preset time intervals after a rainfall event to form a thermal infrared image dataset covering different ordered time periods.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes a method for detecting building seepage points by UAV thermal infrared based on digital twins. Based on the temperature difference between the seepage area and the background area, the Otsu algorithm is used to perform adaptive threshold segmentation, morphological processing and connected component analysis to quickly generate candidate areas for characterizing the potential seepage range, effectively eliminating the interference of large-area background thermal noise and isolated noise points. (2) The candidate region is taken as the region of interest and the YOLO series deep learning object detection model is used for accurate identification. The candidate region is used to assist in the construction of the training dataset, which reduces the cost of large-scale fine annotation and thus improves the accuracy and stability of seepage point detection. (3) The structured state parameters such as the location, type and confidence level of the detected seepage points are input into the digital twin model to dynamically update the seepage state variables of the corresponding building components. Based on the spatial distribution and change trend, risk level assessment and early warning are carried out, providing high-quality and quantifiable dynamic perception data for the digital twin model. Attached Figure Description

[0014] Figure 1 A step-by-step diagram of a UAV thermal infrared detection method for building seepage points based on digital twins; Figure 2 This is a schematic diagram of a digital twin system based on seepage point detection. Detailed Implementation

[0015] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0016] This invention provides a method for detecting building seepage points using UAV thermal infrared imaging, geared towards digital twins. This method aims to address the problems of low operational efficiency, poor detection stability, lack of quantitative analysis indicators, and difficulty in effectively integrating with digital twin systems in existing building seepage detection technologies. Through this method, seepage areas can be automatically and accurately identified from thermal infrared images acquired by UAVs, and the detection results can be updated to the digital twin model in the form of structured parameters, providing scientific data support for building operation and maintenance management. Figure 1 As shown, the method mainly includes the following steps: S1: Acquire thermal infrared images of buildings collected by a thermal infrared imaging device mounted on a drone; S2: Based on the temperature difference between the seepage area and the background area, adaptive threshold segmentation and morphological processing are performed on the thermal infrared image to extract the thermal anomaly area. S3: Perform connected component analysis on the extracted thermal anomaly region to generate at least one candidate region for water seepage to characterize the range of areas where water seepage may exist in the thermal infrared image. S4: Use a pre-trained deep learning object detection model to identify seepage points in the seepage candidate region and output the location and category information of the seepage points in the thermal infrared image; S5: Convert location and category information into structured state parameters and input them into the digital twin model to update the seepage state variables of the corresponding building components in the digital twin model.

[0017] In a specific application scenario, taking the roof leakage detection of a commercial complex in a city as an example, the first step is to prepare for the inspection. A quadcopter drone is selected, equipped with a high-resolution thermal infrared imaging device and a high-precision GPS / IMU integrated navigation system. Based on the geometric dimensions and height information of the building to be inspected, combined with local meteorological conditions (such as wind speed and sunlight), the drone's flight path is pre-planned.

[0018] To balance operational efficiency and spatial resolution of thermal infrared images, the UAV's flight altitude was set at approximately 30 meters to ensure that the ground sampling resolution of the thermal infrared images could clearly distinguish the small structures of the roof. Data acquisition was conducted before sunrise, as the roof surface temperature distribution was relatively uniform after a night of radiative cooling, and it was not subject to strong thermal interference from direct sunlight. The temperature difference between the seepage area, where the low-temperature anomaly caused by water evaporation and cooling, and the background was most significant. Simultaneously, three data acquisition tasks were performed at 24, 48, and 72 hours after the end of a heavy rainfall event, respectively, to obtain a multi-timeframe thermal infrared image sequence reflecting the evolution of the seepage area over time. The UAV performed a full-coverage scan of the building roof according to a pre-set spiral progressive flight path. The acquired raw thermal infrared images were single-channel grayscale images, with the grayscale value of each pixel representing the thermal radiation intensity of the corresponding ground feature. These images, along with the UAV's position and attitude data, were wirelessly transmitted to the ground control station as raw data for subsequent processing.

