Intelligent defect detection method and system for building facade

Abstract

The application relates to the field of building detection, in particular to an intelligent defect detection method and system for a building facade. An intelligent defect detection method for a building facade comprises the following steps: S1: collecting a visible light image sequence, an infrared image sequence and environment data of a building facade of an old residential area; S2: processing the visible light image sequence to extract stain feature data of the building facade; S3: calculating a structural defect index of the building facade according to the infrared image sequence and the environment data; S4: calculating an appearance defect index of the building facade according to the stain feature data and the environment data; S5: obtaining a defect detection result according to the structural defect index and the appearance defect index, and generating a defect distribution space positioning map of the building facade based on the defect detection result. The application realizes accurate, automatic differentiation and visual target positioning of appearance stains and hidden structural defects of the building facade.

Description

Technical Field

[0001] This invention relates to the field of building inspection, and more particularly to an intelligent defect detection method and system for building facades. Background Technology

[0002] With the continuous advancement of urbanization in my country, a large number of old residential communities are gradually entering a stage of functional degradation and aging. Building facades are subjected to multiple factors such as long-term erosion by wind and rain, temperature fluctuations, atmospheric pollutant adhesion, and material fatigue. Surface stains, coating peeling, cracks, hollow areas, and even internal structural damage are becoming increasingly common. These defects not only seriously affect the urban landscape and living quality but also pose a potential threat to public safety. Currently, the main methods for facade inspection are still manual visual inspection and auxiliary tapping and sound listening. Some projects have begun to introduce drones equipped with visible light cameras or infrared thermal imagers for aerial inspection. However, these technologies are mostly designed for specific types of defects and lack the ability to distinguish and jointly identify external stains and structural damage. Therefore, there is an urgent need for an intelligent defect detection method and system for building facades. By integrating visible light and infrared multi-source images and environmental data, an innovative dual-quantitative model of stain penetration depth and heat dissipation can be established. Based on the joint judgment of the two indices, a spatial positioning map can be generated to achieve accurate, automatic distinction and visual targeted positioning of external facade stains and hidden structural defects. Summary of the Invention

[0003] To overcome the shortcomings of the difficulty in automatically distinguishing between appearance defects and stains in the inspection of old building facades, and the lack of unified quantitative standards leading to inaccurate defect location, this invention provides an intelligent defect detection method for building facades.

[0004] The technical implementation scheme of the present invention is as follows: an intelligent defect detection method for building facades, comprising the following steps: S1: Collect visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings; S2: Process the visible light image sequence to extract stain feature data of the facade; S3: Calculate the structural defect index of the facade based on the infrared image sequence and the environmental data; S4: Calculate the appearance defect index of the facade based on the stain characteristic data and environmental data; S5: Obtain defect detection results based on the structural defect index and the appearance defect index, and generate a spatial location map of defect distribution on the facade based on the defect detection results.

[0005] Preferably, the collection of visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings includes: the visible light image sequence being a sequence of visible light images sorted by facade height, the infrared image sequence being a sequence of infrared images sorted by facade height, and the environmental data including wind speed corresponding to facade height, ambient atmospheric particulate matter concentration, and cumulative rainfall.

[0006] Preferably, the process of processing the visible light image sequence to extract stain feature data of the facade includes: obtaining the initial color, apparent thermal diffusivity, saturation penetration depth, and critical cleanable depth threshold of the facade material through standard experiments; segmenting the visible light image sequence using a pre-trained semantic segmentation network model to obtain stain patches; calculating the area of ​​each stain patch using the Monte Carlo algorithm; performing color space analysis on the stain patches; calculating the color heterogeneity index of the stain patches in combination with the initial color of the facade material; performing grayscale texture analysis on the stain patches to calculate the texture roughness of the stain patches; calculating the penetration depth estimate of the stain patches based on the color heterogeneity index, texture roughness, and saturation penetration depth of the facade material; and constructing stain feature data based on the area of ​​the stain patches, the penetration depth estimate, the apparent thermal diffusivity, and the critical cleanable depth threshold of the facade material.

[0007] Preferably, the step of calculating the penetration depth estimate of the stain patch based on the color heterogeneity index of the stain patch, the texture roughness of the stain patch, and the saturation penetration depth of the facade material includes: using the texture roughness of the largest stain patch as a reference texture roughness and using the color heterogeneity index of the largest stain patch as a reference color heterogeneity index. The formula for penetration depth is: ; In the formula, For the first Estimated penetration depth of each stain patch. For the first Saturation penetration depth of facade-like materials These are the weighting coefficients. For the first The color heterogeneity index of individual stain patches, For reference to the color heterogeneity index, For the first The texture roughness of each stain patch, For reference texture roughness.

[0008] Preferably, the step of calculating the structural defect index of the facade based on the infrared image sequence and the environmental data includes: setting a preset grid size, performing gridding processing on the infrared image sequence according to the grid size, obtaining the pixel temperature of each grid, selecting the center position of each grid as the grid position index, calculating the surface temperature gradient of the corresponding grid based on the pixel temperature of each grid, obtaining the thermal diffusivity of the air, and calculating the structural defect index of each grid based on the thermal flux dissipation operator.

[0009] Preferably, the calculation of the structural defect index for each grid based on the thermal flux divergence operator includes: The formula for structural defects is: ; In the formula, For position Structural defect index at the location. The apparent thermal diffusivity of the facade material is given. For heat flux divergence operator, For position Surface temperature gradient at that location The wind speed corresponds to the height of the facade. This is the air-cooled attenuation coefficient. The height of the facade. The thermal diffusivity of air.

[0010] Preferably, the step of calculating the appearance defect index of the facade based on the stain characteristic data and environmental data includes: calculating the average monthly rainfall in the area where the old residential community is located as a reference rainfall, calculating the average monthly ambient atmospheric particulate matter concentration in the area where the old residential community is located as a standard atmospheric particulate matter concentration, selecting the area of ​​the largest stain patch as the area of ​​the reference stain patch, and calculating the appearance defect index by combining the area of ​​the stain patch, the estimated penetration depth of the facade material corresponding to the stain patch, the critical cleanable depth threshold of the facade material corresponding to the stain patch, the ambient atmospheric particulate matter concentration, and the cumulative rainfall.

