An unmanned aerial vehicle inspection device and method for hidden dangers of a high-rise building outer wall

Abstract

This invention relates to the fields of image recognition and building inspection technology, specifically to a drone inspection device and method for detecting hidden dangers in the exterior walls of high-rise buildings. The method acquires visible light and infrared images of the target exterior wall and aligns them to generate a temperature map; it segments the visible light image to extract brick objects, assigning temperature, texture, and neighbor relationship attributes to each brick object; it calculates a temperature difference index based on the temperature and neighbor relationship of each brick object, and uses this to select candidate bricks; it constructs a local thermal network centered on the candidate bricks, determines the heat flux based on the texture and temperature differences of each edge in the network, calculates the heat leakage of the candidate bricks using a maximum flow algorithm, and determines the isolation degree based on the temperature difference index and heat leakage; it performs dual-threshold screening on the candidate bricks based on the isolation degree, outputting a list of bricks diagnosed as hollow and a disease distribution map, thereby reducing the false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of image recognition and building inspection technology, specifically to a drone inspection device and method for detecting hidden dangers in the exterior walls of high-rise buildings. Background Technology

[0002] Exterior wall cladding tiles on high-rise buildings are prone to structural hazards such as hollowing and detachment during long-term service. Traditional manual inspections are inefficient and dangerous, making the use of drones equipped with visible light and infrared equipment for exterior wall inspections a mainstream development trend. Hollow areas, due to the presence of an internal air barrier, will exhibit abnormal temperature characteristics in infrared imaging, which can serve as an important basis for hazard identification.

[0003] Currently, most existing inspection methods directly perform global threshold segmentation on visible light or infrared images. This method has the following obvious limitations: First, heat is conducted laterally within the wall surface, and heat from hollow areas easily diffuses to the surrounding area, forming a blurred thermal halo. If the judgment method relies solely on the temperature difference threshold, it will be difficult to accurately define the boundary of the hollow area, and normal areas affected by heat diffusion are easily misjudged as potential hazard areas, resulting in insufficient detection geometric accuracy. Second, interference factors such as external wall shadows, water stains, and repair marks can also form abnormal temperature rise areas in infrared images. Existing methods cannot distinguish between real hollow areas and non-structural interference based on heat transfer characteristics, resulting in a high false alarm rate. The detection results are difficult to accurately correspond to individual decorative bricks, which cannot meet the needs of refined engineering applications. Summary of the Invention

[0004] To address the technical problems of existing UAV infrared detection methods, such as the difficulty in overcoming boundary blurring caused by thermal diffusion and the inability to effectively distinguish between structural voids and non-structural thermal interference, resulting in a high false alarm rate, this invention provides a UAV inspection device and method for detecting hidden dangers in the exterior walls of high-rise buildings. The specific technical solution adopted is as follows: This invention proposes a drone inspection method for potential hazards in the exterior walls of high-rise buildings, the method comprising: Simultaneously acquire visible light and infrared images of the target's exterior wall; perform dual-light alignment on the visible light and infrared images to generate a temperature mapping map corresponding to the pixel level of the visible light image; segment the visible light image into bricks to extract multiple brick objects; assign a brick temperature value to each brick object based on the temperature mapping map; assign a texture value to each brick object based on the visible light image; determine the neighbor objects of each brick object based on the spatial relationship between them, and form a neighbor table. Based on the brick temperature of each brick object and its neighbor table, determine the temperature difference index of each brick object and screen out candidate bricks; For each candidate brick, a local thermal network is constructed with it as the center; the heat flux of each edge is determined based on the difference in texture value and temperature of the bricks at both ends of each edge in the local thermal network; based on the heat flux, the maximum flow algorithm is executed on the local thermal network to determine the heat leakage from the candidate brick to the network boundary; the isolation degree is determined based on the temperature difference index and heat leakage of the candidate brick. Based on the isolation degree of candidate bricks, a dual threshold screening is performed using statistical threshold and absolute lower limit value, and a list of bricks diagnosed with hollowness and a disease distribution map are output to identify potential hollowness hazards in exterior walls.

[0005] Furthermore, the visible light image and the infrared image are acquired synchronously under preset constraints; the preset constraints include: the UAV hovers or moves vertically in front of the target exterior wall, and the shooting distance is greater than the preset micro-undulation thickness of the target exterior wall cladding bricks, so that the target exterior wall appears as a two-dimensional unfolded plane on the imaging plane; The temperature mapping generation process includes: Obtain the calibration parameters of the visible light camera and the infrared thermal imager; Under preset constraints, the homography matrix between the visible light image and the infrared image is calculated based on the calibration parameters; Using the pixel coordinate system of the visible light image as a reference, the infrared image is transformed by the homography matrix, and the transformed infrared image is resampled by the bilinear interpolation method to generate a temperature mapping map with the same resolution and coordinate system as the visible light image.

[0006] Furthermore, the process of assigning the brick temperature and texture values ​​to each brick object includes: For each brick object, obtain the pixel mask of the brick object in the visible light image. The pixel mask contains the coordinate positions of all pixels covered by the brick object. Based on pixel masks, index all pixels at the same coordinate position in the temperature map; based on pixel masks, index all pixels at the same coordinate position in the visible light image. Read the temperature value and grayscale value corresponding to each indexed pixel; Calculate the arithmetic mean of the temperature values ​​of all read pixels as the brick temperature of the brick object; calculate the arithmetic mean of the grayscale values ​​of all read pixels as the texture value of the brick object.

[0007] Furthermore, the step of determining the neighbor objects of each brick object based on the spatial positional relationship between each brick object and forming a neighbor table includes: For each brick object, identify its boundary pixels. A boundary pixel is a pixel whose any neighboring pixel does not belong to the brick object. The boundary pixels of the brick object are paired with the boundary pixels of every other brick object to form boundary pixel pairs; the Euclidean distance between each pair of boundary pixel pairs is calculated as the interval distance; the minimum interval distance between the brick object and the corresponding other brick objects is extracted. If the minimum interval distance is less than the preset interval threshold, it is determined that the brick object is a neighbor to other corresponding brick objects; all other brick objects that are neighbors to the brick object are recorded as its neighbor brick objects, and the neighbors of the brick object are summarized to form the neighbor table of the brick object.

[0008] Furthermore, the process of determining the temperature difference index includes: For each brick object, obtain the brick temperature of all neighboring brick objects from its neighbor table, calculate the arithmetic mean of the brick temperatures of all neighboring brick objects in its neighbor table, and use it as the average neighbor temperature of that brick object. Calculate the difference between the brick temperature of this brick object and the average temperature of its neighbors, and use this as the temperature difference index of this brick object.

[0009] Furthermore, the candidate bricks are those whose temperature difference index is greater than a preset noise threshold; The process of constructing the local thermal network includes: For each candidate brick, obtain its neighbor table, take all neighbor brick objects in the neighbor table as the first-level node, and initialize the expansion layer number to one; based on the previous layer node, obtain the neighbor table of each brick object in this layer, take the neighbor brick objects that have not been included as the next-level node, and increment the expansion layer number by one; repeat this step until the expansion layer number reaches the preset number. The candidate bricks are merged with all the expanded brick objects to form the node set of the local heat network; Candidate bricks are marked as source points, and all brick objects in the outermost layer of the extended layer are marked as sink points; the connection between source points and sink points is implemented through edges and intermediate nodes in the local heat network, and source points are not directly connected to sink points; Iterate through each brick object in the local heat network node set; for the currently traversed brick object, iterate through each neighbor brick object in its neighbor table; if the neighbor brick object belongs to the local heat network node set, establish an undirected connection edge between the currently traversed brick object and the neighbor brick object; wherein, each pair of neighboring brick objects uniquely corresponds to an undirected connection edge, and no duplicates are established. After traversing all the brick objects in the local heat network node set, the complete edge set of the local heat network is obtained.