[0019] After acquiring the original thermal infrared images, it is necessary to initially separate potentially water-permeable thermal anomaly areas from these images. Because of the temperature difference between the permeable areas and the non-permeable background areas due to water evaporation and cooling, this difference manifests in the thermal infrared images as a significant step or gradient change in grayscale values ​​at the boundaries of the permeable areas. However, different building materials (such as concrete, asphalt, and waterproof membranes) have different heat absorption characteristics during the day and heat dissipation characteristics at night, resulting in significant differences in their absolute temperature values, making it impossible to use a fixed temperature threshold for segmentation. Therefore, this invention employs an adaptive threshold segmentation method, specifically the Otsu algorithm. This algorithm automatically determines the optimal segmentation threshold based on the image's grayscale histogram, using the maximization of the inter-class variance between the target and the background as the criterion. For an input thermal infrared grayscale image, assuming its grayscale range is 0 to 255 and the total number of pixels is... grayscale The corresponding number of pixels is The probability of this gray level appearing is Assuming we use grayscale value T as the candidate segmentation threshold, we divide image pixels into a background class (grayscale value 0 to T) and a target class (grayscale value T+1 to 255). We then calculate the grayscale mean and inter-class variance for each class. By iterating through all possible T values, we find the T that maximizes the inter-class variance; this is the optimal segmentation threshold. Using this threshold, we perform binarization on the image. Pixels with grayscale values ​​greater than T are marked as foreground (thermal anomaly areas), and the rest as background.

[0020] Because rooftop equipment, pipes, and shadows can cause temperature differences, directly segmented binary images often contain many isolated noise points, small holes, and interconnected anomalous regions. Therefore, a series of morphological processing steps are required. First, a closing operation is used, involving dilation followed by erosion, to fill in the small voids within thermal anomaly areas caused by noise or localized temperature inhomogeneities, making the anomaly areas more complete in shape. Next, an opening operation is used, involving erosion followed by dilation, to remove small, isolated noise points, typically caused by sensor noise or tiny hot spots unrelated to water seepage. Finally, an erosion operation is performed using appropriately sized structuring elements to separate adjacent anomalous regions that are interconnected due to gentle temperature gradients, ensuring that each independent anomalous region corresponds to a potential heat source.

[0021] After morphological processing, a relatively clean binary image composed of multiple white pixel blocks (i.e., thermal anomaly regions) is obtained. Next, connected component analysis is performed on these white pixel blocks to filter out candidate regions that are truly likely to correspond to water seepage points. Specifically, the binary image is labeled with connected components, aggregating spatially adjacent pixels with a value of 1 into an independent connected component. For each labeled connected component, the pixel area it covers in the image is calculated. Considering that water seepage areas typically do not occupy a large portion of the entire image, but should not be so small as to be negligible, an area ratio threshold range can be set empirically. For example, the ratio of the connected component pixel area to the total pixel area of ​​the entire image should be between 0.1% and 20%. All connected components are traversed, and those with an area ratio exceeding this range are eliminated. For example, areas with excessively large areas may correspond to large areas of water accumulation or equipment overheating zones, while areas with excessively small areas may be random noise. The regions in the original image corresponding to the retained connected components are extracted as candidate water seepage regions. These candidate regions significantly narrowed the search range that required further detailed detection, while filtering out most irrelevant background thermal interference.

[0022] After obtaining candidate seepage areas, this invention further employs a deep learning object detection model to accurately identify the image content within these candidate areas, in order to determine whether the thermal anomalies are actual seepage points or other heat sources (such as exposed metal pipes, vents, cables, etc.). In this embodiment, the deep learning object detection model uses the YOLOv8 network framework. The YOLOv8 network typically consists of a backbone, a neck, and a head. The backbone is responsible for extracting multi-level feature maps from the input image; the neck uses a feature pyramid structure to fuse features at different scales to enhance the detection capability for multi-scale targets; and the head is responsible for predicting the bounding box, class probability, and confidence score of the target on each grid of the feature map.

