Single-frame infrared thermal imaging building outer wall hollow detection method and system based on heat flow evaluation

Abstract

The application discloses a kind of based on heat flow evaluation's single frame infrared thermal imaging building outer wall cavity detection method and system, and wherein method includes: obtaining the single frame infrared thermal image and visible light image of synchronous acquisition;Based on the Green function of single frame infrared thermal image and heat conduction control equation generates infrared thermal imaging chart sequence;Infrared thermal imaging chart sequence is carried out time-space frequency domain transformation and column harmonic function inversion, again inverse transformation back time-space domain, generates virtual heat flow image sequence;To single frame virtual heat flow image is successively carried out binaryzation processing, discriminant function enhancement processing, obtains outer wall enhancement binaryzation image;Using target detection network identifies interference from visible light image;The result of identification in visible light image is registered with infrared thermal image, and mapping is carried out, eliminates interference in outer wall enhancement binaryzation image, obtains pure outer wall binaryzation image.The application has the advantages of strong environmental adaptability, high recognition precision, and is suitable for the safety detection and maintenance of building outer wall.

Description

Technical Field

[0001] This invention relates to the field of building safety inspection technology, and in particular to a method and system for detecting hollow areas in building exterior walls using single-frame infrared thermal imaging based on heat flow assessment. Background Technology

[0002] Exterior wall detachment from high-rise buildings has become a critical threat to urban public safety. With the increasing density of high-rise buildings in urban built-up areas, incidents of falling exterior wall components such as tiles, paint, and stone are frequent. These incidents not only pose direct hazards such as pedestrian injuries and vehicle damage, but also disrupt the function of urban public facilities, affect residents' sense of security, and severely challenge the city's operational order. Therefore, efficient and accurate exterior wall defect detection technology is a core support for ensuring the structural safety of high-rise buildings and urban public safety.

[0003] Traditional methods for detecting defects in exterior walls primarily rely on manual visual inspection and tapping tests. Manual visual inspection requires inspectors to access the exterior wall surface using suspended platforms, scaffolding, or other high-altitude equipment, and then visually inspect the wall surface for signs such as cracks, hollow areas, and peeling. This method is highly dependent on the experience level of the inspectors; different personnel may have different standards for judging defects, leading to strong subjectivity and a high risk of missed defects. Furthermore, working at height presents a complex environment, constrained by weather factors such as wind and rainfall, resulting in extremely low inspection efficiency—inspecting the exterior wall of a 30-story building often takes several weeks—and carries safety risks such as falls and impacts, making it difficult to guarantee operational safety. Tapping tests involve inspectors tapping the wall with tools and judging the presence of hollow areas based on differences in sound. However, this method relies on the characteristics of sound propagation; differences in wall thickness and material density can lead to variations in sound feedback, making it difficult to accurately identify deep hollow defects. It also requires working at height, presenting both efficiency and safety issues.

[0004] To address the shortcomings of traditional methods, infrared thermal imaging technology has been introduced into the field of exterior wall hollowness detection. This technology indirectly determines the presence of hollow areas by capturing differences in the distribution of thermal radiation on the wall surface. However, in practical applications, this technology still suffers from three major drawbacks, severely hindering its widespread adoption.

[0005] First, there is the issue of sensitivity to temperature interference. Existing infrared thermal imaging detection methods lack a dynamic temperature field model, making it impossible to effectively eliminate the influence of ambient temperature and solar radiation changes on the detection results. During daytime sunlight exposure, the light intensity received by walls facing different directions in high-rise buildings varies significantly, with surface temperature differences between the sun-facing and shaded sides reaching 5-8 degrees Celsius. The surface temperature distribution of the walls exhibits non-uniform dynamic changes. Simultaneously, fluctuations in ambient temperature during the morning and evening, as well as shading from surrounding buildings, further exacerbate the surface temperature disturbances. Due to the lack of precise modeling of the wall's heat conduction process, existing methods cannot distinguish between thermal anomalies caused by hollow areas and normal temperature fluctuations caused by environmental factors. For example, due to the insulating effect of air, the heat conduction speed in hollow areas is slower than in normal walls, exhibiting specific thermal characteristics. However, when the ambient temperature changes abruptly, similar thermal characteristics may also appear in parts of normal walls, leading to a significant increase in the false detection rate of hollow areas, with false detection rates exceeding 40% in some scenarios.

[0006] Secondly, there is a high dependence on labeled data and insufficient generalization ability. Existing infrared thermal imaging detection technologies are mostly based on machine learning models, requiring a large amount of labeled data to train the models to identify hollow features. However, acquiring labeled data is extremely costly, requiring professionals to accurately select hollow areas in infrared images while recording parameters such as the material type, thickness, and construction process of the corresponding wall. This process is not only time-consuming and labor-intensive but also requires labelers to possess both infrared imaging technology and architectural structural expertise. The scarcity of such talent further increases labeling costs. More importantly, the thermal properties of different building materials vary significantly. The thermal conductivity and thermal emissivity of ceramic tile exterior walls, real stone paint exterior walls, and stone curtain walls can differ by 2-3 times. Models trained on data from a single material will experience a significant drop in accuracy when applied to the detection of other wall materials. In actual engineering projects, the same building complex may contain multiple exterior wall materials, and the insufficient generalization ability of the model makes it difficult to guarantee the reliability of the detection results.

[0007] Thirdly, there is the problem of misclassification of non-wall objects. High-rise building exteriors often have non-wall objects attached, such as air conditioner units, pipes, rainwater pipes, and billboards. These objects exhibit unique thermal characteristics in infrared thermal images. For example, the surface temperature of an air conditioner unit is higher than that of the wall during operation, and metal pipes show abnormal temperature zones due to their different thermal conductivity characteristics. Existing detection methods mostly process infrared or visible light images separately, failing to utilize the complementarity of the two modalities. When processing infrared images alone, temperature anomalies in non-wall objects are easily misclassified as hollow defects; when processing visible light images alone, although wall and non-wall structures can be distinguished, internal defects such as hollow defects cannot be identified. This fragmented use of the two modalities leads to a persistently high misclassification rate for non-wall objects, severely reducing detection accuracy, and in some scenarios, reducing the percentage of effective detection results to less than 60%.

[0008] Infrared thermal imaging technology, as a non-contact detection method, should have the advantages of safety and efficiency. However, its existing defects result in low detection accuracy when applied to the detection of hollow areas in building exterior walls, making it difficult to meet the actual needs of safety inspection of exterior walls of high-rise buildings. Summary of the Invention

[0009] This invention provides a method and system for detecting hollow areas in building exterior walls using single-frame infrared thermal imaging based on heat flow assessment, in order to solve the problem of low detection accuracy in existing technologies that use infrared thermal imaging technology to detect hollow areas in building exterior walls.

[0010] Firstly, a single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment is provided, comprising the following steps:

[0011] Acquire single-frame infrared thermal images and visible light images of the building's exterior walls simultaneously;

[0012] Based on a single-frame infrared thermal image and the thermal properties of the wall material, a virtual, time-continuous infrared thermal imaging sequence is generated using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall.