[0010] Furthermore, the process of determining the heat flux of each edge includes: For each local thermal network, obtain the first brick object and the second brick object at both ends of each undirected connection edge; Calculate the absolute difference between the texture values ​​of the first brick object and the second brick object, and use it as the edge gap value of the undirected connection edge; calculate the absolute difference between the brick temperatures of the first brick object and the second brick object, and use it as the edge temperature difference value of the undirected connection edge; repeat the above process to obtain the edge gap value and edge temperature difference value of all undirected connection edges in the local thermal network. Collect the edge gap values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the first range; subtract the minimum value from the edge gap value of each undirected connected edge, and divide by the first range to obtain the normalized gap degree of the edge. Collect the edge temperature difference values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the second range; subtract the minimum value from the edge temperature difference value of each undirected connected edge, and divide by the second range to obtain the normalized temperature difference value of the edge. Based on preset weighting coefficients, the normalized gapness and normalized temperature difference are weighted and summed to obtain the comprehensive barrier value; the result of the exponential function is calculated with the natural constant as the base and the negative of the comprehensive barrier value as the exponent, which is used as the heat flux of each undirected connection edge in the local thermal network.

[0011] Furthermore, the step of performing a maximum flow algorithm on a local thermal network based on heat flux to determine the heat leakage from candidate bricks to the network boundary includes: Use the heat flux of each undirected edge in the local thermal network as the initial capacity of the corresponding edge; initialize the total flow to zero; Repeat the path finding iterative process until no new feasible path can be found. The iterative process includes: searching for a connected path from the source to the sink in the local hot network; if the current capacity of each edge on the connected path is greater than zero, the connected path is recorded as a feasible path; if a feasible path exists, the minimum current capacity of all edges on the feasible path is taken as the component flow of the feasible path, and the component flow is added to the total flow; the current capacity of each edge on the feasible path is subtracted from the component flow, and the capacity of the reverse edge of each edge on the feasible path is increased by the component flow; if no feasible path can be found, the iteration is terminated. After the iteration terminates, the accumulated total flow is used as the heat leakage from the candidate brick to the network boundary.

[0012] Furthermore, the isolation degree determination process includes: For each candidate brick, obtain the temperature difference index and heat leakage of the candidate brick; Calculate the sum of heat leakage and the preset smoothing factor, and use it as the third factor; The ratio of the temperature difference index to the third factor is calculated as the isolation degree of the candidate bricks.

[0013] A drone inspection device for potential hazards in the exterior walls of high-rise buildings is disclosed. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of a drone inspection method for potential hazards in the exterior walls of high-rise buildings.

[0014] The present invention has the following beneficial effects: This invention segments visible light images into individual brick objects and assigns a temperature value to each brick object based on a temperature map, thus improving the detection granularity from the pixel level to the single brick level. Furthermore, it filters candidate bricks by calculating the temperature difference between each brick object and its neighboring bricks, effectively removing thermal halo blur caused by lateral heat conduction and significantly improving the accuracy of locating potential hollow areas. For each candidate brick, a local thermal network is constructed, and the heat flux of each edge is determined based on the texture value and temperature differences between adjacent bricks. A maximum flow algorithm is then executed to calculate heat leakage, which quantifies the relationship between the candidate brick and its neighboring bricks. The degree of heat leakage in the surrounding environment is analyzed, and then the isolation degree is determined by combining the temperature difference index and the amount of heat leakage. Since structural voids form a closed heat insulation ring around them, the amount of heat leakage is minimal and the isolation degree is high. In contrast, non-structural interferences such as shadows and water stains have a thermal connection path in at least one direction, resulting in a larger amount of heat leakage and a lower isolation degree. Therefore, by using the isolation degree, structural voids and non-structural interferences can be distinguished from each other from the topological structure, significantly reducing the false alarm rate. Based on the isolation degree, statistical thresholds and absolute lower limits are used for joint screening, which greatly reduces false detections while ensuring the detection rate. Finally, a reliable list of voided bricks and a distribution map of defects are output, enabling high-precision identification of potential voids in the exterior walls of high-rise buildings. Attached Figure Description

[0015] Figure 1 A flowchart of a method for drone inspection of hidden dangers in the exterior walls of high-rise buildings, provided as an embodiment of the present invention; Figure 2 This is an example diagram illustrating the local thermal network construction process provided in one embodiment of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details the specific scheme of the drone inspection device and method for identifying potential hazards in the exterior walls of high-rise buildings provided by this invention.

[0017] Please see Figure 1 The diagram illustrates a flowchart of a drone inspection method for potential hazards in the exterior walls of high-rise buildings, according to an embodiment of the present invention. The method includes: S101: Simultaneously acquire visible light and infrared images of the target's exterior wall; perform dual-light alignment on the visible light and infrared images to generate a temperature mapping map corresponding to the pixel level of the visible light image; segment the visible light image into bricks to extract multiple brick objects; assign a brick temperature value to each brick object based on the temperature mapping map; assign a texture value to each brick object based on the visible light image; determine the neighbor objects of each brick object based on the spatial relationship between them, and form a neighbor table.

[0018] It's important to understand that in actual inspection scenarios, the exterior wall area of ​​a high-rise building is quite large, making it difficult to cover the entire wall in a single shot. Therefore, the "target exterior wall" referred to in this method can refer to the entire exterior wall to be tested, or it can refer to multiple sub-regions after the entire exterior wall has been divided. The drone flies along the planned route, taking pictures section by section. Each waypoint corresponds to a sub-region, and visible light and infrared images of that sub-region are collected. Subsequent processing is done on a per-image basis, and the detection results of each sub-region are finally stitched together to reconstruct a hazard distribution map of the entire wall.

[0019] Although the drone simultaneously acquires visible light and infrared images, the infrared thermal imager and the visible light camera are physically positioned on the drone's gimbal, and their lens focal lengths and fields of view (FOV) typically differ. This can lead to misalignment of temperature and texture information at the same wall surface point, causing data distortion if subsequent analysis is based solely on the visible light and infrared images. Therefore, to ensure strict temporal alignment of the two image streams (from the infrared thermal imager and the visible light camera) and to enable perspective transformation using a homography matrix to obtain a temperature mapping map with the same resolution and coordinate system as the visible light image, simultaneous acquisition of both images is necessary, along with constraints imposed on the drone's shooting conditions.

[0020] In this embodiment, the visible light image and the infrared image are acquired synchronously under preset constraints. The preset constraints include: the UAV hovers or moves vertically in front of the target exterior wall, and the shooting distance is greater than the preset micro-undulation thickness of the target exterior wall cladding bricks, so that the target exterior wall appears as a two-dimensional unfolded plane on the imaging plane.

[0021] Because yaw, pitch, or roll angles occur during drone filming, the projection relationship between the visible light image and the infrared image will degenerate into a more complex three-dimensional perspective transformation. In this case, the homography matrix cannot accurately describe the pixel coordinate mapping relationship between the two, leading to a decrease in registration accuracy. Therefore, those skilled in the art will understand that it is necessary to control the drone to face the wall directly. Specifically, the drone's camera optical axis should be perpendicular to the wall, meaning the drone should hover directly in front of the wall.