[0011] Preferably, the calculation of the appearance defect index by combining the area of ​​the stain patch, the estimated penetration depth of the corresponding exterior facade material, the critical cleanable depth threshold of the corresponding exterior facade material, the concentration of ambient atmospheric particulate matter, and the cumulative rainfall includes: The formula for appearance defects is: ; In the formula, For the first The appearance defect index of each stain patch, For reference, the area of ​​the stain, Area-weighted index For the first The area of ​​each stain patch For the first The estimated penetration depth of each stain patch on the exterior facade material. For the first The critical cleanable depth threshold of the facade material corresponding to each stain patch. For environmental catalysis weighting coefficients, This refers to the concentration of particulate matter in the ambient atmosphere. Standard atmospheric particulate matter concentration, For cumulative rainfall, For reference rainfall.

[0012] Preferably, the step of obtaining defect detection results based on the structural defect index and the appearance defect index, and generating a spatial location map of the facade defect distribution based on the defect detection results, includes: setting a preset structural defect index threshold and an appearance defect index threshold; comparing the structural defect index of each grid with the structural defect index threshold; if the structural defect index of a grid is greater than or equal to the structural defect index threshold, it is marked as a structural defect; otherwise, it is not marked; performing nearest neighbor analysis on the grids marked as structural defects, merging adjacent grids with structural defects to construct a structural defect distribution location map; comparing the appearance defect index of each stain patch with the appearance defect index threshold; if the appearance defect index of a stain patch is greater than or equal to the appearance defect index threshold, it is marked as an appearance defect; otherwise, it is not marked; statistically analyzing the stain patches marked as appearance defects to construct an appearance defect distribution location map; comparing the structural defect distribution location map and the appearance defect distribution location map, uniformly marking overlapping defects as structural defects to generate a defect distribution spatial location map.

[0013] Preferably, an intelligent defect detection system for building facades further includes: The data acquisition module collects visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings. The feature extraction module processes the visible light image sequence to extract stain feature data from the exterior facade; The structural defect calculation module calculates the structural defect index of the facade based on the infrared image sequence and the environmental data. The appearance defect calculation module calculates the appearance defect index of the facade based on the stain characteristic data and environmental data. The defect detection module obtains defect detection results based on the structural defect index and the appearance defect index, and generates a spatial location map of the defect distribution on the facade based on the defect detection results.

[0014] The present invention has the following advantages: This invention targets visible light image processing, creatively employing a semantic segmentation network combined with a Monte Carlo algorithm to accurately extract the area of ​​stain patches, and deeply mining the color heterogeneity index and texture roughness of the patches to estimate the depth of stain penetration into the exterior facade material. It transforms the subjective severity of stains into an objective penetration depth value, and combines it with the material's critical cleanable depth threshold, enabling the appearance defect index to clearly distinguish between cleanable stains on the interface and permanent damage that has penetrated the material.

[0015] This invention performs fine-grained meshing processing on infrared image sequences, innovatively introduces a heat flow divergence operator to calculate heat flow anomalies, and uses the apparent thermal diffusivity of the facade material to weight the temperature gradient. It also introduces a wind-cooling attenuation coefficient and a wind speed correction term, fully considering the interference of airflow on heat conduction in the actual outdoor environment. This effectively eliminates the influence of weather factors on the misjudgment of infrared detection, and can keenly capture minute heat flow anomalies caused by voids, cracks, or failure of the adhesive layer. It achieves highly sensitive and robust quantitative assessment of hidden structural defects that are not visible to the naked eye.

[0016] This invention constructs a dual-track parallel defect screening mechanism by calculating the appearance defect index and the structural defect index separately and setting independent judgment thresholds; it performs nearest neighbor analysis on the marked defect grid and prioritizes marking defects in overlapping areas as more dangerous structural defects, generating a spatial location map of defect distribution, which maps abstract index data to the real three-dimensional coordinates of the facade, providing precise target construction guidance for subsequent maintenance and reinforcement. Attached Figure Description

[0017] Figure 1 This is a flowchart of an intelligent defect detection method for building facades according to the present invention. Figure 2 This is a structural diagram of an intelligent defect detection system for building facades according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] Example 1: An intelligent defect detection method for building facades, such as Figure 1 As shown, it includes the following steps: S1: Collect visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings; The visible light image sequence is a sequence of visible light images ordered by the height of the facade, and the infrared image sequence is a sequence of infrared images ordered by the height of the facade. The environmental data includes wind speed corresponding to the facade height, ambient atmospheric particulate matter concentration, and cumulative rainfall.

[0020] It should be further explained that, before conducting on-site surveys of the old residential communities to be inspected to determine the building height, facade area, and orientation information, a general inspection of the building's facade is carried out before data collection. A laser rangefinder is used to measure the building's total height, floor height, and facade width. Based on the actual building height and floor height, the facade is divided into multiple continuous height segments along the vertical direction. Each height segment corresponds to a specific height interval, with the floor height as the length of the height interval. After the division, each segment becomes an independent work unit for subsequent image collection.

[0021] For the acquisition of visible light image sequences, a drone equipped with a high-resolution visible light camera was used as the acquisition tool. The drone flew or climbed along the building facade from bottom to top, hovering at each height segment and triggering the camera to capture high-definition visible light images of the corresponding segment. During the shooting process, the camera angle was adjusted so that the camera optical axis was basically perpendicular to the facade surface. Images taken at adjacent height positions maintained a 30% to 50% overlap area. The overlap area was used for subsequent image stitching and accurate calibration of the spatial location of defects. The above method was used to acquire multiple visible light images covering all height segments from the bottom to the top of the building. After acquiring multiple visible light images, all images were arranged in ascending order according to their corresponding height values, that is, from the lowest point to the highest point, forming a visible light image sequence ordered by the height of the facade. Each image in the sequence has a corresponding height label, which is automatically recorded by the altitude sensor carried by the drone during shooting.