[0013] The infrared thermal imaging sequence is subjected to time-space frequency domain transformation and cylindrical harmonic function inversion, and then inversely transformed back to the time-space domain to generate a virtual heat flow image sequence corresponding to a single frame of infrared thermal image;

[0014] Select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall;

[0015] The binarized image of the exterior wall is enhanced by using a discriminant function to obtain an enhanced binarized image of the exterior wall;

[0016] A deep learning-based target detection network is used to process synchronously acquired visible light images to identify and locate non-wall structure interference objects;

[0017] The recognition results in the visible light image are registered and mapped with the infrared thermal image. The regions corresponding to non-wall structure interference objects in the enhanced binarized image of the exterior wall are removed to obtain a pure binarized image of the exterior wall.

[0018] Furthermore, the infrared thermal imaging sequence is generated using the following method:

[0019] The average thermal conductivity at the image scale is calculated based on the spatial resolution parameters of the infrared thermal image, the acquisition frequency, and the average thermal conductivity of the wall.

[0020] Using the acquired single-frame infrared thermal image as input, and combining the Green's function of the heat conduction control equation describing the heat conduction law of the wall with the image-scale average heat conduction coefficient, a virtual infrared thermal imaging sequence of length N, which is continuous and equally spaced in the time dimension, is generated.

[0021] Furthermore, the virtual heat flow image sequence is generated using the following method:

[0022] Gaussian filtering is applied to the infrared thermal imaging sequence;

[0023] Perform Fourier-Hankel transform on the filtered infrared thermal image sequence to transform the image from the spatiotemporal domain to the frequency domain;

[0024] Adjust the ratio of the spatiotemporal frequency domain transformation axis according to the infrared thermal imaging sequence length N and the acquisition time, and recalibrate the time and frequency axis;

[0025] Using cylindrical harmonic functions as the core, the infrared thermal imaging sequence transformed to the frequency domain is inverted.

[0026] The frequency domain infrared thermal imaging sequence after inversion processing is restored to the spatiotemporal domain through inverse transformation to obtain a virtual heat flow image sequence characterizing the heat flow distribution.

[0027] Furthermore, the binarized image of the exterior wall is enhanced by a discriminant function to obtain an enhanced binarized image of the exterior wall, specifically including:

[0028] Enhancement processing based on the sigmoid discriminant function is performed on the binarized image of the exterior wall. The sigmoid discriminant function maps each pixel in the binarized image of the exterior wall to a hollow probability value of 0~1. Then, by comparing with a set probability threshold, binarization is performed again to obtain the enhanced binarized image of the exterior wall.

[0029] Furthermore, the deep learning-based object detection network is obtained through the following method:

[0030] First, train the object detection network on a general object detection dataset to obtain the basic model;

[0031] The base model was transferred to a non-wall structure interference detection dataset for fine-tuning to obtain the final deep learning-based object detection network.

[0032] Furthermore, the purified binary image of the exterior wall is obtained through the following method:

[0033] Select common feature points from visible light images and infrared thermal images;

[0034] The coordinate transformation matrix is ​​calculated by the feature point matching algorithm. This coordinate transformation matrix is ​​then used to convert the coordinates of the non-wall structure interference target boxes identified from the visible light image into the corresponding target box coordinates in the infrared thermal image.

[0035] Based on the coordinates of the corresponding target box in the infrared thermal image, pixel data of non-wall structure interference areas are removed from the enhanced binarized image of the exterior wall to obtain a clean binarized image of the exterior wall.

[0036] Furthermore, the single-frame infrared thermal image and visible light image of the building's exterior wall were acquired using the following method:

[0037] Based on the geographical location, orientation, and inspection date of the target building's exterior wall, the theoretical solar radiation intensity is calculated, and the inspection time window is determined in conjunction with the temperature field model describing the heat conduction law of the wall.

[0038] The drone, equipped with a dual-light gimbal for visible light and infrared thermal imaging, is controlled to collect image data of the wall along a preset path within the detection time window.

[0039] Furthermore, it also includes: calculating the overall wall hollowness rate based on a clean binary image of the exterior wall, and completing the risk assessment.

[0040] Furthermore, for exterior walls with tiled surfaces, the following are also included:

[0041] Using a template matching algorithm, a clean binary image of the exterior wall is matched with a standard tile template without hollow areas to filter out tile areas without hollow areas.

[0042] Based on the matching results, the hollow rate of individual tiles in the unmatched areas and the hollow rate of the entire wall surface are calculated to complete the risk assessment.

[0043] Secondly, a single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment is provided, comprising:

[0044] The image data acquisition module is used to acquire single-frame infrared thermal images and visible light images of the building's exterior walls that are collected synchronously.

[0045] The infrared thermal imaging sequence generation module is used to generate a virtual, time-continuous infrared thermal imaging sequence based on a single-frame infrared thermal image and the thermal property parameters of the wall material, using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall.

[0046] The virtual heat flow image sequence generation module is used to perform time-space frequency domain transformation and cylindrical harmonic function inversion on the infrared thermal imaging sequence, and then inversely transform it back to the time-space domain to generate a virtual heat flow image sequence.

[0047] The binarization module is used to select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall;

[0048] The binarization enhancement module is used to enhance the binarized image of the exterior wall through a discriminant function to obtain an enhanced binarized image of the exterior wall.

[0049] The interference identification module is used to process synchronously acquired visible light images using a deep learning-based target detection network to identify and locate non-wall structure interference objects.

[0050] The interference removal module is used to register and map the recognition results in the visible light image with the infrared thermal image, remove the areas of non-wall structure interference in the enhanced binarized image of the exterior wall, and obtain a clean binarized image of the exterior wall.

[0051] Thirdly, a single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment is provided, including:

[0052] Drones;

[0053] A dual-light gimbal for visible light and infrared thermal imaging, mounted on the UAV, is used to simultaneously acquire single-frame infrared thermal images and visible light images of the building's exterior walls;

[0054] The controller is communicatively connected to the UAV and the visible light and infrared thermal imaging dual-light gimbal, and is used to control the UAV and the visible light and infrared thermal imaging dual-light gimbal, and record the acquired image data in real time.

[0055] The processing module is used to acquire recorded image data from the controller and is configured to perform the single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment, as described above.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) This invention establishes a temperature field model (heat conduction control equation), fully considers the heat conduction characteristics of the wall material, the law of environmental temperature change and the influence of sunlight on the surface temperature of the wall, simulates the temperature distribution difference between normal walls and hollow walls under different environmental conditions, thereby compensating for the interference of environmental factors on the detection, avoiding misjudgment of hollow areas due to environmental fluctuations, and improving the detection accuracy of hollow walls in building exteriors.

[0058] (2) The present invention adopts a multimodal feature fusion method to combine the wall thermal features reflected by infrared thermal images with the object morphology features presented by visible light images. Through the complementary verification of the two modal information, non-wall structural interference objects and wall defects are clearly distinguished, thereby eliminating the interference of non-wall structural interference objects on the detection results and improving the detection accuracy of hollow building exterior walls.