[0022] It should be noted that the preset micro-undulation thickness refers to the maximum deviation distance of the surface of the facing brick relative to the average plane of its base layer (i.e., the overall reference plane formed between the adhesive layer of the exterior wall facing brick and the structural wall). This typically includes the thickness of the facing brick itself and the thickness of the adhesive layer, but this embodiment does not specify a particular thickness. For example, according to common exterior wall facing brick construction specifications, the thickness of the ceramic tile is generally 5 to 10 mm, and the thickness of the adhesive layer is 3 to 5 mm. Therefore, the total micro-undulation thickness is usually in the range of 8 to 15 mm. Thus, the preset micro-undulation thickness can be set to 1 cm, which covers the thickness range of most exterior wall facing bricks. In actual operation, since modern UAV flight control systems support high-precision hovering and flight path planning, after the UAV is aligned with the wall using the attitude lock function, the distance to the wall can be measured in real time using the laser rangefinder or ultrasonic sensor onboard the UAV. This ensures that the shooting distance (usually greater than 5 meters) is much greater than the preset micro-undulation thickness. At this point, the projection error caused by the wall's micro-undulations in the image is less than one pixel and can be ignored.

[0023] It is important to note that the phrase "making the target exterior wall appear as a two-dimensional unfolded plane on the imaging plane" in this method does not require the wall surface to be absolutely flat in a physical sense. Rather, it means that, at a given shooting distance and camera resolution, the imaging error caused by the microscopic undulations of the wall surface can be ignored, thus mathematically approximating the two-dimensional plane transformation model. This approximation is a common technique in photogrammetry and computer vision. In practice, users can pre-calculate the shooting distance that satisfies the two-dimensional plane assumption based on the drone's camera resolution and shooting distance. This shooting distance should be greater than the pre-defined thickness of the microscopic undulations in the target exterior wall's cladding bricks, thereby ensuring the effectiveness of this approximation.

[0024] To generate a complete temperature mapping map, as an example, the calibration parameters of the visible light camera and the infrared thermal imager are obtained. Under preset constraints, the homography matrix between the visible light image and the infrared image is calculated based on the calibration parameters. Using the pixel coordinate system of the visible light image as a reference, the infrared image is transformed using the homography matrix, and the transformed infrared image is resampled using bilinear interpolation to generate a temperature mapping map with the same resolution and coordinate system as the visible light image.

[0025] Furthermore, in this embodiment, the calculation of the homography matrix depends not only on the calibration parameters (i.e., intrinsic and extrinsic parameters) of the visible light camera and the infrared thermal imager, but also on the distance to the wall measured in real time by the laser rangefinder or ultrasonic sensor mounted on the aforementioned UAV. Based on the above, since the target wall is approximated as a two-dimensional plane under preset constraints, the spatial position of this two-dimensional plane is uniquely determined by the shooting distance. Therefore, based on the calibration parameters and the shooting distance, the homography matrix from the infrared image coordinate system to the visible light image coordinate system can be uniquely calculated. The specific calculation process of the homography matrix is ​​well-known technology and will not be elaborated here.

[0026] Calibration parameters are a set of data describing the internal imaging geometry of a camera and the relative positional relationships between cameras, including intrinsic and extrinsic parameters. In practice, calibration parameters can be obtained through offline calibration, a technique well-known to those skilled in the art, and will not be elaborated upon in this embodiment. Specifically, in a laboratory environment, a calibration board is used to calibrate a visible light camera and an infrared thermal imager, respectively. By capturing multiple sets of images of the calibration board in different poses, the intrinsic and extrinsic parameters are calculated using classic algorithms such as the Zhang Zhengyou calibration method.

[0027] The homography matrix is ​​a 3×3 projection transformation matrix that describes the mapping relationship between image coordinates of the same plane under two different camera viewpoints. Its calculation process is a well-known technique in the art and will not be detailed in this embodiment. Specifically, based on the acquired intrinsic and extrinsic parameters of the visible light camera and infrared thermal imager, combined with the assumption of a two-dimensional plane on the target's outer wall, the homography matrix from the infrared image coordinate system to the visible light image coordinate system can be calculated. That is, for each pixel in the infrared image, its pixel coordinates are multiplied by the homography matrix to obtain its mapped coordinates in the visible light image coordinate system.

[0028] Pixel coordinates are numerical values ​​describing the position of each pixel in an image. They are used to locate the position of each pixel in a visible light image and a temperature map, so that temperature information and texture information can be accurately mapped to the same point on the wall. Specifically, the origin is usually the top left corner of the image, with the x-axis pointing to the right and the y-axis pointing downwards. For example, in an image with a resolution of 1920×1080, the coordinates of the first pixel in the top left corner are (0, 0), and the coordinates of the 100th pixel to the right and the 200th pixel downwards are (100, 200).

[0029] It is important to understand that the aforementioned calibration parameter acquisition, homography matrix calculation, perspective transformation, and bilinear interpolation resampling are all mature technologies in the fields of image processing and computer vision. The improvement of this invention lies not in these steps themselves, but in applying them to the specific scenario of detecting hollow areas in the exterior walls of high-rise buildings, and combining them with subsequent techniques such as brick segmentation, temperature difference screening, local thermal network construction, and maximum flow verification. Therefore, this embodiment does not elaborate on the aforementioned prior art; those skilled in the art can implement these steps based on common knowledge.

[0030] It should be noted that segmenting a visible light image into bricks and extracting multiple brick objects typically involves steps such as edge detection, morphological closing operations, connected component extraction, and size prior screening. These are all mature technologies in the fields of image processing and computer vision. Specifically, edge detection (e.g., Canny and Sobel operators) is used to extract grout lines; morphological closing operations are used to connect broken gaps; connected component extraction is used to mark closed regions as bricks; and size prior screening is used to remove false targets such as stains. Therefore, it is understandable that the improvement of this method does not lie in these segmentation steps themselves, but in associating the segmented brick objects with a temperature map at the pixel level, assigning temperature and texture attributes to each brick object, and combining this with subsequent techniques such as neighbor table construction, temperature difference filtering, and local thermal networks. Therefore, this embodiment will not elaborate on the above-mentioned prior art parts in detail; those skilled in the art can implement brick segmentation and brick object extraction based on common knowledge.

[0031] The process of assigning brick temperature and texture values ​​to each brick object includes the following steps: 1) For each brick object, obtain the pixel mask of the brick object in the visible light image. The pixel mask contains the coordinate positions of all pixels covered by the brick object.

[0032] It should be noted that the "coordinate position" or "pixel coordinate" referred to in this embodiment refers to the position coordinates of a pixel in the image, with the upper left corner of the image as the origin, the rightward direction as the horizontal coordinate, and the downward direction as the vertical coordinate.

[0033] In this process, each extracted brick object corresponds to a closed pixel region. The coordinates of all pixels within this region are recorded to form a binary mask. In this mask, pixel positions belonging to the brick object are marked as 1, and those not belonging are marked as 0. For example, if a brick covers a rectangular region in the image with horizontal coordinates from 100 to 150 and vertical coordinates from 200 to 250, then its pixel mask contains all integer coordinate positions within that rectangular region.