[0022] For the acquisition of infrared image sequences, a drone equipped with an infrared thermal imager was used as the acquisition tool. After the acquisition of visible light images, infrared image acquisition was carried out to ensure that the acquisition time of visible light images and infrared images at the same height position was close to ensure the consistency of external environmental conditions. The infrared thermal imager moved along the facade from bottom to top segment by segment, stopping and taking pictures at each height segment. The angle of the thermal imager was adjusted so that the optical axis was basically perpendicular to the facade surface. Before acquisition, the emissivity parameters of the thermal imager were set according to the actual material type of the facade. During the shooting, the specific height value corresponding to each frame of infrared image was recorded. After the infrared images of all height segments were acquired, they were also arranged in ascending order according to the height value corresponding to each frame of image to form an infrared image sequence sorted by facade height. The visible light image sequence and the infrared image sequence adopted the same sorting rules, so that the visible light images and infrared images at the same height position could be directly called by each other through the height index, which facilitated subsequent joint interpretation and analysis.

[0023] Environmental data collection included measurements of wind speed corresponding to the building facade height, measurements of atmospheric particulate matter concentration, and cumulative rainfall statistics. Wind speed measurements were performed by deploying wind speed sensors at various heights corresponding to different data collection sections on the building facade. The installation height of each sensor was consistent with the height division of the aforementioned data collection sections. Wind speed values ​​at each height were read simultaneously with visible light and infrared image acquisition. Atmospheric particulate matter concentration measurements were performed by setting up an atmospheric particulate matter monitor in an open area surrounding the building. The concentration was continuously monitored and recorded throughout the image acquisition period, with the unit of measurement being micrograms per cubic meter. Cumulative rainfall statistics were based on records from the meteorological station in the area where the old residential community was located, calculating the total rainfall prior to the detection date, with the unit of measurement being millimeters.

[0024] S2: Process the visible light image sequence to extract stain feature data of the facade; The initial color, apparent thermal diffusivity, saturation penetration depth, and critical cleanable depth threshold of the facade material were obtained through standard experiments. A pre-trained semantic segmentation network model was used to segment the visible light image sequence to obtain stain patches. The area of ​​each stain patch was calculated using the Monte Carlo algorithm. Color space analysis was performed on the stain patches, and the color heterogeneity index of the stain patches was calculated based on the initial color of the facade material. Gray-scale texture analysis was performed on the stain patches to calculate their texture roughness. Based on the color heterogeneity index, texture roughness, and saturation penetration depth of the facade material, the penetration depth estimate of the stain patches was calculated. Stain feature data was constructed based on the stain patch area, penetration depth estimate, apparent thermal diffusivity, and critical cleanable depth threshold of the facade material.

[0025] The texture roughness of the largest stain patch is used as the reference texture roughness, and the color heterogeneity index of the largest stain patch is used as the reference color heterogeneity index. The formula for penetration depth is: ; In the formula, For the first Estimated penetration depth of each stain patch. For the first Saturation penetration depth of facade-like materials These are the weighting coefficients. For the first The color heterogeneity index of individual stain patches, For reference to the color heterogeneity index, For the first The texture roughness of each stain patch, For reference texture roughness.

[0026] It should be further explained that the basic physical properties of the building facade material were obtained in advance through standard experiments. The specific procedures for these standard experiments are as follows: standard specimens were prepared using materials of the same batch number and specifications as the building facade to be tested. These samples or specimens underwent systematic physical property testing in the laboratory. The original colorimetric value of the material samples was measured using a colorimeter as the initial color of the facade material. This initial color represents the material's natural color in a brand-new or clean state. The thermal properties of the material were measured using the laser flash method to obtain the apparent thermal diffusivity of the facade material. This coefficient reflects the rate of temperature field change when the material is heated. The material samples were immersed in a standard contamination solution, and the penetration depth was measured at different immersion times. A curve showing the penetration depth versus time was plotted until the penetration depth no longer increased. The maximum penetration depth after stabilization is the saturated penetration depth of the facade material. Multiple sets of material specimens with artificial stains of different depths were prepared, and each specimen was cleaned using standard cleaning processes. The maximum stain depth that restored the color difference of the material surface to more than 95% of the initial color after cleaning was determined as the critical cleanable depth threshold of the facade material.

[0027] After obtaining the aforementioned basic physical property parameters, the visible light image sequence is processed. Each frame of the visible light image sequence is input into a pre-trained semantic segmentation network model. The semantic segmentation network model adopts a deep convolutional neural network structure and is pre-trained using a labeled facade image dataset. The training samples contain pixel-level annotations of various types of stain areas, material areas, and background areas, enabling the model to accurately identify stain pixel areas and non-stain pixel areas in the input image. After the model completes inference for each frame of the visible light image, it outputs a pixel-level classification result image with the same size as the input image. The continuous pixel areas classified as stains in the classification result image are the stain patches. After all frames of images are segmented, all stain patches on the building facade are obtained.

[0028] After obtaining the stain patches, the area of ​​each stain patch is calculated. Since the stain patches appear as irregular polygons or arbitrarily shaped closed regions in the image, their area cannot be directly calculated using simple geometric formulas. Therefore, the Monte Carlo algorithm is used to numerically solve the area of ​​each stain patch. The specific method of the Monte Carlo algorithm to solve the area is to randomly place a large number of uniformly distributed sampling points within the bounding rectangle of each stain patch, count the ratio of the number of sampling points falling inside the outline of the stain patch to the total number of sampling points, and then multiply it by the actual area of ​​the bounding rectangle. This product is the estimated area of ​​the stain patch. Color space analysis is performed on each stain patch. The pixel values ​​of all pixels in the stain patch area are converted from the RGB color space to the CIELab uniform color space. The average chromaticity value of all pixels in the stain patch area is calculated in the CIELab color space. This average value is then compared with the chromaticity value of the initial color of the facade material obtained in the aforementioned standard experiment in the CIELab space to calculate the color difference between the two. This color difference is the color heterogeneity index of the stain patch. The larger the color heterogeneity index, the more significant the difference between the stain patch and the original color of the material in the clean state.