[0059] (3) This invention utilizes transfer learning technology, which only requires fine-tuning the model with a small number of labeled samples of the target building to enable the model to quickly adapt to the characteristics of the target building. It can achieve accurate identification without relying on a large number of samples, effectively solving the problem of small sample learning.

[0060] (4) This invention forms a complete automated detection system through the synergistic effect of temperature field model, multimodal feature fusion method and transfer learning technology, that is, through the organic combination of physical model and data-driven method. This system covers the entire process from early data collection, mid-term defect identification to late risk assessment. It does not require much manual intervention, greatly improves detection accuracy and environmental adaptability, thereby promoting the large-scale application of infrared thermal imaging technology in the field of building safety inspection and providing strong technical support for the structural safety of high-rise buildings. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of a single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment, provided in an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram of the dual-light data registration result provided in an embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram of the statistical results of category detection accuracy of the target detection network provided in this embodiment of the invention;

[0065] Figure 4 This is a schematic diagram of an infrared thermal image and a binarization processing result provided in an embodiment of the present invention, wherein (a) is an infrared thermal image and (b) is a schematic diagram of the binarization processing result;

[0066] Figure 5 This is a template matching result diagram provided in an embodiment of the present invention;

[0067] Figure 6 This is a schematic diagram of the binarization result after hollowing out, provided in an embodiment of the present invention;

[0068] Figure 7 This is a schematic diagram of the calculation results of the void ratio in a region provided by an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0070] This invention addresses three key problems inherent in traditional detection technologies: First, traditional techniques are significantly affected by environmental temperature fluctuations and changes in sunlight, easily leading to deviations in detection results. Second, when dealing with hollow features in different building materials, the need for a large number of labeled samples to train the model, coupled with the difficulty of obtaining samples in actual engineering projects, results in poor adaptability to small-sample scenarios. Third, common non-wall objects in visible light images, such as air conditioner outdoor units and pipes, are often misidentified as wall defects, significantly reducing detection accuracy. Through targeted design, this invention proposes a single-frame infrared thermal imaging method and system for detecting hollow building exterior walls based on heat flow assessment. This method achieves high-precision identification of hollow defects and, based on accurate defect location, can quantitatively assess the potential safety risks posed by the defects, providing a reliable basis for subsequent operation and maintenance. The technical solution of this invention will be specifically described below with reference to specific embodiments.

[0071] This invention provides a single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment. This method combines temperature field modeling, transfer learning, and multimodal feature fusion for detecting hollow areas in high-rise building exterior walls. Its technical logic revolves around "accurately eliminating interference and improving recognition capabilities": This method uses single-frame heat flow estimation to enhance and evaluate the infrared thermal imaging detection results. Compared to the temperature data in the original infrared thermal imaging, the estimated heat flow data is more sensitive to changes in physical properties caused by factors such as hollowing and delamination. Simultaneously, visual information is added to lock onto the detection target, effectively solving problems such as tile color and structural interference encountered in practical solar thermal imaging detection. Utilizing transfer learning technology, the model first masters the common characteristics of non-wall structural interference objects using existing data, and then quickly adapts to new scenarios using a small number of target samples. Even with limited sample conditions, it can still accurately capture the characteristics of non-wall structural interference objects in the target building, achieving accurate recognition. Furthermore, the method integrates multi-source information, combining features from both infrared thermal images and visible light images. It can combine visible light images collected by drones with a target detection network to verify whether an object belongs to the wall structure, completely eliminating interference from non-wall objects. By integrating these technologies, automated detection and risk assessment of hollow defects can be achieved, eliminating the need for manual inspection and significantly reducing labor costs and operational risks, while improving detection efficiency and accuracy.

[0072] Specifically, such as Figure 1As shown, the method for detecting hollow areas in the exterior walls of a building includes the following steps:

[0073] S1: Constructing a temperature field model: The core of constructing a temperature field model is to accurately describe the heat conduction law of the wall (including hollow areas and non-hollow areas) through mathematical equations, and to simulate the temperature distribution differences under different environments based on this.

[0074] The temperature field model can be expressed by the heat conduction governing equation as follows:

[0075] ;

[0076] ;

[0077] Where T is the wall temperature; The thermal diffusivity; Represents the Laplace operator; Indicate boundary conditions; For the heat source term, specifically referring to the intensity of thermal radiation received by the wall surface, this value needs to be calculated in conjunction with the latitude and longitude of the wall: due to the differences in solar altitude angle, sunshine duration, and atmospheric transparency in different latitude and longitude regions, for example, the solar radiation at noon is stronger in low latitude regions and the solar radiation in winter is weaker in high latitude regions, it is necessary to use a solar radiation model associated with latitude and longitude (which is a known model and will not be elaborated here) to correct the influence of geographical location on the intensity of thermal radiation, so as to ensure that the calculation of Q fits the actual environment; This represents the heat source term at latitude x and longitude y at time t.

[0078] After establishing the equations, environmental parameters (temperature, orientation, wind speed, solar radiation intensity, etc.) need to be input for simulation calculations: ambient temperature directly affects the initial temperature baseline of the wall; wind speed changes the air heat conduction efficiency, thus affecting the heat exchange on the wall surface, and together with solar radiation intensity, determines the dynamic change of the heat source term Q, i.e. ,in, This represents the theoretical solar radiation intensity at the current moment and under the current orientation (calculated based on a solar radiation model). The convective heat transfer coefficient at the current wind speed. The current ambient temperature. This represents the current wall surface temperature. By substituting these parameters into the heat conduction control equation, the simulated temperature of each pixel within the wall surface region at different time points can be calculated.

[0079] S2: Calculate the theoretical solar radiation intensity based on the geographical location, orientation, and inspection date of the target building's exterior wall, and determine the inspection time window in conjunction with the temperature field model.

[0080] The generation of the detection window period requires consideration of three core factors: the latitude and longitude of the wall's location, the detection date, and the wall's orientation. The core objective is to avoid periods of drastic temperature fluctuations and strong solar interference to ensure stable detection data. Latitude and longitude determine the solar trajectory at the detection location. For example, in mid-latitude regions of the Northern Hemisphere, the sun rises in the northeast and sets in the northwest in summer, and in the southeast and sets in the southwest in winter. The detection date affects the solar altitude angle; at the same location, the solar altitude angle is higher in summer, resulting in concentrated solar radiation, while the opposite is true in winter. The wall's orientation directly determines the duration and intensity of sunlight it receives. For example, south-facing walls receive the strongest sunlight at noon, while north-facing walls receive weaker sunlight. By inputting these three factors into a solar radiation model and combining them with a temperature field model to analyze the temperature variation of the wall surface at different times, the detection window is selected based on periods of small temperature fluctuations and relatively stable solar radiation. For example, a north-facing wall experiences relatively small temperature fluctuations from 10:00 AM to 2:00 PM in spring and autumn, making it suitable as a detection window.