[0034] 2) Based on the pixel mask, index all pixels at the same coordinate position in the temperature map; based on the pixel mask, index all pixels at the same coordinate position in the visible light image.

[0035] 3) Read the temperature value and grayscale value corresponding to each indexed pixel.

[0036] Since the temperature map and the visible light image have the exact same resolution and coordinate system, the same pixel mask can be directly applied to both the temperature map and the visible light image without coordinate transformation. Specifically, each pixel coordinate in the pixel mask is traversed, the pixel with the same coordinates is located in the temperature map, and the temperature value and grayscale value stored in that pixel are read.

[0037] It should be noted that if the visible light image is a color visible light image, its red, green, and blue channel pixel values ​​must first be converted to grayscale values. The specific conversion process can use the standard grayscale conversion formula, which is an existing technical method and will not be elaborated further.

[0038] 4) Calculate the arithmetic mean of the temperature values ​​of all read pixels as the brick temperature of the brick object; calculate the arithmetic mean of the grayscale values ​​of all read pixels as the texture value of the brick object.

[0039] Texture values ​​reflect the average brightness or color depth of a brick surface. A greater difference in texture values ​​between two adjacent bricks indicates a more pronounced visual boundary between them (such as grout lines).

[0040] To accurately determine the neighboring objects of each brick object, as a preferred implementation, for each brick object, its boundary pixels are identified. A boundary pixel is a pixel whose neighboring pixel does not belong to the brick object. The boundary pixels of the brick object are paired with the boundary pixels of each other brick object to form boundary pixel pairs. The Euclidean distance between each pair of boundary pixel pairs is calculated as the interval distance. The minimum interval distance between the brick object and the corresponding other brick objects is extracted. If the minimum interval distance is less than a preset interval threshold, it is determined that the brick object and the corresponding other brick objects constitute a neighbor relationship. All other brick objects that constitute a neighbor relationship with the brick object are recorded as its neighbor brick objects, and the neighbors of the brick object are summarized to form the neighbor table of the brick object.

[0041] Neighborhood pixels refer to other pixels in adjacent positions around a given pixel. In this embodiment, a four-neighborhood definition is preferred. For example, each pixel in the pixel mask of a brick object is traversed, and the four adjacent directions (i.e., the four-neighborhood) of the pixel are checked. If a pixel in any of the adjacent directions does not belong to the pixel mask of the brick object, then the pixel is marked as a boundary pixel.

[0042] In this context, "corresponding other brick objects" refers to all other brick objects besides the brick object itself. When constructing the neighbor table, the brick object needs to be paired with each of these other brick objects individually, calculating the minimum distance between them and determining if they are adjacent. For example, with 100 brick objects, brick object number 5 needs to be paired and analyzed one by one with the 99 brick objects numbered 1 to 4 and 6 to 100.

[0043] It should be noted that the Euclidean distance calculation method is a common technique, which can be directly applied in this embodiment and will not be elaborated here.

[0044] It should be noted that the specific value of the preset interval threshold can be adjusted according to the resolution of the visible light image and the shooting distance; this embodiment does not impose a specific limitation. For example, if the shooting distance is 10 meters and the camera focal length is 25 millimeters, each pixel corresponds to an actual length of approximately 5 millimeters. Assuming the width of the sealant is 6 millimeters, this can be converted to 1.2 pixels. Considering the segmentation error, the preset interval threshold can be rounded down to 2 pixels.

[0045] It's important to understand that in actual testing, bricks located at wall edges, corners, or opening edges may have fewer than two neighboring bricks. For these bricks with too few neighboring bricks, the reliability of subsequent temperature difference calculations will be low, often lacking statistical significance, and they cannot form an effective multi-layered topology when expanding into a local thermal network. Therefore, these bricks with too few neighboring bricks can be marked as edge bricks and excluded from the candidate brick selection process. They should be listed separately in the final test report or manually reviewed.

[0046] As one possible implementation, if the number of neighboring brick objects in the neighbor table of a certain brick object is less than the preset minimum number of neighbors, the brick object is marked as an edge brick.

[0047] It should be noted that the specific value of the preset minimum number of neighbors can be determined based on industry experience, and this embodiment does not impose a specific limitation. For example, the preset minimum number of neighbors is usually set to 2.

[0048] S102: Based on the brick temperature of each brick object and its neighbor table, determine the temperature difference index of each brick object and screen out candidate bricks.

[0049] It is important to understand that this method is generally preferred for testing during the day when there is ample sunlight and the weather is clear. Because the hollow areas have higher thermal resistance, they absorb and accumulate heat during the warming process under sunlight, causing their surface temperature to be significantly higher than the surrounding normal wall surface, creating a clear thermal contrast. Therefore, testing is usually conducted during sunny days. If testing is performed on cloudy days, rainy days, or in the early morning or late evening when thermal contrast is insufficient, the temperature difference in the hollow areas may be masked by environmental noise, affecting the testing results. Those skilled in the art will understand that the specific testing time can be adjusted appropriately based on local climate, wall orientation, and other factors.

[0050] In a preferred embodiment, for each brick object, the brick temperatures of all neighboring brick objects are obtained from its neighbor table, and the arithmetic mean of the brick temperatures of all neighboring brick objects in its neighbor table is calculated as the average neighbor temperature of the brick object; the difference between the brick temperature of the brick object and the average neighbor temperature is calculated as the temperature difference index of the brick object.

[0051] It should be noted that if the inspection process takes place at night or on a cloudy day, the absolute value of the difference between the temperature of the brick in question and the average temperature of the neighboring bricks can be taken as the temperature deviation, so as to make it easier to uniformly measure the degree of deviation.

[0052] The neighbor average temperature reflects the average thermal state of the local environment surrounding a particular brick object.

[0053] Because the air gap in the hollow area has a high thermal resistance, it hinders heat conduction into the wall during the warming process caused by sunlight. This results in the surface temperature of the hollow area being significantly higher than the surrounding normal wall surface; while the normal wall surface has a similar temperature to the neighboring walls, with no significant difference. Therefore, by calculating the difference between the temperature of a particular brick and the average temperature of its neighbors, the degree of thermal deviation of that brick relative to the local environment can be quantified. A larger temperature difference (i.e., a larger positive value) indicates that the brick is much hotter than its neighbors, and the higher the probability of it being a hollow area. If the temperature difference is close to zero or negative, it indicates that the temperature of the brick is similar to or lower than its neighbors, and the higher the probability of it being a normal wall surface or cold spot interference.

[0054] It should be noted that if the temperature difference index of a certain brick object is negative, it means that the temperature of the brick object is lower than the average temperature of its surrounding neighbors, exhibiting a "cold spot" characteristic. As described above, this method is mainly applicable to daytime sunshine heating conditions. Hollow areas accumulate heat due to their higher thermal resistance, usually showing a positive temperature difference index. Therefore, for negative temperature difference indices, the absolute value can be taken depending on the actual detection scenario. For example, in nighttime or cloudy conditions, the absolute difference is calculated. However, in this embodiment, the original difference is retained and uniformly processed during subsequent screening using a preset noise threshold.

[0055] In this embodiment, the candidate brick is the brick object whose temperature difference index is greater than a preset noise threshold.

[0056] The preset noise threshold is used to distinguish whether the temperature difference index of the brick object is significantly greater than the normal thermal fluctuation range.