[0029] When performing grayscale texture analysis on stain patches, the color image of the stain patch area is converted into a grayscale image. After conversion, each pixel retains only one grayscale value, which typically ranges from 0 to 255. After conversion, the grayscale co-occurrence matrix (GCMM) method is used to perform statistical calculations on the texture features of the grayscale image. The calculation process of the GCMM method is as follows: the joint distribution frequency of grayscale values ​​of any two pixels in the grayscale image occurring simultaneously under a specific spatial relationship is statistically analyzed. A spatial displacement direction and displacement step size are pre-defined, such as a horizontal displacement of one pixel, a vertical displacement of one pixel, or a diagonal displacement of one pixel. For all pixels in the image, the grayscale value combinations of each pixel and its adjacent pixels after displacement in the set direction are statistically analyzed. The number of occurrences of each grayscale value combination on all pixels is recorded, and the occurrence counts of all grayscale value combinations are arranged into a two-dimensional matrix, which is the grayscale co-occurrence matrix.

[0030] After obtaining the gray-level co-occurrence matrix (GLCM), several second-order statistics are calculated based on this matrix. The contrast statistic reflects the degree to which non-zero elements in the GLCM deviate from the diagonal; a high contrast value indicates a large difference in gray levels between adjacent pixels, while a low contrast value indicates a smooth change in gray levels. The correlation statistic reflects the linear dependence of elements in the GLCM along the row and column directions; a high correlation value indicates a clear directional texture. The energy statistic reflects the sum of squares of the values ​​of each element in the GLCM; a high energy value indicates a uniform texture, while a low energy value indicates a chaotic texture. The weighted sum of the contrast, correlation, and energy statistics according to preset weights yields the texture roughness, which reflects the characteristics of stains and patches. The determination of texture roughness weights is divided into three levels: target layer, criterion layer, and scheme layer, considering the drastic changes in surface grayscale and structural complexity. The target layer determines the comprehensive weights for texture roughness. The criterion layer includes various statistics participating in the weighted summation, such as contrast, correlation, and energy statistics. The scheme layer determines the final weight values ​​for each statistic. A judgment matrix is ​​constructed, and experts in building inspection and image processing are invited to compare the importance of any two statistics in the criterion layer relative to texture roughness evaluation pairwise. A nine-level scaling method is used, where scale values ​​from one to nine represent the importance of two elements from equal to absolutely important, with the reciprocal indicating a reverse comparison. All pairwise comparison results are arranged into a square matrix, forming the judgment matrix. The weight vector is calculated using the eigenvalue method on the constructed judgment matrix. The mathematical process is as follows: the largest eigenvalue of the judgment matrix and its corresponding eigenvector are obtained, and each element in the eigenvector is divided by the sum of all elements in the vector for normalization. The normalized eigenvector is the weight allocation result for each statistic. A consistency check is performed, and a consistency index is calculated. This index is the difference between the largest eigenvalue and the order of the judgment matrix, divided by the order minus one. Then, a consistency ratio is calculated, which is the ratio of the consistency index to the random consistency index. When the consistency ratio is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency, and the calculated weight vector is acceptable. If the consistency ratio is greater than or equal to 0.1, it indicates a contradiction in the judgment matrix. The process is returned to experts to re-perform pairwise comparisons and adjust the values ​​of the judgment matrix until the consistency ratio meets the requirements.

[0031] After obtaining the color heterogeneity index and texture roughness of all stain patches, the stain patch with the largest texture roughness value is identified from all stain patches, and this largest texture roughness is set as the reference texture roughness; the stain patch with the largest color heterogeneity index value is identified from all stain patches, and this largest color heterogeneity index is set as the reference color heterogeneity index. The reference texture roughness and the reference color heterogeneity index are used as normalized benchmark values ​​in subsequent penetration depth estimation calculations.

[0032] For each stain patch, the estimated penetration depth is calculated by combining the stain patch's color heterogeneity index, texture roughness, the saturation penetration depth corresponding to the type of facade material in the area where the stain patch is located, and the aforementioned reference color heterogeneity index and reference texture roughness. The calculation logic of the penetration depth formula is as follows: Divide the stain patch's color heterogeneity index by the reference color heterogeneity index to obtain the color deviation ratio; divide the stain patch's texture roughness by the reference texture roughness to obtain the texture deviation ratio; multiply the color deviation ratio and texture deviation ratio by a weighting coefficient bound to this type of facade material to obtain a comprehensive penetration factor; then, take the negative value of the comprehensive penetration factor and perform exponential operation with the natural constant e as the base to obtain an exponential decay term; subtract the exponential decay term from one to obtain a penetration saturation coefficient between zero and one; multiply the penetration saturation coefficient by the saturation penetration depth of this type of facade material; the product is the estimated penetration depth of the stain patch. In the formula, the weighting coefficient is a coefficient bound to the color heterogeneity index, texture roughness, and texture roughness. The empirical coefficients tied to material type are determined through experimental calibration. For each type of facade material, multiple sets of standard stain samples with known penetration depths are prepared. The actual penetration depth of each set of samples is measured experimentally, along with the color heterogeneity index and texture roughness. The experimentally measured actual penetration depth is used as the target value, and the color heterogeneity index and texture roughness are used as input variables. A nonlinear regression method is employed to fit the parameters of the formula. During the fitting process, the sum of squared errors between the estimated penetration depth and the actual measured value is minimized, thereby determining the weight coefficient values ​​that achieve the best fitting effect. The weight coefficients for different material types are calibrated independently. After calculating the penetration depth estimates for all stain patches, corresponding stain feature data is constructed for each stain patch. The stain feature data for each stain patch includes the area of ​​the stain patch, the estimated penetration depth of the stain patch, the apparent thermal diffusivity of the facade material in the area where the stain patch is located, and the critical cleanable depth threshold of the facade material in the area where the stain patch is located.