[0081] Specifically, the calculation of theoretical solar radiation intensity needs to be based on the above parameters: combining latitude and longitude to determine the solar declination angle, the solar hour angle corresponding to the detection date and the solar azimuth angle corresponding to the wall orientation, as well as the atmospheric attenuation coefficient of solar radiation (affected by cloud cover and air quality, which can be set based on the meteorological conditions of the detection date), and using the solar radiation model, the theoretical solar radiation intensity of the wall surface at different times is calculated.

[0082] S3: Control the drone equipped with a dual-light gimbal for visible light and infrared thermal imaging to collect image data of the wall according to a preset path within the detection time window, and record the spatial position and attitude parameters during the collection.

[0083] The dual-light gimbal on the drone needs to integrate both infrared thermal imaging and visible light imaging modules to ensure simultaneous acquisition of image data in both modalities during push-broom scanning. Path planning must be tailored to the structural characteristics of the building facade. For example, for the concave and convex facades of high-rise residential buildings, a zigzag or stepped path should be planned to ensure the drone can move close to the wall surface and achieve full-coverage, spot-point shooting. During spot-point shooting, the drone head must be directly facing the wall, and the angle between the drone head normal and the wall normal must not exceed ±15°. The core reason for this angle limitation is that if the angle is too large, the infrared thermal imaging lens will deviate in measuring the wall surface temperature due to oblique illumination. The thermal radiation received by the lens will be attenuated due to the incident angle, and the visible light image will also experience perspective distortion, which is detrimental to subsequent dual-light registration.

[0084] Recording the shooting distance is crucial for calculating the thermal imaging line resolution: thermal imaging line resolution reflects the smallest wall size that the lens can distinguish per unit distance. By recording the shooting distance in real time by the drone—that is, the vertical distance between the drone and the wall surface—and combining this with the inherent resolution of the infrared thermal imaging module (such as the number of pixels and focal length), the actual wall size corresponding to each pixel, i.e., the line resolution, can be calculated. For example, when the shooting distance is close, the line resolution value is small, allowing for the capture of more subtle wall defects; when the shooting distance is far, the line resolution value increases, requiring adjustments to the lens focal length or a reduction in flight altitude to ensure resolution accuracy and prevent small-area void defects from being overlooked due to low line resolution.

[0085] S4: Sequentially acquire single-frame infrared thermal images and visible light images of the building's exterior wall to sequentially detect hollow areas in each region of the building's exterior wall.

[0086] S5: Based on a single-frame infrared thermal image and the thermal properties of the wall material, a virtual, time-continuous infrared thermal imaging sequence is generated using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall.

[0087] Specifically, the infrared thermal imaging sequence is generated by the following method:

[0088] S51: Calculate the average thermal conductivity at the image scale based on the spatial resolution parameters, field of view, acquisition frequency, and average thermal conductivity of the wall in the infrared thermal image.

[0089] Image-scale average thermal conductivity It is expressed as follows:

[0090] ;

[0091] Image-scale average thermal conductivity The calculation transforms the thermal diffusivity α of the wall material into an image domain parameter adapted for infrared thermal image processing, providing a quantitative basis for subsequent heat flow simulation. The physical meaning and conversion logic of each parameter in the formula are as follows: S (spatial resolution parameter, unit m / pix) represents the actual wall area corresponding to each pixel. Dividing by S² converts the area unit (m²) in physical space into the pixel area unit (pix²) in the image domain, ensuring that the spatial dimension of the thermal conductivity coefficient matches the image pixel; σ (acquisition frequency, unit fram / s) represents the number of thermal image frames acquired per unit time. Dividing by σ converts the physical time unit (s) into the time unit (fram) in the image domain, achieving adaptation of the time dimension.

[0092] Calculated using this formula (Unit: pixel² / frame) essentially reflects the efficiency of heat transfer per unit time (per frame) through a unit pixel area in the image domain. This parameter needs to be precisely matched with the actual thermal diffusivity α of the wall material. For example, the α value of a ceramic tile wall is higher than that of a stone paint wall, and its corresponding... The value will also be higher, which means that in the infrared thermal imaging sequence, the temperature change of the tile wall is transmitted at the pixel level faster. This parameter will be directly used for the generation of subsequent infrared thermal imaging sequences and heat flow inversion calculations.

[0093] S52: Using the acquired single-frame infrared thermal image as input, combined with the Green's function of the heat conduction control equation describing the heat conduction law of the wall and the image-scale average heat conduction coefficient, a virtual infrared thermal imaging sequence of length N, which is continuous and equally spaced in the time dimension, is generated.

[0094] The core of this step is the Green's function based on a single-frame infrared thermal image and the heat conduction control equation. This method extends and generates a continuous sequence of infrared thermal images, compensating for potential data gaps at specific time points in actual detection and ensuring the continuous simulation of the heat conduction process. The Green's function, used in the heat conduction control equations to solve for temperature distribution under non-homogeneous boundary conditions, is physically defined as "the temperature response of a unit heat source at a specific point in time." The Green's function can be used to calculate the temperature diffusion of a single-frame infrared thermal image at different time points, thus extending the simulation into a sequence of length N.

[0095] During the calculation process, the image-scale average thermal conductivity coefficient obtained in step S51 needs to be used. Substituting the Green's function ensures that the temperature changes in the image sequence conform to the heat conduction laws in the image domain. "Equivalent completion" refers to the following: if the actual acquired single-frame infrared thermal image corresponds to a specific time point, the temperature changes before and after that time point need to be simulated based on the temperature field model to complete the missing time node images, so that the generated image sequence can fully reflect the dynamic temperature changes of the wall within the detection window period. For example, if only a single-frame image of the middle 30 minutes is acquired within one hour from the start to the end of the detection, the temperature changes of the first 30 minutes and the last 30 minutes can be simulated using the Green's function to complete the image sequence into N consecutive frames, providing complete spatiotemporal data for subsequent frequency domain transformation.

[0096] S6: Perform time-space frequency domain transformation and cylindrical harmonic function inversion on the infrared thermal imaging sequence, and then inversely transform it back to the time-space domain to generate a virtual heat flow image sequence corresponding to a single frame of infrared thermal image.

[0097] Specifically, the virtual heat flow image sequence is generated by the following method:

[0098] S61: Perform Gaussian filtering on the infrared thermal imaging sequence.

[0099] First, the generated infrared thermal image sequence is processed by Gaussian filtering with a window size of 3×3. The core function of Gaussian filtering is to smooth image noise (such as pixel temperature fluctuations caused by slight drone shaking, and noise generated by environmental electromagnetic interference) while preserving the detailed features of the wall temperature distribution. The choice of 3×3 window size needs to balance the noise reduction effect and the preservation of details: if the window is too small, the noise reduction will be incomplete, and if it is too large, the subtle temperature differences in the hollow areas will be blurred. Therefore, this size can achieve the optimal balance between the two.