[0057] It should be noted that the specific value of the preset noise threshold can be determined based on the noise equivalent temperature difference of the infrared thermal imager, and this embodiment does not impose a specific limitation. Specifically, the noise equivalent temperature difference is an inherent performance parameter of the infrared thermal imager, representing the smallest temperature difference that the thermal imager can distinguish. Considering the existence of various random noises in actual detection (such as sensor noise, atmospheric disturbances, etc.), directly using the noise equivalent temperature difference as the preset noise threshold would lead to a large number of misjudgments. Therefore, the preset noise threshold can be set as a multiple of the noise equivalent temperature difference. For example, in this embodiment, the preset noise threshold is set to 3 to 5 times the noise equivalent temperature difference. For instance, if the noise equivalent temperature difference of a certain infrared thermal imager is 0.05℃, taking 4 times yields a preset noise threshold of 0.2℃.

[0058] S103: For each candidate brick, construct a local thermal network centered on it; determine the heat flux of each edge based on the difference in texture value and temperature of the bricks at both ends of each edge in the local thermal network; based on the heat flux, execute the maximum flow algorithm on the local thermal network to determine the heat leakage from the candidate brick to the network boundary; determine the isolation degree based on the temperature difference index and heat leakage of the candidate brick.

[0059] It's important to understand that the temperature difference index of a single candidate brick only reflects its temperature deviation relative to neighboring bricks, and cannot determine whether the abnormal heat involved in that candidate brick has been effectively isolated by the surrounding environment. The core characteristic of structural voids is that heat is trapped by surrounding air gaps, forming "thermal islands," while non-structural disturbances have thermal connectivity paths in at least one direction. Therefore, a local thermal network can be constructed centered on the candidate brick. By simulating the process of heat diffusion from the center to the surroundings, the degree of heat leakage can be quantified from a topological perspective, thus distinguishing between structural voids and non-structural disturbances.

[0060] The process of constructing a local thermal network is as follows: Figure 2 As shown, it includes: S103-1: For each candidate brick, obtain its neighbor table, take all neighbor brick objects in the neighbor table as the first-level node, and initialize the expansion layer number to one; based on the previous layer node, obtain the neighbor table of each brick object in this layer, take the neighbor brick objects that have not been included as the next-level node, and increment the expansion layer number by one; repeat this step until the expansion layer number reaches the preset number of layers.

[0061] It is understandable that the brick objects associated with the first-level nodes are the bricks that are directly adjacent to the candidate bricks.

[0062] It should be noted that "brick object" is a general term, including all extracted individual facing brick units; "neighbor brick object" specifically refers to other brick objects that have a spatial adjacency relationship with a certain brick object. In the process of constructing the local thermal network, the nodes obtained by expanding layer by layer include both the direct neighbors of the candidate brick itself (i.e., the first layer) and the neighbors of these neighbors (i.e., the second layer and above). Although these nodes are all obtained by expanding through "neighbor relationships", they are not all direct neighbors of the candidate brick.

[0063] S103-2: Merge the candidate bricks with all the expanded brick objects to form the node set of the local thermal network.

[0064] It should be noted that the specific value of the preset number of layers can be determined based on industry experience to ensure sufficient coverage of the hollow area boundary and its surrounding background area. For example, a typical preset number of layers is 3.

[0065] S103-3: Mark the candidate bricks as source points and mark all brick objects in the outermost layer of the extended layer as sink points; the connection between the source point and the sink point is realized through the edges and intermediate nodes in the local heat network, and the source point is not directly connected to the sink point.

[0066] The source point represents the starting point of the virtual heat flow, that is, the starting point where heat spreads from the candidate brick to the surrounding area.

[0067] Since all the outermost brick objects are far from the candidate bricks, their temperature is minimally affected by the heat diffusion from the candidate bricks. Therefore, these outermost brick objects can be approximated as a background heat pool. Thus, the sink represents the point where the virtual heat flow dissipates, i.e., the location where heat ultimately leaks.

[0068] S103-4: Traverse each brick object in the local heat network node set; for the currently traversed brick object, traverse each neighbor brick object in its neighbor table; if the neighbor brick object belongs to the local heat network node set, establish an undirected connection edge between the currently traversed brick object and the neighbor brick object; wherein, each pair of mutually neighboring brick objects uniquely corresponds to an undirected connection edge, and no duplicates are established.

[0069] It's important to note that there are no direct edges connecting the source and sink nodes. The connections between them are established through intermediate nodes in the local heat network (i.e., the brick objects in the first, second, and other intermediate layers) and the edges between them. Specifically, heat from the source node first flows to the first-layer nodes, then from the first-layer nodes to the second-layer nodes, and so on, eventually reaching the outermost sink node.

[0070] An undirected edge means that heat can be transferred in two directions, but this method mainly focuses on the direction of diffusion from the source point (candidate brick) to the outside. An undirected edge can be equivalent to two directed edges in opposite directions in the maximum flow algorithm.

[0071] Neighboring brick objects refer to two brick objects that are spatially adjacent to each other. That is, one brick object's neighbor list contains the other brick object, and vice versa. Since the neighbor list is constructed based on the symmetry of spatial relationships (if A and B are adjacent, then B and A are also adjacent), neighboring brick objects are actually any pair of adjacent brick objects. For example, if brick A and brick B are neighbors, when traversing to A, an edge AB is created; when traversing to B, if edge AB already exists, it is not created again.

[0072] S103-5: After traversing all the brick objects in the local heat network node set, the complete edge set of the local heat network is obtained.

[0073] It's important to understand that due to the physical gaps at the boundary of a hollow area and the significant temperature difference between its two sides, heat is difficult to leak outwards through this boundary. In contrast, non-structural disturbances (such as shadows, water stains, and thermal bridges), although exhibiting localized temperature anomalies, do not have physical gaps at their boundaries, allowing heat to conduct outwards along the continuous material. Therefore, a heat flux can be defined for each edge in the local thermal network to quantify the ease with which heat can pass through the boundary. A smaller heat flux indicates stronger obstruction and greater difficulty for heat to pass through; a larger heat flux indicates weaker obstruction and easier heat to pass through. By assigning heat flux to each edge, the subsequent maximum flow algorithm can use these heat fluxes as capacity constraints to simulate the process of heat diffusion from candidate bricks to their surroundings, thereby distinguishing between structural hollow areas surrounded by high-obstruction boundaries and non-structural disturbances with low-obstruction connected paths.

[0074] The process of determining the heat flux of each edge includes the following steps: 1) For each local thermal network, obtain the first brick object and the second brick object at both ends of each undirected connection edge.

[0075] 2) Calculate the absolute difference between the texture values ​​of the first brick object and the second brick object, and use it as the edge gap value of the undirected connection edge; calculate the absolute difference between the brick temperatures of the first brick object and the second brick object, and use it as the edge temperature difference value of the undirected connection edge; repeat the above process to obtain the edge gap value and edge temperature difference value of all undirected connection edges in the local thermal network.

[0076] The edge gap value represents the degree of texture difference between two adjacent brick objects in a visible light image. A larger edge gap value indicates a more obvious visual boundary line (such as grout lines) between the two bricks, meaning a more obvious physical gap.

[0077] The edge temperature difference value indicates the degree of surface temperature difference between two adjacent brick objects. The larger the edge temperature difference value, the more significant the thermal break between the two bricks, meaning that heat is more difficult to conduct through the boundary.