[0023] To train the model, a training dataset containing a large number of thermal infrared seepage samples is needed. Since obtaining large-scale, high-quality manually labeled thermal infrared seepage images in practical engineering is costly and inefficient, this invention utilizes seepage candidate regions generated in the aforementioned steps to assist in dataset construction. Specifically, for a batch of acquired thermal infrared images, a candidate region generation algorithm is first run to obtain a series of candidate regions. Then, the annotators only need to judge and label within these candidate regions, marking areas confirmed as seepage points as positive samples and clearly non-seepage thermal anomalies (such as pipes and equipment) as negative samples. Since the candidate regions have eliminated most background areas, the workload of annotation is significantly reduced, and the samples are more targeted. After annotation, image patches with bounding boxes and category labels are input as training samples into the YOLOv8 network for training. During training, a stochastic gradient descent optimizer is used, and data augmentation techniques (such as random flipping, rotation, and color jitter) are applied to improve the model's generalization ability. After multiple rounds of iterative training, when the average accuracy of the model on the validation set reaches a preset threshold (e.g., 0.85), a seepage point detection model suitable for practical detection is obtained. During the model inference phase, the thermal infrared image to be detected, along with the seepage candidate regions generated in step S3, are input into the model. The model does not perform global detection on the entire image; instead, it infers only the image content within the candidate regions. For each candidate region, the model outputs a series of bounding boxes with confidence scores. Each bounding box corresponds to a location identified as a seepage point, and the model also outputs the category of the seepage point (e.g., it can be classified as point seepage, linear seepage, or area seepage). Because the search range has been significantly narrowed, the model's inference speed is fast, and the false alarm rate is significantly lower than that of directly detecting on the entire image.

[0024] After obtaining the location and category information of the seepage points in the thermal infrared image, this information needs to be transformed into structured state parameters that can be understood and used by the digital twin model. First, the pixel coordinates of the seepage points in the image coordinate system need to be converted into three-dimensional spatial coordinates in the world coordinate system (usually a geocentric or engineering coordinate system) of the digital twin model. This conversion relies on the position and attitude data of the UAV when capturing the image frame, including the UAV's GPS coordinates (longitude, latitude, altitude), pitch angle, roll angle, and yaw angle, as well as the intrinsic parameter matrix of the thermal infrared camera (such as focal length, principal point coordinates, and distortion coefficients). A mapping relationship between image pixel coordinates and object space coordinates can be established using collinearity equations or direct linear transformation algorithms. Since the roof surface can usually be approximated as a plane or has a known BIM model, the digital elevation model of the roof can be further utilized to improve the mapping accuracy. For each detected seepage point, its corresponding three-dimensional spatial coordinates are calculated based on its pixel coordinates and the UAV pose, and the area of ​​the seepage region (converted from the pixel area within the bounding box) and the identification confidence level are recorded. These data collectively constitute a structured set of state parameters. This set of state parameters is then input into the digital twin model of the target building.

[0025] A digital twin model is a digital mirror image of a building, built on BIM (Building Information Modeling) and containing its geometric, physical, and functional information. Each building component in the model (such as a specific roof panel, waterproofing layer, structural beam, etc.) corresponds to a unique identifier and is associated with several state variables, such as "leakage status," "leakage area," and "last inspection time." By calling the digital twin model's API interface, the calculated spatial coordinates of the seepage point are spatially matched with the components in the model to find the affected component. Then, the corresponding state variables of that component are updated; for example, the seepage status is updated from "normal" to "leaking," the seepage area is updated to the inspection value, and the timestamp and confidence level of this inspection are recorded.