[0100] S62: Perform Fourier-Hankel transform on the filtered infrared thermal image sequence to transform the image from the spatiotemporal domain to the frequency domain.

[0101] Fourier transform is used to convert the time-domain signal (temperature change over time) of an image into a frequency-domain signal (temperature fluctuation components of different frequencies), while Hankel transform is optimized for radially symmetrical signals (such as temperature diffusion on a wall surface centered on a certain point), and is suitable for processing the heat conduction characteristics of circular or near-circular hollow areas.

[0102] S63: Adjust the ratio of the spatiotemporal frequency domain transformation axis according to the length N of the infrared thermal imaging sequence and the acquisition time, and recalibrate the time and frequency axis.

[0103] In the Fourier-Hankel transform process, it is necessary to define the time-frequency domain transformation axis ω (corresponding to the frequency of temperature change) and the axial spatial frequency domain transformation axis ξ (corresponding to the spatial frequency of temperature change), and adjust the ratio of the two according to the sequence length N (i.e., the number of time nodes) and the acquisition time τ: for example, if the sequence contains N time nodes, the acquisition time is from τ1 to τ2. N It is necessary to map τ1 to the time 00:00 through proportional adjustment to unify the time base of the entire graph sequence, which will facilitate subsequent analysis of the temperature change patterns between different graph sequences and eliminate the interference of acquisition time differences on frequency domain analysis.

[0104] S64: Using cylindrical harmonic functions as the core, the infrared thermal imaging sequence transformed to the frequency domain is inverted.

[0105] The inversion process requires the use of column harmonic functions. The transformation algorithm is based on this, where, It is the time-frequency domain transform axis after adjustment of the scale. This represents an imaginary number. The columnar harmonic function is introduced because the heat conduction process in the wall exhibits a columnar harmonic function distribution characteristic in the radial direction (such as the heat diffusion in the hollow area resembling circular wave diffusion). It can accurately describe the radial diffusion law of heat flow in the frequency domain. By inverting the frequency domain sequence after Fourier-Hankel transform using this core, the heat conduction attenuation error in the frequency domain signal can be corrected, and the true heat flow distribution characteristics can be restored.

[0106] S65: The frequency-domain infrared thermal imaging sequence after inversion processing is restored to the spatiotemporal domain through inverse Fourier-Hankel transform to obtain a virtual heat flow image sequence characterizing the heat flow distribution. The core value of the virtual heat flow image is to transform the heat flow changes inside the wall, which cannot be directly observed, into intuitive changes in image pixel intensity. Because of the presence of an air layer inside, the heat conduction efficiency of hollow areas is lower than that of normal walls. In the virtual heat flow image, these areas will show abnormal heat flow intensity, such as pixel brightness lower than normal areas, thus providing clear heat flow characteristics for subsequent hollow area identification.

[0107] S7: Select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall.

[0108] During binarization, a reasonable pixel threshold needs to be set for the virtual heat flow image at 00:00 (at which time the time base is unified by step S6, the heat flow distribution is relatively stable, and interference is minimal). When the pixel value of a certain pixel is not less than the pixel threshold, the pixel is determined to be a "normal wall pixel" and the pixel value of the pixel is set to 255 (white). When the pixel value of a certain pixel is less than the pixel threshold, the pixel is determined to be a "suspected hollow pixel" and the pixel value of the pixel is set to 0 (black). This achieves the binarization of the virtual heat flow image, obtains the binarized image of the exterior wall, and initially distinguishes the potential hollow area from the normal wall area.

[0109] S8: Enhance the binarized image of the exterior wall by using a discriminant function to obtain an enhanced binarized image of the exterior wall.

[0110] In this embodiment, the sigmoid function is chosen as the discriminant function because it possesses smooth non-linear mapping characteristics, which can avoid the misjudgment problem of traditional fixed threshold binarization (such as the Otsu algorithm) in temperature fluctuation scenarios. The output value of the sigmoid function changes continuously between 0 and 1. The sigmoid discriminant function maps each pixel in the binarized image of the exterior wall to a smooth hollow probability value of 0 to 1: pixels that conform to the characteristics of a normal wall have an output probability close to 1; pixels in hollow areas have an output probability close to 0.

[0111] In the process of enhancing the binarized image of the exterior wall, a reasonable probability threshold (such as 0.5) needs to be set for further binarization. Pixels with a hollow probability value not lower than the probability threshold are judged as "normal wall pixels" and assigned a value of 255 (white); pixels with a probability value lower than the probability threshold are judged as "suspected hollow pixels" and assigned a value of 0 (black), thus obtaining the enhanced binarized image of the exterior wall. Through this enhancement method, more reliable hollow areas can be screened based on probability, which is more accurate than the "one-size-fits-all" approach of direct binarization. It clearly distinguishes between normal areas and suspected hollow areas, making the boundaries more accurate and reducing misjudgments. Essentially, it is an optimization and purification of the binarized image, which can lay the foundation for subsequent template matching and hollow rate calculation.

[0112] S9: A deep learning-based target detection network is used to process synchronously acquired visible light images to identify and locate non-wall structure interference objects.

[0113] The target detection network can be selected as needed, such as the YOLO series models. In this embodiment, the improved pruned YOLOv9 is selected, whose core optimization directions are "lightweighting" and "enhanced generalization ability". The pruning technique analyzes the contribution of each convolutional kernel in the network and removes redundant convolutional kernels (such as kernels that contribute very little to distinguishing between walls and non-walls). While reducing the number of network parameters and computational load, it retains the core feature extraction capability. This optimization enables the algorithm to run efficiently on the UAV embedded platform, avoids detection delay caused by excessive computation, and improves the detection speed in complex backgrounds, such as when facing dense air conditioner outdoor units and pipes, it can still process images quickly.

[0114] During the model training phase, a transfer learning strategy is employed: first, a basic YOLOv9 model is trained on a general object detection dataset (such as the COCO dataset) to master common object detection features (such as object edges and shape features); then, the basic model is transferred to a dedicated dataset for detecting non-wall structure interference (including samples of doors and windows, fences, tile detachment, repair areas, and roofs) for fine-tuning, resulting in an object detection network. By adjusting network parameters, the model is adapted to the features of wall scenes (such as the rectangular outlines of doors and windows, and texture differences in tile repair areas), enhancing its generalization ability to non-wall areas in complex backgrounds. For example, even when faced with irregular fence structures in old buildings, the model can still accurately identify and locate them, ultimately outputting detection results with bounding boxes. Each bounding box corresponds to a class of non-wall structure interference areas, and its category is labeled (such as "air conditioner outdoor unit" or "door and window"), providing accurate non-wall area location information for subsequent two-light registration. Finally, the trained YOLOv9-based object detection network is pruned to achieve model lightweighting.

[0115] S10: Register and map the recognition results in the visible light image with the infrared thermal image, remove the areas in the enhanced binarized image of the exterior wall that correspond to non-wall structure interference objects, and obtain a clean binarized image of the exterior wall.