[0033] S3: Calculate the structural defect index of the facade based on the infrared image sequence and the environmental data; The infrared image sequence is meshed according to a preset grid size to obtain the pixel temperature of each grid. The center position of each grid is selected as the grid position index. The surface temperature gradient of the corresponding grid is calculated based on the pixel temperature of each grid. The thermal diffusivity of the air is obtained. The structural defect index of each grid is calculated based on the thermal flux dissipation operator.

[0034] The formula for structural defects is: ; In the formula, For position Structural defect index at the location. The apparent thermal diffusivity of the facade material is given. For heat flux divergence operator, For position Surface temperature gradient at that location The wind speed corresponds to the height of the facade. This is the air-cooled attenuation coefficient. The height of the facade. The thermal diffusivity of air.

[0035] It should be further explained that before calculating the structural defect index, a grid size is preset. The grid size is set based on the spatial resolution of the infrared image and the positioning accuracy required for detecting defects on the building facade. The grid size is set to a range of 5 to 20 centimeters square, corresponding to the actual physical dimensions. After setting the grid size, each frame of the infrared image sequence is gridded. The entire infrared image is divided into several non-overlapping regular grid units according to the preset grid size. Each grid unit corresponds to a specific area on the facade surface in space. Each grid unit contains multiple pixels. The arithmetic mean of the temperature values ​​of all pixels falling within the grid unit is taken to obtain the pixel temperature of the grid unit. Each frame of the infrared image sequence is gridded in the above manner to obtain the pixel temperature of each grid.

[0036] After meshing, the center position of each mesh cell is selected as its spatial index. The two-dimensional coordinates of the mesh center position are determined by the horizontal position number and vertical height number of the mesh on the facade plane. These coordinates are the position identifier of the mesh in the facade space. For each mesh, based on the pixel temperature values ​​of the mesh itself and its adjacent meshes, the rate of temperature change in both the horizontal and vertical directions is calculated using the finite difference method to obtain the surface temperature gradient at the location of the mesh. The surface temperature gradient reflects the degree of drastic change in the surface temperature of the facade in space; a larger temperature gradient value indicates a significant temperature abrupt change at that location.

[0037] The thermal diffusivity of air is obtained. The thermal diffusivity of air is an inherent thermal property parameter of air. It is obtained from the engineering thermophysics handbook based on the ambient temperature and air compression conditions at the time of detection. The structural defect index of each grid is calculated based on the heat flux divergence operator. The heat flux divergence operator is a spatial differential operator used to describe the degree of divergence or convergence of heat flux density in space. The physical meaning of this operator is: first, calculate the product of the apparent thermal diffusivity of the facade material and the surface temperature gradient. This product represents the heat flux density vector at that location. Then, calculate the spatial divergence of this heat flux density vector, that is, calculate the sum of the rates of change of heat flux density in the horizontal and vertical directions. Finally, take the absolute value of the divergence value. This absolute value reflects the degree of non-uniformity of heat flux in space at that location.

[0038] The specific calculation method of the structural defect index is as follows: For a grid with a specific coordinate, the apparent thermal diffusivity of the facade material at that grid location is multiplied by the surface temperature gradient at that grid location to obtain a heat flux density vector. Then, the heat flux density vector is applied to the heat flux divergence operator to obtain the spatial divergence and the absolute value is taken to obtain the basic value of the heat flux divergence at that location. The wind-cooling attenuation factor is constructed by multiplying the facade height by the corresponding wind speed and then dividing by the air thermal diffusivity to obtain a dimensionless ratio. This dimensionless ratio is multiplied by the aforementioned wind-cooling attenuation factor to obtain a product term. Then, 1 is added to this product term to form a denominator term. The reciprocal of this denominator term is the wind-cooling attenuation factor. Finally, the aforementioned basic value of heat flux dissipation is multiplied by the wind-cooling attenuation factor, and the resulting product is the structural defect index at that grid location. The wind-cooling attenuation coefficient in the formula is determined by experimental calibration. The physical meaning of this coefficient is the degree of interference of wind speed on the infrared thermographic temperature measurement results. Infrared thermographic images of the building facade with the same known structural state are collected under different wind speed conditions. The basic value of heat flux dissipation under windless correction is calculated and compared with the known structural state of the facade. The wind-cooling attenuation coefficient value that has the strongest correlation with the known structural state is obtained by fitting the least squares method.

[0039] The underlying physical principle of the above calculation process is as follows: If structural defects such as hollow areas, cracks, detached adhesive layers, or loose materials exist inside the building facade, these defects will alter the heat conduction path and efficiency within the wall. The apparent thermal diffusivity of the defective areas differs from that of the intact areas, leading to localized abnormal heating or cooling zones in the facade surface temperature field. These abnormal zones manifest as drastic changes in temperature gradient and a significant increase in heat flux. The heat flux operator can effectively capture these abnormal heat flux changes. External wind cooling effects alter the convective heat transfer conditions of the facade surface; the higher the wind speed, the faster the surface dissipates heat. The surface temperature measured by the infrared thermal imager is more significantly affected by wind speed. Therefore, in the structural... The structural defect index calculation incorporates a wind-cooling attenuation factor to correct for wind speed interference, making the calculated structural defect indices under different wind speed conditions comparable. The facade height appears as a characteristic length in the wind-cooling attenuation factor because the wind speed distribution is different at different heights of the building facade. The wind speed increases with height, and the wind-cooling effect is more significant at higher locations than at lower locations. Including the facade height in the correction term makes the calculation of the structural defect index more consistent with actual physical laws. The thermal diffusivity of air reflects the ability of air to conduct heat. The cooling effect of wind speed on the surface is closely related to the thermal diffusivity of the air itself. Therefore, the thermal diffusivity of air also appears in the denominator of the wind-cooling attenuation factor.

[0040] After calculating the structural defect index for each grid in the infrared image sequence in the manner described above, the structural defect indices of all grids constitute the spatial distribution data of the structural defect indices of the entire building facade. This data is used in subsequent steps to compare with the structural defect index threshold to determine which grids have structural defects.