[0026] Furthermore, this embodiment of the invention also includes a step of performing state analysis and early warning based on the updated digital twin model (i.e., step S6). The digital twin model integrates the results of multiple historical detections and can display the changing trend of the seepage status of the same component in a time series, such as whether the area of ​​the seepage zone is expanding and whether the confidence level is increasing. Through spatial analysis, a heat map of the distribution of seepage points on the roof can be displayed, intuitively identifying high-risk areas. Combining building structural information and local meteorological data (such as future rainfall forecasts), the impact of seepage on building safety and functionality can be assessed, and risk levels can be classified, for example, areas where the seepage area is continuously expanding and located above major load-bearing components can be marked as high-risk. The system can automatically generate early warning information according to preset rules and send it to building operation and maintenance management personnel via SMS, email, or APP push, reminding them to arrange on-site verification or repair in a timely manner. At the same time, the detection results can also be used to guide the next drone inspection plan, such as to conduct intensive observation of high-risk areas, thereby forming a closed-loop management mechanism.

[0027] In actual implementation, the parameters, models, thresholds, etc., in the above embodiments can be adaptively adjusted according to specific application scenarios. For example, for rooftops in different climate zones and with different building types, the flight altitude, data collection period, and grayscale range of the Otsu algorithm can be adjusted; for scenarios requiring higher detection accuracy, YOLOv8 can be replaced with other types of deep learning detection networks (such as Faster R-CNN or Transformer-based detectors); for the interface form of the digital twin model, a common data exchange standard such as IFC or CityGML can be adopted. These adjustments do not depart from the scope of protection of this invention.

[0028] Based on the above, the technical solution of the present invention is further improved by... Figure 2 The integrated system shown achieves closed-loop interaction between physical space and digital twin space. In specific implementation, such as... Figure 2As shown, the UAV conducts an inspection flight along a planned route over the roof of the target building, simultaneously acquiring thermal infrared images via a thermal infrared camera and transmitting them to a ground processing system. The system first performs preliminary screening of the thermal infrared images to obtain potential seepage areas, generating candidate regions that may contain seepage points. These candidate regions are then input into a pre-trained deep learning object detection model to accurately identify seepage points and output seepage point state parameters, including the pixel location of the seepage point in the image, the estimated area of ​​the seepage region, and the model's output recognition confidence level. Based on this, and combined with the position and attitude data recorded by the high-precision positioning system onboard the UAV, these data are mapped to the three-dimensional spatial coordinates of the seepage region in a digital twin model through coordinate transformation and written into the corresponding building component state variables, thus completing the real-time synchronization of physical inspection data to the virtual model. The built-in state analysis and assessment module of the digital twin model comprehensively analyzes the spatial distribution density of seepage points, confidence thresholds, and historical state change trends based on updated state variables. It dynamically classifies the current seepage risk level and, combined with the building material properties and load conditions stored in the model, determines whether to generate early warning information for maintenance personnel. When the assessment results reach the preset early warning threshold, the system automatically feeds back the early warning information and the spatial coordinates of the seepage area to the physical system's maintenance terminal, guiding on-site personnel to conduct targeted inspections and repairs. Simultaneously, this feedback can also be used to dynamically adjust the flight path planning for subsequent drone inspections, such as adding waypoints to high-risk areas or shortening the repetitive inspection cycle. Thus, through a complete chain of "data acquisition - candidate area generation - seepage point detection - state parameter extraction - twin model update - risk analysis and feedback," a closed-loop update mechanism for the digital twin model is formed, continuously improving the intelligence level of seepage detection and maintenance.

[0029] In summary, this invention acquires thermal infrared images of buildings using a drone equipped with a thermal infrared imaging device. First, based on the temperature difference between the seepage area and the background area, the Otsu algorithm is used for adaptive threshold segmentation, combined with morphological processing and connected component analysis, to quickly generate candidate regions to characterize the potential seepage range, effectively eliminating interference from large-area background thermal noise and isolated noise points. Then, this candidate region is used as the region of interest, and the YOLO series of deep learning object detection models are used for accurate identification. Furthermore, the candidate region is used to assist in building a training dataset, reducing the cost of large-scale fine annotation, thereby improving the accuracy and stability of seepage point detection. Finally, the structured state parameters such as the location, category, and confidence level of the detected seepage points are input into a digital twin model to dynamically update the seepage state variables of the corresponding building components. Based on the spatial distribution and change trends, risk level assessment and early warning are performed, providing high-quality, quantifiable, and dynamically perceptual data for the digital twin model.