[0116] Specifically, the core of dual-light registration is to establish a coordinate mapping relationship between the visible light image (including the target bounding box in non-wall areas) and the infrared thermal image, achieving spatial alignment of the two modalities. In practice, common feature points (such as the corners of walls, the edges of doors and windows, and the joints of tiles) need to be selected in both images first. The coordinate transformation matrix is ​​then calculated using a feature point matching algorithm (such as SIFT or ORB). This matrix can convert the target bounding box coordinates in the visible light image into the corresponding target bounding box coordinates in the infrared thermal image, achieving accurate superposition of the target bounding box in the infrared thermal image.

[0117] After registration, based on the target bounding box coordinates in the converted infrared thermal image, pixel data from non-wall structural interference areas are removed from the enhanced binarized image of the exterior wall. For example, all infrared pixels within the "air conditioner outdoor unit" target bounding box are removed, retaining only the infrared pixels of the wall outside the target bounding box. The purpose of this step is to eliminate thermal feature interference from non-wall areas—non-wall structural interference (such as the high temperature of an air conditioner outdoor unit during operation or the low temperature of metal pipes) may appear as hollow-like temperature anomalies in the infrared thermal image. By removing these areas, a pure exterior wall binarized image is obtained, ensuring that subsequent defect identification targets only the wall area, thus improving detection accuracy. Since the infrared thermal image and the enhanced binarized image of the exterior wall correspond in size and pixels, the target bounding box coordinates in the converted infrared thermal image correspond to the location bounding boxes of non-wall structural interference in the enhanced binarized image of the exterior wall.

[0118] In some preferred embodiments, the method further includes:

[0119] S11: Calculate the regional void rate based on the clean binary image of the exterior wall. Regional void rate = (number of void pixels / number of pixels in the area) × 100%, and complete the risk assessment.

[0120] In other embodiments, such as for exterior walls with tiles, the method further includes:

[0121] S12: Using a template matching algorithm, the clean binary image of the exterior wall is matched with a standard tile template without hollow areas to filter out the tile areas without hollow areas.

[0122] The core of template matching is to use a standard tile template to filter out areas of tiles without hollow spots in a clean binary image of the exterior wall, providing a reference for subsequent location of hollow areas. First, the optimal standard tile template needs to be selected: in a clean binary image of the exterior wall, tiles without hollow spots, with uniform texture and stable heat flow characteristics are selected as the standard template. This template must be representative (e.g., the size and material must be the same as the tile to be detected) to avoid matching errors caused by template differences.

[0123] Then, the range of black and white pixel values ​​for the standard template was determined: tiles without hollow spots appear as pure white (pixel value 255) in the binarized image, but due to slight noise, there may be a small number of gray pixels. Therefore, a reasonable range of values ​​(such as pixel value 240-255) needs to be set to ensure that tiles without hollow spots are accurately included in the matching range.

[0124] The matching process employs a batch processing algorithm (such as the normalized cross-correlation matching algorithm): a standard template is slid across a clean binary image of the exterior wall, and the similarity between the template and each region of the image is calculated. Regions with a similarity higher than a set threshold (e.g., 0.9) are identified as "regions without hollow areas," and the tile positions in these regions are marked. Regions with a similarity lower than the threshold are tentatively designated as "suspected hollow areas." Through batch matching, most regions without hollow areas can be quickly filtered out, narrowing the scope of subsequent hollow rate calculations and improving detection efficiency.

[0125] S13: Based on the matching results, calculate the hollow rate of individual tiles and the hollow rate of the area in the unmatched area to complete the risk assessment.

[0126] For a single tile, in a clean binary image of the exterior wall, the pixel values ​​of tiles without hollow areas are all within the range of the standard template (e.g., 240-255). Tiles with hollow areas will have some or all pixel values ​​below this range (e.g., 0-239, all black or alternating black and white). By counting the number of pixels below the range within a single tile (i.e., the number of hollow pixels), dividing this number by the total number of pixels in the tile, and then multiplying by 100%, we can obtain the hollow rate of a single tile. The formula can be expressed as: Hollow rate of a single tile = (Number of hollow pixels / Total number of pixels in a single tile) × 100%.

[0127] To determine the area hollowness rate, the calculation area (such as a wall or a floor) must first be determined. The total number of pixels of all tiles in the area (i.e., S_total, the pixel value corresponding to the total area) must be counted. Then, the total number of hollow pixels of all suspected hollow tiles in the area (i.e., S_void, the total area minus the total number of pixels in the area without hollow areas) must be counted. The area hollowness rate is then calculated using the formula S_void / S_total × 100%.

[0128] The advantage of this calculation method is that it is based on pixel-level accurate statistics, which can avoid the subjective error of manual measurement. At the same time, the single tile hollow rate can be used to judge the risk of a single tile falling off (e.g., a tile with a hollow rate of more than 80% has an extremely high risk of falling off), and the area hollow rate can be used to assess the safety status of the entire wall (e.g., an area hollow rate of more than 30% requires overall inspection), providing a quantitative basis for subsequent operation and maintenance decisions.

[0129] The single-frame infrared thermal imaging method for detecting hollow exterior walls based on heat flow assessment provided in the above embodiments is supported by three core aspects: First, by establishing an accurate temperature field model, fully considering the thermal conductivity characteristics of the wall material, the law of environmental temperature change, and the influence of sunlight on the wall surface temperature, the model simulates the temperature distribution differences between normal walls and hollow walls under different environmental conditions, thereby compensating for the interference caused by environmental factors and avoiding misjudgment of hollow areas due to environmental fluctuations, thus improving the accuracy of hollow exterior wall detection. Second, by employing a multimodal feature fusion method, the thermal characteristics of the wall reflected by the infrared thermal image are combined with the object morphology characteristics presented by the visible light image. Through the complementary verification of the two modal information, non-wall structural interference objects and wall defects are clearly distinguished, thereby eliminating the interference of non-wall structural interference objects on the detection results and improving the accuracy of hollow exterior wall detection. Third, by utilizing transfer learning technology, the model can be quickly adapted to the characteristics of the target building by only fine-tuning it with a small number of labeled samples of the target building, achieving accurate identification without relying on a large number of samples, effectively solving the problem of small sample learning. By leveraging the synergistic effects of temperature field models, multimodal feature fusion methods, and transfer learning techniques—that is, by organically combining physical models with data-driven methods—a complete automated detection system has been formed. This system covers the entire process from initial data acquisition and mid-term defect identification to post-risk assessment, requiring minimal human intervention and significantly improving detection accuracy and environmental adaptability. This promotes the large-scale application of infrared thermal imaging technology in the field of building safety inspection, providing strong technical support for the structural safety of high-rise buildings. It can be widely applied to various specific scenarios such as urban building safety monitoring, old community renovation assessment, and historical building protection.

[0130] This invention also provides a single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment, in order to realize the single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment described in the above embodiments. The system includes a drone, a dual-light gimbal for visible light and infrared thermal imaging, a controller, and a processing module.