[0078] It should be noted that the principle behind using the absolute difference in texture values ​​between adjacent bricks as the edge gap in this embodiment is as follows: In typical exterior wall cladding brick scenarios, the color of the grout between adjacent bricks is usually significantly different from the brick surface color. Therefore, the overall texture value of the brick itself can effectively reflect the visual salience of the gap. However, for the very few cases where the brick surface color is similar to the grout color, the edge gap value may approach zero. But in this method, the calculation of heat flux also depends on the edge temperature difference. If there is a significant thermal break at the hollow boundary (i.e., a large edge temperature difference), even if the edge gap value is small, the overall barrier value is still large. Therefore, even if the heat flux approaches zero, it does not affect the determination of the hollow boundary. Based on the above, those skilled in the art will understand that the design of this embodiment represents a reasonable trade-off between accuracy and efficiency.

[0079] 3) Collect the edge gap values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the first range; subtract the minimum value from the edge gap value of each undirected connected edge, and divide by the first range to obtain the normalized gap degree of the edge.

[0080] 4) Collect the edge temperature difference values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the second range; subtract the minimum value from the edge temperature difference value of each undirected connected edge, and divide by the second range to obtain the normalized temperature difference value of the edge.

[0081] It should be noted that during the min-max normalization process, if all edges in a local thermal network have the same gap value (i.e., the maximum value equals the minimum value), then the normalized gap value for all edges in that local thermal network is uniformly assigned to 0. Similarly, if all edges have the same temperature difference value (i.e., the maximum value equals the minimum value), then the normalized temperature difference value is uniformly assigned to 0. Based on the above processing, it is ensured that the normalization result always lies in the interval between 0 and 1, and the denominator is not zero.

[0082] Normalized gapness represents the gap value of an undirected connection edge in a local thermal network relative to the overall distribution. Specifically, if the normalized gapness of an undirected connection edge in a local thermal network is greater than 1, it indicates that the gap value is higher than the average level.

[0083] The normalized temperature difference value represents the degree to which the temperature difference value of a certain undirected connecting edge in a local thermal network is relative to the overall distribution level of the temperature difference values ​​of all edges in the local thermal network. Specifically, if the normalized temperature difference value of a certain undirected connecting edge in a local thermal network is greater than 1, it means that the temperature difference value of that edge is higher than the average level of the network, that is, the temperature difference between the two sides of the boundary is relatively significant.

[0084] 5) Based on the preset weighting coefficients, the normalized gap degree and the normalized temperature difference value are weighted and summed to obtain the comprehensive barrier value; the result of the exponential function is calculated with the natural constant as the base and the negative number of the comprehensive barrier value as the exponent, and is used as the heat flux of each undirected connection edge in the local thermal network.

[0085] The preset weighting coefficients are used to adjust the contribution ratio of normalized gap size and normalized temperature difference in the overall barrier value. It should be noted that the specific values ​​of the preset weighting coefficients can be adjusted according to the type of exterior wall material of different buildings; this embodiment will not elaborate further. Specifically, for walls with obvious visual gap characteristics (such as dark-colored grout), the weight of normalized gap size can be appropriately increased; for walls with inconspicuous visual gap characteristics (such as seamless stone), the weight of normalized gap size can be appropriately decreased. For example, the weighting coefficients for both normalized gap size and normalized temperature difference are set to 1, meaning they are equally important.

[0086] For example, assuming the preset weighting coefficients include a first weighting coefficient and a second weighting coefficient, where the first weighting coefficient is the weighting coefficient corresponding to the normalized gap size and the second weighting coefficient is the weighting coefficient corresponding to the normalized temperature difference value, then the overall barrier value = first weighting coefficient × normalized gap size + second weighting coefficient × normalized temperature difference value.

[0087] The overall barrier value comprehensively reflects the degree to which a certain edge in a local thermal network hinders heat conduction. Specifically, a larger overall barrier value for a certain edge in a local thermal network indicates that the edge has both a greater gap difference and a greater temperature difference, meaning a stronger barrier to heat passing through that edge. Conversely, a smaller overall barrier value for a certain edge in a local thermal network indicates that both the gap difference and temperature difference are smaller, meaning a weaker barrier to heat passing through that edge.

[0088] It's important to understand that a higher overall barrier value results in a heat flux closer to zero (meaning heat cannot pass through as much as possible); conversely, a lower overall barrier value results in a heat flux closer to one (meaning heat can pass through more easily). Since the overall barrier value is a weighted sum of the normalized gap size and the normalized temperature difference, it's understandable that a higher gap size and temperature difference result in a higher overall barrier value and a heat flux closer to zero; conversely, a lower gap size and temperature difference result in a lower overall barrier value and a heat flux closer to one.

[0089] In this embodiment, the heat flux of each undirected connected edge in the local thermal network is used as the initial capacity of the corresponding edge; the total flow is initialized to zero; the path finding iteration process is repeated until no new feasible path can be found; wherein, the iteration process includes: searching for a connected path from the source to the sink in the local thermal network; if the current capacity of each edge on the connected path is greater than zero, the connected path is recorded as a feasible path; if a feasible path exists, the minimum value of the current capacity of all edges on the feasible path is taken as the component flow of the feasible path, and the component flow is added to the total flow; the current capacity of each edge on the feasible path is subtracted from the component flow, and the capacity of the reverse edge of each edge on the feasible path is increased by the component flow; if no feasible path can be found, the iteration is terminated; after the iteration terminates, the accumulated total flow is used as the heat leakage from the candidate brick to the network boundary.

[0090] Since the maximum flow algorithm requires setting a capacity limit for each edge, representing the maximum flow that edge can transmit, this method directly uses the heat flux of each edge as its capacity constraint. For example, if the heat flux of an edge is 0.8, it means that the edge can transmit a maximum of 0.8 units of heat; if the heat flux of another edge is 0.1, it means that the edge can only transmit a maximum of 0.1 units of heat, which is almost blocked.

[0091] Furthermore, in the maximum flow algorithm, each original edge corresponds to two directed edges in opposite directions: a forward edge and a reverse edge. Initially, the capacity of a forward edge equals its heat flux, while the capacity of a reverse edge is zero. For example, if a feasible path carries a component flow *f*, the forward capacity of each edge on that feasible path decreases by *f*, while the capacity of its reverse edge increases by *f*. This allows the previously allocated flow to be "reclaimed" in subsequent iterations via the reverse edge, thus finding the global maximum flow and avoiding the inability to find a globally optimal solution due to premature use of shared edges.

[0092] It should be noted that the reverse edge mechanism is a standard component of the maximum flow algorithm and is publicly known information, so it can be applied directly here. Specifically, after an edge has carried some flow, the addition of reverse edges allows the previously occupied flow to be "reclaimed" in subsequent iterations, thereby reallocating resources.

[0093] It is understandable that in each iteration, a connected path from the source to the sink needs to be searched in the local hot network. The search method can be breadth-first search or depth-first search, that is, starting from the source, exploring outward layer by layer along the edges with the current capacity greater than zero until any sink node is reached.

[0094] Since the flow that the entire feasible path can transport is limited by the edge with the smallest capacity (i.e., the bottleneck edge), the minimum current capacity of all edges on the feasible path is taken as the component flow of that feasible path, and the component flow represents the transport capacity of that feasible path.

[0095] It should be noted that in each iteration, after updating the capacity of each edge on the feasible path, the edge with the smallest capacity will become zero (because the value equal to its capacity has been subtracted), indicating that the edge has been fully utilized and is no longer available in subsequent iterations; the capacity of other edges will be reduced accordingly, but may still be greater than zero, and can continue to be used in subsequent iterations.