[0041] S4: Calculate the appearance defect index of the facade based on the stain characteristic data and environmental data; The average monthly rainfall in the area where the old residential community is located is used as a reference rainfall, and the average monthly ambient atmospheric particulate matter concentration in the area is used as the standard atmospheric particulate matter concentration. The area of ​​the largest stain is selected as the area of ​​the reference stain. The appearance defect index is calculated by combining the area of ​​the stain, the estimated penetration depth of the corresponding exterior facade material, the critical cleanable depth threshold of the corresponding exterior facade material, the ambient atmospheric particulate matter concentration, and the cumulative rainfall.

[0042] The formula for appearance defects is: ; In the formula, For the first The appearance defect index of each stain patch, For reference, the area of ​​the stain, Area-weighted index For the first The area of ​​each stain patch For the first The estimated penetration depth of each stain patch on the exterior facade material. For the first The critical cleanable depth threshold of the facade material corresponding to each stain patch. For environmental catalysis weighting coefficients, This refers to the concentration of particulate matter in the ambient atmosphere. Standard atmospheric particulate matter concentration, For cumulative rainfall, For reference rainfall.

[0043] It should be further explained that before calculating the appearance defect index, the average monthly rainfall in the area where the old residential community is located is calculated and used as a reference rainfall. Rainfall records for each month over the past three or five consecutive years are obtained from the meteorological observation station in the area. The rainfall for all months is summed up and divided by the total number of months to obtain the average monthly rainfall, which is the reference rainfall. The average monthly atmospheric particulate matter concentration in the area where the old residential community is located is calculated and used as the standard atmospheric particulate matter concentration. Daily atmospheric particulate matter concentration monitoring data for the past three or five consecutive years are obtained from the environmental monitoring station in the area. The daily average value is summed up and divided by the total number of days to obtain the daily average value. The daily average value is then multiplied by 30 to obtain an approximate monthly atmospheric particulate matter concentration, which is the standard atmospheric particulate matter concentration. The largest stain patch is identified from all stain patches, and its area is used as the area of ​​the reference stain patch.

[0044] After completing the above three preliminary statistical tasks, for each stain patch, the appearance defect index of the stain patch is calculated by combining the area of ​​the stain patch, the estimated penetration depth of the corresponding facade material, the critical cleanable depth threshold of the corresponding facade material, the ambient atmospheric particulate matter concentration and the cumulative rainfall collected above.

[0045] The Appearance Defect Index formula is used to calculate the negative impact of the i-th stain patch on the appearance quality of the facade. The formula is obtained by multiplying three parts: area factor, penetration factor, and environmental factor. The area factor is calculated as follows: divide the area of ​​the stain patch by the area of ​​a reference stain patch to obtain an area ratio, then perform an exponential operation on the area ratio, with the exponent being the area weight index. The penetration factor is calculated as follows: divide the estimated penetration depth of the stain patch by the critical cleanable depth threshold to obtain a penetration ratio, take the negative value of the penetration ratio, perform an exponential operation, and then subtract the exponential result by 1. The environmental factor is calculated as follows: divide the ambient atmospheric particulate matter concentration by the standard atmospheric particulate matter concentration to obtain a particulate matter ratio, divide the cumulative rainfall by the reference rainfall to obtain a rainfall ratio, multiply the particulate matter ratio and the rainfall ratio, multiply by the environmental catalytic weight coefficient, and finally add the product by 1. In the formula, the appearance defect index formula... The area weight index and environmental catalytic weight coefficient in the formula are also determined by experimental calibration. The area weight index controls the sensitivity of the area ratio in the index calculation. The determination method is as follows: select multiple groups of stain patch samples with different areas but the same other conditions, have professional inspectors subjectively score the severity of appearance defects, and then perform regression fitting on the area ratio and subjective scores. The resulting index coefficient is the area weight index value. The environmental catalytic weight coefficient controls the overall influence of environmental factors on the appearance defect index. The determination method is as follows: conduct long-term tracking observation of the same batch of stain patches under different atmospheric particulate matter concentrations and different cumulative rainfall conditions, record the appearance change rate of the stain patches, and obtain the value of the environmental catalytic weight coefficient through multiple linear regression fitting. The appearance defect index is a dimensionless positive number. The larger the value, the more serious the negative impact of the stain patch on the appearance quality of the facade.

[0046] S5: Obtain defect detection results based on the structural defect index and the appearance defect index, and generate a spatial location map of defect distribution on the facade based on the defect detection results.

[0047] Preset thresholds for structural defect index and appearance defect index. Compare the structural defect index of each grid with the threshold. If the structural defect index of a grid is greater than or equal to the threshold, it is marked as a structural defect; otherwise, it is not marked. Perform nearest neighbor analysis on the grids marked as structural defects, and merge adjacent grids with structural defects to construct a structural defect distribution map. Compare the appearance defect index of each stain patch with the threshold. If the appearance defect index of a stain patch is greater than or equal to the threshold, it is marked as an appearance defect; otherwise, it is not marked. Statistically count the stain patches marked as appearance defects to construct an appearance defect distribution map. Compare the structural defect distribution map and the appearance defect distribution map, and mark overlapping defects as structural defects to generate a defect distribution spatial location map.

[0048] It should be further explained that, for the pre-set structural defect index threshold and appearance defect index threshold, a control area with the same material type as the building to be inspected and known to be free of structural defects is selected. The structural defect index of all grids in this control area is obtained according to the aforementioned structural defect index calculation method. The average value is calculated and double the standard deviation is added as the upper limit of normal fluctuation. Then, a verification area with known typical structural defects such as hollowness, cracks, or debonding is selected, and the minimum value of the structural defect index in this area is calculated as the lower limit of defect judgment. The upper limit of normal fluctuation and the lower limit of defect judgment are compared. When there is a significant gap between the two, the median value is taken as the preset threshold; when there is overlap, the lower limit of defect judgment is used as the preset threshold. The setting process for the appearance defect index threshold is based on the mapping relationship between subjective evaluation and objective index. A large number of stain and patch samples with different appearance degrees are collected, and the appearance defect index of each sample is calculated. Simultaneously, several professionals with building appearance inspection qualifications are invited to conduct blind evaluations and scores of the samples. Statistical regression analysis is performed on the calculated appearance defect index value and the subjective scores to determine the critical value of the appearance defect index corresponding to the change from acceptable to unacceptable in subjective evaluation. This critical value is the appearance defect index threshold. After the threshold is set, the determination and marking of structural defects begins. The structural defect index of each grid calculated above is compared with the preset structural defect index threshold one by one. For a certain grid, if the structural defect index of the grid is greater than or equal to the structural defect index threshold, it indicates that the degree of heat flow anomaly at the location of the grid has reached or exceeded the minimum standard for judging it as a structural defect, and the grid is marked as a structural defect grid. If the structural defect index of the grid is less than the structural defect index threshold, it indicates that the degree of heat flow anomaly at the location has not yet reached the judgment standard for structural defects, and the grid is not marked. After all grids are compared in the above manner, all grids marked as structural defects constitute the preliminary screening results of structural defects.