[0131] Drones, serving as mobile platforms for data acquisition, primarily function to carry infrared thermal imaging and other detection equipment to systematically scan building facades according to a pre-defined detection plan. Considering the complex structure of high-rise building facades and the varying detection requirements in different areas, drones must possess stable flight control capabilities to ensure stable flight even under slight wind or other external interference in high-altitude environments, preventing data acquisition interruptions or deviations due to flight swaying. Simultaneously, they must have precise positioning capabilities, accurately correlating each set of collected data with specific areas of the wall to ensure accurate tracing of defects during subsequent analysis. During the scanning process, the drone must fly along a pre-planned path that conforms to the building facade shape, ensuring complete coverage of the entire facade without omissions or duplications, while maintaining consistent data location to allow for effective comparison of data collected at different times and in different batches.

[0132] Visible and infrared thermal imaging dual-light gimbal: Mounted on the aforementioned drone, it is used to simultaneously acquire single-frame infrared thermal images and visible light images of the building's exterior walls. Since the temperature difference between hollow areas and normal walls is subtle, the gimbal needs sufficient temperature resolution to capture these minute temperature changes and clearly present the temperature difference between the two. Simultaneously, it needs appropriate spatial resolution to ensure that the infrared thermal images clearly show the details of the wall surface, preventing small-area hollow defects from being overlooked due to insufficient resolution. Furthermore, drones inevitably experience attitude swaying during flight. If the gimbal shifts with the drone's movement, it will cause image blurring, affecting the detection effect. Therefore, the gimbal needs attitude stabilization capabilities, using its own adjustment mechanism to counteract the effects of drone swaying, ensuring that clear and stable images are acquired throughout the data acquisition process, providing a high-quality data foundation for subsequent defect identification.

[0133] The controller communicates with the UAV and the visible light and infrared thermal imaging dual-light gimbal. On one hand, it controls the UAV and the dual-light gimbal to ensure synchronized operation of all components during the detection process. For example, when the UAV flies to a designated location, the gimbal can adjust its angle and start acquisition in time, avoiding any disconnect between components. On the other hand, it records the acquired image data in real time, including acquisition time, distance between the UAV and the wall, acquisition height, and gimbal orientation angle. This information corresponds one-to-one with the image data. These recorded parameters are crucial for subsequent temperature field modeling and void identification: the acquisition time can be correlated with the ambient temperature and sunlight conditions, providing a basis for compensating for environmental interference in the temperature field model; distance, height, and gimbal orientation angle help calibrate the correspondence between the thermal image and the actual area of ​​the wall, clarifying the wall position corresponding to each pixel in the thermal image, providing necessary spatial and attitude information for data analysis, and ensuring the accuracy of defect location.

[0134] The processing module is used to acquire recorded image data from the controller and is configured to execute the single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment as described in the foregoing embodiments.

[0135] This invention also provides a single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment, characterized in that it includes:

[0136] The image data acquisition module is used to acquire single-frame infrared thermal images and visible light images of the building's exterior walls that are collected synchronously.

[0137] The infrared thermal imaging sequence generation module is used to generate a virtual, time-continuous infrared thermal imaging sequence based on a single-frame infrared thermal image and the thermal property parameters of the wall material, using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall.

[0138] The virtual heat flow image sequence generation module is used to perform time-space frequency domain transformation and cylindrical harmonic function inversion on the infrared thermal imaging sequence, and then inversely transform it back to the time-space domain to generate a virtual heat flow image sequence.

[0139] The binarization module is used to select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall;

[0140] The binarization enhancement module is used to enhance the binarized image of the exterior wall through a discriminant function to obtain an enhanced binarized image of the exterior wall.

[0141] The interference identification module is used to process synchronously acquired visible light images using a deep learning-based target detection network to identify and locate non-wall structure interference objects.

[0142] The interference removal module is used to register and map the recognition results in the visible light image with the infrared thermal image, remove the areas of non-wall structure interference in the enhanced binarized image of the exterior wall, and obtain a clean binarized image of the exterior wall.

[0143] It should be understood that the functional unit modules in the various embodiments of the present invention can be concentrated in one processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated into one unit module, and can be implemented in hardware or software.

[0144] The technical solution of the present invention will be further explained below with reference to a specific example.

[0145] The data used is a building and garden facade dataset from a location in Changsha, Hunan Province. Data was collected via a drone using dual-light technology. Due to differences in the field of view of the dual-light dataset carried by the drone, focus learning and alignment were employed to achieve registration of the dual-light datasets. The results are as follows: Figure 2 As shown.

[0146] The registered visible light images were used to train and test a YOLOv9-based object detection network. The training and testing datasets were in a 9:1 ratio, containing 14 categories, including fences, alcoves, clothes racks, air conditioners, pillars, and flaws. Detection on the visible light dataset prepared the groundwork for removing decorative backgrounds from subsequent thermal imaging images, improving the accuracy of template matching. The statistical results of category detection accuracy are shown below. Figure 3 As shown.

[0147] The infrared thermal image was binarized for testing, such as... Figure 4 As shown, it is clearly observable that tiles without hollow areas appear as a complete white region, while areas with hollow areas are represented as black because their heat flow information differs from that of healthy, intact tiles. This image feature allows for the extraction of a standard, undamaged template. A rigorous threshold matching mechanism is then applied, using OpenCV's template matching algorithm to match the image with the template. Areas that do not match are considered problematic. Figure 5 As shown.

[0148] Depend on Figure 5 As can be seen, the decoration removal step is affected by the accuracy of image registration and the detection accuracy of the target detection model. In some non-detection areas such as windows, fences, and pillars, there will be incorrect template matching results. By adjusting the coordinate mapping parameters, the decorations are "hollowed out" from the original image, improving the template matching accuracy. Figure 6 The image shown is a schematic diagram of the result after the binarization process has been hollowed out.

[0149] Based on the existing binarization results, calculate the hollow rate of individual tiles and regional tiles, and save the results in the root directory, recording the time of hollow detection, dataset name, and hollow rate results. Figure 7 The diagram shows the calculation results of the hollow tile rate in a certain area, which is 62.3%. Testing revealed that the accuracy of hollow tile detection and template matching is high in building exterior walls with smaller dimensions and high tile uniformity.