[0096] For example, if the current capacities of three edges on a feasible path are 0.5, 0.3, and 0.4 respectively, then the component flow is 0.3. After the update, the capacities of the three edges will become 0.2, 0, and 0.1 respectively. The edges whose capacities become zero will no longer be considered in subsequent searches.

[0097] It is worth noting that the local heat network constructed by this method is a directed graph with source and sink nodes and edges with non-negative capacity (i.e., an undirected edge can be equivalent to two directed edges in opposite directions), which fully satisfies the applicability conditions of the maximum flow algorithm. The maximum flow algorithm, starting from the source node, searches layer by layer along the remaining edges with capacity greater than zero, and will inevitably find all feasible paths. Therefore, the maximum flow algorithm can be directly used in this scenario, and the specific process can be understood by those skilled in the art based on common knowledge; it will not be elaborated here.

[0098] Heat leakage reflects the maximum amount of heat that can be transferred from the source to the sink under a given network capacity constraint. A smaller heat leakage for a candidate brick indicates that the brick is surrounded by a high-barrier boundary, making heat leakage more difficult, which implies a stronger correlation with the characteristics of a structurally hollow thermal island. Conversely, a larger heat leakage for a candidate brick indicates the existence of at least one low-barrier path, allowing heat to leak more smoothly, which is considered a non-structural disturbance.

[0099] As a preferred implementation, for each candidate brick, the temperature difference index and heat leakage of the candidate brick are obtained; the sum of the heat leakage and the preset smoothing factor is calculated as the third factor; the ratio of the temperature difference index to the third factor is calculated as the isolation degree of the candidate brick.

[0100] Since the temperature difference index reflects the degree of thermal deviation of a candidate brick relative to its local environment, a larger temperature difference index indicates that the candidate brick is hotter than its neighbors. Heat leakage reflects the degree to which the candidate brick is isolated by its surrounding boundaries; a smaller heat leakage indicates that the candidate brick is more likely to be surrounded by a high-barrier boundary, making heat leakage more difficult. Therefore, by comprehensively considering both the temperature difference index and heat leakage, the isolation degree of the candidate brick can be determined, further distinguishing between structural voids and non-structural disturbances.

[0101] Isolation reflects the degree of temperature rise of a candidate brick under a unit heat leakage. A higher isolation level indicates less heat leakage at the same temperature rise, meaning the heat is more effectively contained within the brick, better matching the "thermal island" characteristic of structural voids. Conversely, a lower isolation level suggests that while the brick may experience a temperature rise, heat can leak outwards through at least one connecting path, making it more likely to be a non-structural disturbance.

[0102] It should be noted that the preset smoothing factor should be much smaller than the normal heat leakage rate to avoid having a substantial impact on the calculation results. For example, since the heat flux is between 0 and 1, the heat leakage rate in a local thermal network is usually also in the range of 0 to 1. Therefore, setting the preset smoothing factor to 0.000001 can effectively prevent division by zero errors without significantly changing the value of the heat leakage rate.

[0103] S104: Based on the isolation degree of candidate bricks, a dual threshold screening is performed using statistical threshold and absolute lower limit value, and a list of bricks diagnosed with hollowness and a disease distribution map are output to identify potential hollowness hazards in the exterior wall.

[0104] In this embodiment, the isolation degrees of all candidate bricks are collected to form an isolation degree dataset; the values ​​in the isolation degree dataset are arranged in ascending order to obtain a sorted isolation degree sequence; the proportion of isolation degrees located at a preset high quantile in the sorted isolation degree sequence is used as a statistical threshold; the maximum value between the statistical threshold and the absolute lower limit is taken as a judgment threshold; candidate bricks with isolation degrees greater than the judgment threshold are identified as bricks diagnosed as hollow; a list of hollow bricks containing the location information of the diagnosed hollow bricks is output, and a disease distribution map with hollow markings is generated.

[0105] It should be noted that the specific value of the preset high-resolution locus proportion can be adjusted based on the tolerance of the missed detection rate and false alarm rate for the survey task. Those skilled in the art can determine it in combination with the actual scenario, and this embodiment does not impose a specific limitation. For example, in order to reduce the false alarm rate, a higher proportion can be adopted, such as a preset high-resolution locus proportion of 95%.

[0106] It should be noted that the specific value of the absolute lower limit can be derived based on a large amount of experimental data. Specifically, in typical testing scenarios, the isolation degree of an exterior wall with known hollowness is tested and calculated. A histogram of the isolation degree distribution between hollow and non-hollow bricks is plotted, and the boundary value that clearly distinguishes the two types of samples is taken as the absolute lower limit. For example, the isolation degree of true hollowness is usually greater than 5, while the isolation degree of non-hollowness is usually less than 5, so the absolute lower limit can be taken as 5.

[0107] Typical testing scenarios refer to the testing environment recommended in step S102 above, which mainly includes: testing under conditions of sufficient sunlight and clear weather during the day; the drone hovering or moving vertically directly in front of the wall, with the shooting distance being much greater than the preset micro-undulation thickness of the target exterior wall cladding bricks, etc.

[0108] It should be noted that the absolute lower limit value can be readjusted according to different detection scenarios (such as different infrared thermal imager models and different building types).

[0109] For example, the hollow brick list is a structured report. The process of creating the hollow brick list is as follows: count the number of brick objects diagnosed as hollow; obtain the preset physical length and physical width of a single brick, and multiply the two to obtain the physical area of ​​a single brick; multiply the physical area of ​​a single brick by the number of bricks diagnosed as hollow to obtain the total hollow area; export the structured report, in which each row corresponds to a brick object diagnosed as hollow, and at least includes the grid coordinates, isolation degree and total hollow area of ​​the brick object.

[0110] The preset physical length and width of a single brick can be directly read according to the actual specifications of the facing bricks used; this embodiment does not impose specific limitations. For example, the physical length of a single brick can be 0.2 meters, and the physical width can be 0.1 meters.

[0111] For example, the process of creating a disease distribution map is as follows: acquire the original visible light image; create a semi-transparent layer of the same size as the original visible light image; traverse all confirmed hollow brick objects and obtain the pixel area of ​​each confirmed hollow brick object in the visible light image; in the semi-transparent layer, fill the pixel area corresponding to each confirmed hollow brick object with a highlight color and mark the grid coordinates of the brick object at the center of the area; overlay the semi-transparent layer with the original visible light image to generate a disease distribution map.

[0112] It is important to understand that the list of hollow areas and the distribution map of defects are the final outputs of this method, and their specific generation methods are conventional techniques in this field, and are not limited thereto. Those skilled in the art can use any suitable report export and image annotation method to achieve the above outputs according to actual engineering needs. The above description is only one optional implementation method.

[0113] A drone inspection device for potential hazards in the exterior walls of high-rise buildings includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of a drone inspection method for potential hazards in the exterior walls of high-rise buildings.