[0049] Nearest neighbor analysis is performed on the grids marked as structural defects. The specific procedure for nearest neighbor analysis is as follows: each grid marked as a structural defect is examined one by one, and its adjacent grids in the top, bottom, left, right, and diagonal directions are found to belong to the grid marked as a structural defect. If two grids marked as structural defects are spatially adjacent, they are classified into the same connected cluster. Through repeated iterative traversal of this adjacency relationship, all grids marked as structural defects are divided into several independent connected clusters according to spatial adjacency. Each connected cluster represents a continuous structural defect area on the facade. The spatial range covered by these connected clusters is extracted, that is, the outline boundaries of all grids in each connected cluster are drawn to form a structural defect distribution map. The structural defect distribution map uses the unfolded plan of the building facade as the base map, and the outline and position of each connected cluster are marked on the base map with specific colors or patterns to visually show the distribution of all structural defect areas on the facade.

[0050] The appearance defect index of each stain patch calculated above is compared with the preset appearance defect index threshold one by one. For a certain stain patch, if the appearance defect index of the stain patch is greater than or equal to the appearance defect index threshold, it indicates that the negative impact of the stain patch on the appearance quality of the facade has reached or exceeded the minimum standard for judging it as an appearance defect, and the stain patch is marked as an appearance defect patch. If the appearance defect index of the stain patch is less than the appearance defect index threshold, it indicates that the impact of the stain patch on the appearance of the facade has not reached the judgment standard for appearance defect, and the stain patch is not marked. After comparing all stain patches in the above manner, all stain patches marked as appearance defects constitute the screening results of appearance defects.

[0051] All stain patches marked as appearance defects are statistically analyzed, and the spatial location and outline boundary information of each appearance defect patch are extracted. The spatial location information of the appearance defect patch comes from the actual facade coordinates corresponding to the stain patch in the visible light image. These coordinates are obtained by mapping the visible light image and the spatial coordinates of the building facade through the calibration relationship. The outline boundaries of all stain patches marked as appearance defects are drawn on the plan development of the building facade to form an appearance defect distribution map. The appearance defect distribution map is based on the same base map, and the outline and location of each appearance defect patch are marked on the base map with a different color or pattern than the structural defects, which intuitively shows the distribution of all appearance defect areas on the facade.

[0052] After obtaining the structural defect distribution location map and the appearance defect distribution location map, the two are compared in space. The specific comparison method is as follows: under a unified building facade spatial coordinate system, check whether there is any overlap between the spatial range covered by each structural defect area in the structural defect distribution location map and the spatial range covered by each appearance defect area in the appearance defect distribution location map. If a certain location appears in both a structural defect area in the structural defect distribution location map and an appearance defect area in the appearance defect distribution location map, then the location belongs to the overlapping location of the two defect definitions.

[0053] For defects in these overlapping locations, they should be uniformly marked as structural defects. Structural defects involve the safety of building structures, and their danger and maintenance priority are far higher than those of simple appearance defects. When an area meets both the criteria for determining structural defects and the criteria for determining appearance defects, from the perspective of engineering safety, the area should be regarded as a structural defect and treated according to the level of structural defects to avoid underestimating safety risks due to misjudging structural hazards as appearance defects.

[0054] After the above overlap processing, areas that were originally appearance defects but did not overlap with structural defects are retained as appearance defect markers, and areas that were originally structural defects are all retained as structural defect markers. Areas where both types overlap are uniformly classified under structural defect markers. This results in a complete defect classification result that integrates both structural and appearance defects without overlap. This complete defect classification result is then plotted on the same building facade plan according to their respective spatial locations and outline boundaries, generating a spatial location map of defect distribution. The spatial location map of defect distribution uses two different marking methods to represent structural and appearance defects respectively. Each defect has clear spatial coordinates and outline boundaries, allowing inspection personnel to intuitively determine the specific location, distribution range, and severity of various defects on the facade, providing precise spatial guidance for subsequent repair and reinforcement scheme design and construction positioning.

[0055] Example 2: Based on Example 1, an intelligent defect detection system for building facades, such as... Figure 2 As shown, it also includes: The data acquisition module collects visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings. The feature extraction module processes the visible light image sequence to extract stain feature data from the exterior facade; The structural defect calculation module calculates the structural defect index of the facade based on the infrared image sequence and the environmental data. The appearance defect calculation module calculates the appearance defect index of the facade based on the stain characteristic data and environmental data. The defect detection module obtains defect detection results based on the structural defect index and the appearance defect index, and generates a spatial location map of the defect distribution on the facade based on the defect detection results.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent defect detection of building facades, characterized in that, Includes the following steps: S1: Collect visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings; S2: Process the visible light image sequence to extract stain feature data of the facade; S3: Calculate the structural defect index of the facade based on the infrared image sequence and the environmental data; S4: Calculate the appearance defect index of the facade based on the stain characteristic data and environmental data; S5: Obtain defect detection results based on the structural defect index and the appearance defect index, and generate a spatial location map of defect distribution on the facade based on the defect detection results.

2. The intelligent defect detection method for building facades as described in claim 1, characterized in that, The collection of visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings includes: the visible light image sequences being visible light image sequences sorted by facade height; the infrared image sequences being infrared image sequences sorted by facade height; and the environmental data including wind speed corresponding to facade height, ambient atmospheric particulate matter concentration, and cumulative rainfall.

