Building engineering quality detection method and system based on image big data analysis

Abstract

This invention relates to the field of image data processing technology and discloses a method and system for inspecting the quality of building engineering based on image big data analysis. The method includes: acquiring digital image data of the building surface and sensor pose parameters; extracting the frequency domain response using a multi-scale logarithmic Gaussian filter to generate a phase consistency feature map with significant edge characteristics; quantifying the anisotropic features of pixels through eigenvalue decomposition to identify defect pixel sets; obtaining three-dimensional geometric parameters through projection transformation and mapping them to the model parameter space to reconstruct topological description data, generating geometric entity data containing physical fracture boundaries. This invention utilizes phase information to break the constraint of illumination contrast on feature extraction, ensuring detection stability under non-uniform illumination; distinguishing material texture and defect boundaries through geometric dimensions; establishing a correlation path between optical grid and vector topology; and achieving deep coupling between detection results and design models.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to a method and system for inspecting the quality of building engineering based on image big data analysis. Background Technology

[0002] Currently, in the process of building inspection, the automatic detection of concrete cracks or steel structure defects is achieved by acquiring images of the building surface and determining the pixel areas with gray-level gradient changes. This is a common method in the field. In the inspection environment of high-rise steel structures or large-volume concrete, the change in the angle of natural light projection causes non-uniform lighting shadows in the image. The gradient fluctuations in the gray-level space of pores or hydration patches on the concrete surface are similar to real defects. The intrinsic texture of the material and the boundary of structural failure are in an overlapping state, making it difficult for the extraction mechanism to separate the edges of small defects from the background noise.

[0003] The gradient distribution of a grayscale image depends on the light reflection intensity of the material surface. When the reflectivity of the material surface is uneven, the grayscale intensity-based processing introduces unstructured texture interference. When capturing microcracks by adjusting the segmentation threshold, false alarms may occur. In addition, there is a topological heterogeneity between optical raster data and computer-aided design models, and optical features cannot be directly converted into geometric parameters for mechanical evaluation. For example, Chinese invention patent application CN120259281A discloses a method and system for monitoring the construction quality of water conservancy projects based on image recognition. It uses the spatial correlation between pixel intensity and grayscale gradient to analyze crack characteristics. The underlying logic of the algorithm depends on the grayscale brightness gradient distribution. Building materials such as concrete surfaces naturally have pores and hydration patches. The gradient features generated in the grayscale domain are similar to those of real defects. Existing solutions lack analysis of image phase information and geometric anisotropy. Under non-uniform lighting interference, it is difficult to decouple the structural failure boundary from the intrinsic texture, resulting in a high false alarm rate. Furthermore, statistical analysis at the raster pixel level cannot solve the topological heterogeneity problem between optical pixel data and three-dimensional vector design models, and the detection results cannot be directly converted into geometric entity data for mechanical evaluation.

[0004] Therefore, how to reconstruct the image feature extraction mechanism, achieve the separation of illumination components and structural components in the grayscale matrix while maintaining the existing physical light source, and establish the correspondence between detection data and the topological mesh of the design model, has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a method for inspecting the quality of building construction projects based on image big data analysis, comprising the following steps:

[0006] Step 101: Acquire digital image data of the building surface and the pose parameters of the sensors;

[0007] Step 102: Extract the frequency domain response of digital image data using a multi-scale logarithmic Gaussian filter bank, and calculate the local energy based on the odd symmetric component and the even symmetric component to generate a phase consistency feature map with edge saliency features.

[0008] Step 103: Calculate the local structure tensor eigenvalues ​​of the phase consistency feature map, quantize the anisotropic feature values ​​of the pixels based on the polarity distribution of the first and second eigenvalues, so as to distinguish the intrinsic texture of the material from the defect boundary in the geometric gradient dimension and identify the defect pixel set.

[0009] Step 104: Based on the sensor's optical intrinsic parameter matrix and pose parameters, perform a spatial projection transformation on the defect pixel set to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system.

[0010] Step 105: Map the three-dimensional geometric parameters to the parameterized space of the three-dimensional structural model. By calculating the intersection constraints between the three-dimensional geometric parameters and the topological elements of the three-dimensional structural model, perform topological description data reconstruction in the database of the three-dimensional structural model to generate geometric entity data containing physical fracture boundaries.

[0011] Preferably, the process of generating the phase consistency feature map in step 102 includes: performing multi-scale convolution processing on the digital image data using a multi-scale log-Gaussian filter bank to obtain the extended complex response of the pixel at each scale; calculating the sum of local energy and the sum of local amplitude of the pixel using the real and imaginary parts of the extended complex response; introducing a preset noise power compensation parameter to perform offset correction on the sum of local energy, and determining the ratio of the corrected sum of local energy to the sum of local amplitude as the response value of the pixel in the phase consistency feature map.

[0012] Preferably, the response value is measured by the following phase consistency measure. Sure: Where W(x) is the frequency expansion weighting coefficient, This represents the local energy value at the nth scale. Let be the local amplitude value at the nth scale, T be the noise power compensation parameter, ϵ be a small constant to prevent the denominator from being zero, and ⌊⋅⌋ be an operator that takes a value of 0 when the value in the parentheses is negative.