Claims

1. A method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings, characterized in that, The method includes: Simultaneously acquire visible light and infrared images of the target's exterior wall; perform dual-light alignment on the visible light and infrared images to generate a temperature mapping map corresponding to the pixel level of the visible light image; segment the visible light image into bricks to extract multiple brick objects; assign a brick temperature value to each brick object based on the temperature mapping map; assign a texture value to each brick object based on the visible light image; determine the neighbor objects of each brick object based on the spatial relationship between them, and form a neighbor table. Based on the brick temperature of each brick object and its neighbor table, determine the temperature difference index of each brick object and screen out candidate bricks; For each candidate brick, a local thermal network is constructed with it as the center; the heat flux of each edge is determined based on the difference in texture value and temperature of the bricks at both ends of each edge in the local thermal network; based on the heat flux, the maximum flow algorithm is executed on the local thermal network to determine the heat leakage from the candidate brick to the network boundary; the isolation degree is determined based on the temperature difference index and heat leakage of the candidate brick. Based on the isolation degree of candidate bricks, a dual threshold screening is performed using statistical threshold and absolute lower limit value, and a list of bricks diagnosed with hollowness and a disease distribution map are output to identify potential hollowness hazards in exterior walls.

2. The method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 1, characterized in that, The visible light image and the infrared image are acquired synchronously under preset constraints. The preset constraints include: the UAV hovers or moves vertically in front of the target exterior wall, and the shooting distance is greater than the preset micro-undulation thickness of the target exterior wall cladding bricks, so that the target exterior wall appears as a two-dimensional unfolded plane on the imaging plane. The temperature mapping generation process includes: Obtain the calibration parameters of the visible light camera and the infrared thermal imager; Under preset constraints, the homography matrix between the visible light image and the infrared image is calculated based on the calibration parameters; Using the pixel coordinate system of the visible light image as a reference, the infrared image is transformed by the homography matrix, and the transformed infrared image is resampled by the bilinear interpolation method to generate a temperature mapping map with the same resolution and coordinate system as the visible light image.

3. The method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 2, characterized in that, The process of assigning the brick temperature and texture values ​​to each brick object includes: For each brick object, obtain the pixel mask of the brick object in the visible light image. The pixel mask contains the coordinate positions of all pixels covered by the brick object. Based on pixel masks, index all pixels at the same coordinate position in the temperature map; based on pixel masks, index all pixels at the same coordinate position in the visible light image. Read the temperature value and grayscale value corresponding to each indexed pixel; Calculate the arithmetic mean of the temperature values ​​of all read pixels as the brick temperature of the brick object; calculate the arithmetic mean of the grayscale values ​​of all read pixels as the texture value of the brick object.

4. The method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 1, characterized in that, The step of determining the neighbor objects of each brick object based on the spatial positional relationship between each brick object and forming a neighbor table includes: For each brick object, identify its boundary pixels. A boundary pixel is a pixel whose any neighboring pixel does not belong to the brick object. The boundary pixels of the brick object are paired with the boundary pixels of every other brick object to form boundary pixel pairs; the Euclidean distance between each pair of boundary pixel pairs is calculated as the interval distance; the minimum interval distance between the brick object and the corresponding other brick objects is extracted. If the minimum interval distance is less than the preset interval threshold, it is determined that the brick object is a neighbor to other corresponding brick objects; all other brick objects that are neighbors to the brick object are recorded as its neighbor brick objects, and the neighbors of the brick object are summarized to form the neighbor table of the brick object.

5. A method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 4, characterized in that, The process for determining the temperature difference index includes: For each brick object, obtain the brick temperature of all neighboring brick objects from its neighbor table, calculate the arithmetic mean of the brick temperatures of all neighboring brick objects in its neighbor table, and use it as the average neighbor temperature of that brick object. Calculate the difference between the brick temperature of this brick object and the average temperature of its neighbors, and use this as the temperature difference index of this brick object.

6. The method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 1, characterized in that, The candidate brick is a brick object whose temperature difference index is greater than a preset noise threshold. The process of constructing the local thermal network includes: For each candidate brick, obtain its neighbor table, take all neighbor brick objects in the neighbor table as the first-level node, and initialize the expansion layer number to one; based on the previous layer node, obtain the neighbor table of each brick object in this layer, take the neighbor brick objects that have not been included as the next-level node, and increment the expansion layer number by one; repeat this step until the expansion layer number reaches the preset number. The candidate bricks are merged with all the expanded brick objects to form the node set of the local heat network; Candidate bricks are marked as source points, and all brick objects in the outermost layer of the extended layer are marked as sink points; the connection between source points and sink points is implemented through edges and intermediate nodes in the local heat network, and source points are not directly connected to sink points; Iterate through each brick object in the local heat network node set; for the currently traversed brick object, iterate through each neighbor brick object in its neighbor table; if the neighbor brick object belongs to the local heat network node set, establish an undirected connection edge between the currently traversed brick object and the neighbor brick object; wherein, each pair of neighboring brick objects uniquely corresponds to an undirected connection edge, and no duplicates are established. After traversing all the brick objects in the local heat network node set, the complete edge set of the local heat network is obtained.

7. A method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 6, characterized in that, The process of determining the heat flux of each edge includes: For each local thermal network, obtain the first brick object and the second brick object at both ends of each undirected connection edge; Calculate the absolute difference between the texture values ​​of the first brick object and the second brick object, and use it as the edge gap value of the undirected connection edge; calculate the absolute difference between the brick temperatures of the first brick object and the second brick object, and use it as the edge temperature difference value of the undirected connection edge; repeat the above process to obtain the edge gap value and edge temperature difference value of all undirected connection edges in the local thermal network. Collect the edge gap values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the first range; subtract the minimum value from the edge gap value of each undirected connected edge, and divide by the first range to obtain the normalized gap degree of the edge. Collect the edge temperature difference values ​​of all undirected connected edges in the local thermal network, find the maximum and minimum values; calculate the difference between the maximum and minimum values ​​as the second range; subtract the minimum value from the edge temperature difference value of each undirected connected edge, and divide by the second range to obtain the normalized temperature difference value of the edge. Based on preset weighting coefficients, the normalized gapness and normalized temperature difference are weighted and summed to obtain the comprehensive barrier value; the result of the exponential function is calculated with the natural constant as the base and the negative of the comprehensive barrier value as the exponent, which is used as the heat flux of each undirected connection edge in the local thermal network.

8. A method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 6, characterized in that, The step of performing a maximum flow algorithm on a local thermal network based on heat flux to determine the heat leakage from candidate bricks to the network boundary includes: Use the heat flux of each undirected edge in the local thermal network as the initial capacity of the corresponding edge; initialize the total flow to zero; Repeat the path finding iterative process until no new feasible path can be found; the iterative process includes: searching for a connected path from the source to the sink in the local hot network; if the current capacity of each edge on the connected path is greater than zero, the connected path is recorded as a feasible path; If a feasible path exists, take the minimum current capacity of all edges on the feasible path as the component flow of the feasible path, and add the component flow to the total flow; subtract the component flow from the current capacity of each edge on the feasible path, and increase the capacity of the reverse edge of each edge on the feasible path by the component flow; if no feasible path can be found, terminate the iteration. After the iteration terminates, the accumulated total flow is used as the heat leakage from the candidate brick to the network boundary.

9. A method for unmanned aerial vehicle (UAV) inspection of potential hazards in the exterior walls of high-rise buildings according to claim 8, characterized in that, The isolation degree determination process includes: For each candidate brick, obtain the temperature difference index and heat leakage of the candidate brick; Calculate the sum of heat leakage and the preset smoothing factor, and use it as the third factor; The ratio of the temperature difference index to the third factor is calculated as the isolation degree of the candidate bricks.

10. A drone inspection device for potential hazards in the exterior walls of high-rise buildings, characterized in that, The apparatus includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 9.

Images