3. The intelligent defect detection method for building facades as described in claim 1, characterized in that, The process of processing the visible light image sequence to extract stain feature data of the facade includes: obtaining the initial color, apparent thermal diffusivity, saturation penetration depth, and critical cleanable depth threshold of the facade material through standard experiments; segmenting the visible light image sequence using a pre-trained semantic segmentation network model to obtain stain patches; calculating the area of ​​each stain patch using the Monte Carlo algorithm; performing color space analysis on the stain patches; calculating the color heterogeneity index of the stain patches based on the initial color of the facade material; calculating the texture roughness of the stain patches through grayscale texture analysis; calculating the penetration depth estimate of the stain patches based on the color heterogeneity index, texture roughness, and saturation penetration depth of the facade material; and constructing stain feature data based on the area of ​​the stain patches, the estimated penetration depth, the apparent thermal diffusivity, and the critical cleanable depth threshold of the facade material.

4. The intelligent defect detection method for building facades as described in claim 3, characterized in that, The method of calculating the penetration depth estimate of a stain patch based on the color heterogeneity index of the stain patch, the texture roughness of the stain patch, and the saturation penetration depth of the facade material includes: using the texture roughness of the largest stain patch as the reference texture roughness and using the color heterogeneity index of the largest stain patch as the reference color heterogeneity index. The formula for penetration depth is: ； In the formula, For the first Estimated penetration depth of each stain patch. For the first Saturation penetration depth of facade-like materials These are the weighting coefficients. For the first The color heterogeneity index of individual stain patches, For reference to the color heterogeneity index, For the first The texture roughness of each stain patch, For reference texture roughness.

5. The intelligent defect detection method for building facades as described in claim 1, characterized in that, The step of calculating the structural defect index of the facade based on the infrared image sequence and the environmental data includes: setting a preset grid size, performing gridding processing on the infrared image sequence according to the grid size, obtaining the pixel temperature of each grid, selecting the center position of each grid as the grid position index, calculating the surface temperature gradient of the corresponding grid based on the pixel temperature of each grid, obtaining the thermal diffusivity of the air, and calculating the structural defect index of each grid based on the thermal flux dissipation operator.

6. The intelligent defect detection method for building facades as described in claim 5, characterized in that, The calculation of the structural defect index for each grid based on the thermal flux divergence operator includes: The formula for structural defects is: ； In the formula, For position Structural defect index at the location. The apparent thermal diffusivity of the facade material. For heat flux divergence operator, For position Surface temperature gradient at that location The wind speed corresponds to the height of the facade. This is the air-cooled attenuation coefficient. The height of the facade. The thermal diffusivity of air.

7. The intelligent defect detection method for building facades as described in claim 3, characterized in that, The calculation of the facade appearance defect index based on the stain characteristic data and environmental data includes: calculating the average monthly rainfall in the area where the old residential community is located as a reference rainfall, calculating the average monthly ambient atmospheric particulate matter concentration in the area where the old residential community is located as a standard atmospheric particulate matter concentration, selecting the area of ​​the largest stain patch as the area of ​​the reference stain patch, and calculating the appearance defect index by combining the area of ​​the stain patch, the estimated penetration depth of the facade material corresponding to the stain patch, the critical cleanable depth threshold of the facade material corresponding to the stain patch, the ambient atmospheric particulate matter concentration, and the cumulative rainfall.

8. The intelligent defect detection method for building facades as described in claim 7, characterized in that, The calculation of the appearance defect index, which combines the area of ​​the stain patch, the estimated penetration depth of the corresponding exterior facade material, the critical cleanable depth threshold of the corresponding exterior facade material, the concentration of ambient atmospheric particulate matter, and the cumulative rainfall, includes: The formula for appearance defects is: ； In the formula, For the first The appearance defect index of each stain patch, For reference, the area of ​​the stain, Area-weighted index For the first The area of ​​each stain patch For the first The estimated penetration depth of each stain patch on the exterior facade material. For the first The critical cleanable depth threshold of the facade material corresponding to each stain patch. For environmental catalysis weighting coefficients, This refers to the concentration of particulate matter in the ambient atmosphere. Standard atmospheric particulate matter concentration, For cumulative rainfall, For reference rainfall.

9. The intelligent defect detection method for building facades as described in claim 1, characterized in that, The step of obtaining defect detection results based on the structural defect index and the appearance defect index, and generating a spatial location map of the facade defect distribution based on the defect detection results, includes: setting a preset structural defect index threshold and an appearance defect index threshold; comparing the structural defect index of each grid with the structural defect index threshold; if the structural defect index of a grid is greater than or equal to the structural defect index threshold, it is marked as a structural defect, otherwise it is not marked; performing nearest neighbor analysis on the grids marked as structural defects, merging adjacent grids with structural defects to construct a structural defect distribution location map; comparing the appearance defect index of each stain patch with the appearance defect index threshold; if the appearance defect index of a stain patch is greater than or equal to the appearance defect index threshold, it is marked as an appearance defect, otherwise it is not marked; statistically analyzing the stain patches marked as appearance defects to construct an appearance defect distribution location map; comparing the structural defect distribution location map and the appearance defect distribution location map, uniformly marking overlapping defects as structural defects to generate a defect distribution spatial location map.

10. An intelligent defect detection system for building facades, used to implement the intelligent defect detection method for building facades as described in any one of claims 1-9, characterized in that it further includes... include: The data acquisition module collects visible light image sequences, infrared image sequences, and environmental data of the exterior facades of old residential buildings. The feature extraction module processes the visible light image sequence to extract stain feature data from the exterior facade; The structural defect calculation module calculates the structural defect index of the facade based on the infrared image sequence and the environmental data. The appearance defect calculation module calculates the appearance defect index of the facade based on the stain characteristic data and environmental data. The defect detection module obtains defect detection results based on the structural defect index and the appearance defect index, and generates a spatial location map of the defect distribution on the facade based on the defect detection results.

Images