[0030] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0031] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0032] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0033] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A method for detecting building seepage points using unmanned aerial vehicle (UAV) thermal infrared technology for digital twins, characterized in that, Including the following steps: S1: Acquire thermal infrared images of buildings collected by a thermal infrared imaging device mounted on a drone; S2: Based on the temperature difference between the seepage area and the background area, adaptive threshold segmentation and morphological processing are performed on the thermal infrared image to extract the thermal anomaly area. S3: Perform connected component analysis on the extracted thermal anomaly region to generate at least one candidate region for water seepage to characterize the range of areas where water seepage may exist in the thermal infrared image. S4: Use a pre-trained deep learning object detection model to identify seepage points in the seepage candidate region and output the location and category information of the seepage points in the thermal infrared image; S5: Convert location and category information into structured state parameters and input them into the digital twin model to update the seepage state variables of the corresponding building components in the digital twin model.

2. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 1, characterized in that, In step S2, the adaptive threshold segmentation uses the Otsu algorithm, specifically: Based on the criterion of maximizing the inter-class variance, the inter-class variance of each gray value is calculated in the gray-level range of the thermal infrared image as a candidate threshold. The gray value corresponding to the maximum value of the inter-class variance is determined as the segmentation threshold. The thermal infrared image is divided into a binary image of thermal anomaly region and background region using this segmentation threshold.

3. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 2, characterized in that, In step S2, the morphological processing specifically includes: Closing operations are used to fill the voids inside the thermal anomaly region, opening operations are used to remove isolated noise points, and erosion operations are used to separate adjacent thermal anomaly regions.

4. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 2, characterized in that, The S3 step is specifically as follows: Connected component labeling is performed on the binary image corresponding to the extracted thermal anomaly region. Adjacent pixels with the same pixel value in the binary image are aggregated into independent connected components. The pixel area occupied by each connected component in the binary image is calculated. Based on the ratio of the pixel area of ​​each connected component to the pixel area of ​​the entire thermal infrared image, connected components whose area ratio exceeds a preset threshold range are removed. The image regions corresponding to the remaining connected components are used as the water seepage candidate regions.

5. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 1, characterized in that, In step S4, the deep learning object detection model is the YOLO series object detection network.

6. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 5, characterized in that, The training process of the deep learning object detection model includes: The candidate regions for seepage generated in step S3 are used to filter or label thermal infrared images, a training dataset containing seepage target samples is constructed, and the deep learning target detection model is iteratively trained based on the training dataset.

7. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 1, characterized in that, In step S5, the structured state parameters include the spatial coordinates for locating the seepage point in the digital twin model, the area of ​​the seepage region, and the identification confidence level.

8. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 1, characterized in that, The step S5 is followed by the following step: S6: Based on the updated seepage state variables, analyze the spatial distribution and changing trend of seepage state and compare it with historical state, assess the seepage risk level, and provide early warning information to operation and maintenance personnel based on the assessment results.

9. The method for detecting building seepage points using UAV thermal infrared technology for digital twins as described in claim 1, characterized in that, In step S1, when the UAV collects thermal infrared images, the shooting distance is preset according to the requirements of work efficiency and spatial resolution of thermal infrared images; the collection period is selected after the surface temperature of the building roof and the ambient temperature reach thermal equilibrium and the solar altitude angle is lower than a preset threshold; and multiple collections are performed at preset time intervals after a rainfall event to form a thermal infrared image dataset covering different ordered time periods.

Images