[0150] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0151] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0152] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting hollow areas in building exterior walls using single-frame infrared thermal imaging based on heat flow assessment, characterized in that, Includes the following steps: Acquire single-frame infrared thermal images and visible light images of the building's exterior walls simultaneously; Based on a single-frame infrared thermal image and the thermal properties of the wall material, a virtual, time-continuous infrared thermal imaging sequence is generated using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall. The infrared thermal imaging sequence is subjected to time-space frequency domain transformation and cylindrical harmonic function inversion, and then inversely transformed back to the time-space domain to generate a virtual heat flow image sequence; the virtual heat flow image sequence is generated by the following method: Gaussian filtering is applied to the infrared thermal imaging sequence; Perform Fourier-Hankel transform on the filtered infrared thermal image sequence to transform the image from the spatiotemporal domain to the frequency domain; Adjust the ratio of the spatiotemporal frequency domain transformation axis according to the infrared thermal imaging sequence length N and the acquisition time, and recalibrate the time and frequency axis; Using cylindrical harmonic functions as the core, the infrared thermal imaging sequence transformed to the frequency domain is inverted. The frequency domain infrared thermal imaging sequence after inversion processing is restored to the spatiotemporal domain through inverse transformation to obtain a virtual heat flow image sequence characterizing the heat flow distribution. Select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall; The binarized image of the exterior wall is enhanced by using a discriminant function to obtain an enhanced binarized image of the exterior wall, specifically including: Enhancement processing based on the sigmoid discriminant function is performed on the binarized image of the exterior wall. The sigmoid discriminant function maps each pixel in the binarized image of the exterior wall to a hollow probability value of 0 to 1. Then, by comparing with a set probability threshold, binarization is performed again to obtain the enhanced binarized image of the exterior wall. A deep learning-based target detection network is used to process synchronously acquired visible light images to identify and locate non-wall structure interference objects; The recognition results in the visible light image are registered and mapped with the infrared thermal image. The regions corresponding to non-wall structure interference objects in the enhanced binarized image of the exterior wall are removed to obtain a pure binarized image of the exterior wall.

2. The method for detecting hollow areas in building exterior walls based on single-frame infrared thermal imaging with heat flow assessment according to claim 1, characterized in that, The infrared thermal image sequence is generated using the following method: The average thermal conductivity at the image scale is calculated based on the spatial resolution parameters of the infrared thermal image, the acquisition frequency, and the average thermal conductivity of the wall. Using the acquired single-frame infrared thermal image as input, and combining the Green's function of the heat conduction control equation describing the heat conduction law of the wall with the image-scale average heat conduction coefficient, a virtual infrared thermal imaging sequence of length N, which is continuous and equally spaced in the time dimension, is generated.

3. The method for detecting hollow areas in building exterior walls based on single-frame infrared thermal imaging with heat flow assessment according to claim 1, characterized in that, The deep learning-based object detection network is obtained through the following method: First, train the object detection network on a general object detection dataset to obtain the basic model; The base model was transferred to a non-wall structure interference detection dataset for fine-tuning to obtain the final deep learning-based object detection network.

4. The single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment according to claim 1, characterized in that, The clean binarized image of the exterior wall is obtained through the following method: Select common feature points from visible light images and infrared thermal images; The coordinate transformation matrix is ​​calculated by the feature point matching algorithm. This coordinate transformation matrix is ​​then used to convert the coordinates of the non-wall structure interference target boxes identified from the visible light image into the corresponding target box coordinates in the infrared thermal image. Based on the coordinates of the corresponding target box in the infrared thermal image, pixel data of non-wall structure interference areas are removed from the enhanced binarized image of the exterior wall to obtain a clean binarized image of the exterior wall.

5. The method for detecting hollow areas in building exterior walls using single-frame infrared thermal imaging based on heat flow assessment according to any one of claims 1 to 4, characterized in that, The single-frame infrared thermal image and visible light image of the building's exterior wall were acquired using the following method: Based on the geographical location, orientation, and inspection date of the target building's exterior wall, the theoretical solar radiation intensity is calculated, and the inspection time window is determined in conjunction with the temperature field model describing the heat conduction law of the wall. The drone, equipped with a dual-light gimbal for visible light and infrared thermal imaging, is controlled to collect image data of the wall along a preset path within the detection time window.

6. The method for detecting hollow areas in building exterior walls using single-frame infrared thermal imaging based on heat flow assessment according to claim 1, characterized in that, For exterior walls with tiles, the following are also included: Using a template matching algorithm, a clean binary image of the exterior wall is matched with a standard tile template without hollow areas to filter out tile areas without hollow areas. Based on the matching results, the hollow rate of individual tiles in the unmatched areas and the hollow rate of the entire wall surface are calculated to complete the risk assessment.

7. A single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment, characterized in that, include: The image data acquisition module is used to acquire single-frame infrared thermal images and visible light images of the building's exterior walls that are collected synchronously. The infrared thermal imaging sequence generation module is used to generate a virtual, time-continuous infrared thermal imaging sequence based on a single-frame infrared thermal image and the thermal property parameters of the wall material, using the Green's function of the heat conduction control equation that describes the heat conduction law of the wall. A virtual heat flow image sequence generation module is used to perform time-space frequency domain transformation and cylindrical harmonic function inversion processing on infrared thermal imaging sequences, and then inversely transform them back to the time-space domain to generate a virtual heat flow image sequence; the virtual heat flow image sequence is generated by the following method: Gaussian filtering is applied to the infrared thermal imaging sequence; Perform Fourier-Hankel transform on the filtered infrared thermal image sequence to transform the image from the spatiotemporal domain to the frequency domain; Adjust the ratio of the spatiotemporal frequency domain transformation axis according to the infrared thermal imaging sequence length N and the acquisition time, and recalibrate the time and frequency axis; Using cylindrical harmonic functions as the core, the infrared thermal imaging sequence transformed to the frequency domain is inverted. The frequency domain infrared thermal imaging sequence after inversion processing is restored to the spatiotemporal domain through inverse transformation to obtain a virtual heat flow image sequence characterizing the heat flow distribution. The binarization module is used to select a frame of virtual heat flow image and perform binarization processing to obtain a binarized image of the exterior wall; The binarization enhancement module is used to enhance the binarized image of the exterior wall using a discriminant function, resulting in an enhanced binarized image of the exterior wall. Specifically, it includes: Enhancement processing based on the sigmoid discriminant function is performed on the binarized image of the exterior wall. The sigmoid discriminant function maps each pixel in the binarized image of the exterior wall to a hollow probability value of 0 to 1. Then, by comparing with a set probability threshold, binarization is performed again to obtain the enhanced binarized image of the exterior wall. The interference identification module is used to process synchronously acquired visible light images using a deep learning-based target detection network to identify and locate non-wall structure interference objects. The interference removal module is used to register and map the recognition results in the visible light image with the infrared thermal image, remove the areas of non-wall structure interference in the enhanced binarized image of the exterior wall, and obtain a clean binarized image of the exterior wall.

8. A single-frame infrared thermal imaging system for detecting hollow areas in building exterior walls based on heat flow assessment, characterized in that, include: Drones; A dual-light gimbal for visible light and infrared thermal imaging, mounted on the UAV, is used to simultaneously acquire single-frame infrared thermal images and visible light images of the building's exterior walls; The controller is communicatively connected to the UAV and the visible light and infrared thermal imaging dual-light gimbal, and is used to control the UAV and the visible light and infrared thermal imaging dual-light gimbal, and record the acquired image data in real time. The processing module is used to acquire recorded image data from the controller and is configured to perform the single-frame infrared thermal imaging method for detecting hollow areas in building exterior walls based on heat flow assessment as described in any one of claims 1 to 6.

Images