[0013] Preferably, the process of identifying the defective pixel set in step 103 includes: constructing a local second-order moment matrix using a phase consistency feature map, obtaining a first eigenvalue representing the local structural evolution direction and a second eigenvalue perpendicular to it through eigenvalue decomposition; calculating the distribution ratio of the first eigenvalue and the second eigenvalue to obtain a linearity index; and when the linearity index exceeds a preset defect judgment threshold and the principal axis direction of its anisotropic feature has spatial continuity, classifying the pixel into the defective pixel set.

[0014] Preferably, the process of obtaining the three-dimensional geometric parameters of the defect boundary in the object coordinate system in step 104 includes: obtaining the 6-DOF pose parameters of the image acquisition sensor during sampling, constructing a homography mapping matrix from the pixel coordinate system to the object coordinate system; inputting the contour coordinates of the defect pixel set into the homography mapping matrix, and combining the preset surface depth constraints of the building component to output a defect boundary point cloud sequence with physical scale dimension.

[0015] Preferably, the process of performing topological description data reconstruction in step 105 includes: retrieving the target facets in the three-dimensional structural model that correspond to the three-dimensional geometric parameters, using an interpolation algorithm to fit the three-dimensional geometric parameters into an analytical curve in the surface parameter space; locating the intersection points of the analytical curves and the existing topological edges in the target facets, and dynamically inserting new topological vertices and topological half-edges into the database for the target facets.

[0016] Preferably, after dynamically inserting new topological vertices and topological half-edges into the database, the following steps are also included: Step 701, based on the newly inserted topological vertices and topological half-edges, performing sub-face segmentation on the target facet to generate defective sub-faces; Step 702, according to the depth components corresponding to each pixel in the defective pixel set, calculating the normal displacement deviation value of the defective sub-face relative to the target facet, and adjusting the spatial position of the defective sub-face according to the normal displacement deviation value, so as to reconstruct the geometry of the damaged area in the three-dimensional structural model.

[0017] Preferably, after generating geometric entity data containing physical fracture boundaries, the method further includes the following steps: calling the finite element analysis module to perform mesh generation on the three-dimensional structural model, defining the physical fracture boundaries as geometric discontinuities; applying preset building load conditions to the three-dimensional structural model, and outputting the remaining bearing capacity index characterizing the safety state of the building structure.

[0018] Preferably, when performing mesh generation, the finite element analysis module increases the mesh distribution density within a preset neighborhood of the physical fracture boundary and performs material physical parameter correction on the mesh elements containing the physical fracture boundary through a stiffness reduction factor. The process of acquiring digital image data of the building surface in step 101 is performed by an unmanned aerial vehicle (UAV) system. The UAV system uses an image sensor to perform overlay image acquisition on the building surface and synchronously records the real-time three-dimensional pose data corresponding to each frame of image.

[0019] A construction engineering quality inspection system based on image big data analysis includes:

[0020] The data acquisition module acquires digital image data of the building surface and the pose parameters of the sensors;

[0021] The feature map generation module uses a multi-scale logarithmic Gaussian filter bank to extract the frequency domain response of digital image data, and calculates the local energy based on odd symmetric components and even symmetric components to generate a phase consistency feature map with edge saliency features.

[0022] The defect identification module calculates the local structure tensor feature value of the phase consistency feature map, quantizes the anisotropic feature quantity of the pixel based on the polarity distribution of the first feature value and the second feature value, so as to distinguish the intrinsic texture of the material and the defect boundary in the geometric gradient dimension and identify the defect pixel set.

[0023] The projection transformation module performs spatial projection transformation on the defect pixel set based on the sensor's optical intrinsic parameter matrix and pose parameters to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system.

[0024] The model reconstruction module maps 3D geometric parameters to the parameterized space of the 3D structural model. By calculating the intersection constraints between the 3D geometric parameters and the topological elements of the 3D structural model, it performs topological description data reconstruction in the database of the 3D structural model to generate geometric entity data containing physical fracture boundaries.

[0025] The beneficial effects of this invention are:

[0026] 1. In the quality inspection of building engineering, the synergistic effect of multi-scale frequency domain transformation and phase consistency analysis is used to break the constraint of image gray-level distribution on structural feature extraction. By constructing a multi-scale logarithmic filter bank to extract the frequency domain response, and calculating local energy and amplitude based on odd and even symmetric components, the gray-level intensity information of the original image is converted into a scalar spectrum characterizing the obviousness of the structure. This transformation mechanism from the intensity domain to the structure domain makes the extraction process of defect pixels independent of image contrast and illumination intensity. Even in non-uniform illumination or environments with complex light and shadow interference, the detection sensitivity and response stability of defect edges can still be maintained.

[0027] 2. Based on the ratio and polarity distribution of local tensor eigenvalues, a physical feature separation mechanism for intrinsic texture and structural defects of materials is established. A local matrix is ​​constructed using phase-consistent feature maps and eigenvalues ​​are solved. The tubularity metric function is calculated according to the distribution law of the first and second eigenvalues ​​to achieve quantitative evaluation of the anisotropic features of pixels. This mechanism distinguishes between isotropic pore noise and anisotropic linear cracks from the topological geometric dimension. Without relying on physical light source intervention, it solves the problem of aliasing of intrinsic texture and defect boundaries in spatial gradient and improves the signal-to-noise ratio of image feature segmentation.

[0028] 3. By establishing a logical closed loop of parametric mapping and inverse perspective transformation, a correlation path between optical raster data and vector topology is established. With the help of sensor pose matrix and projection parameters, the coordinates of extracted two-dimensional defect pixels are mapped to the parametric coordinate space of the initial model of the building project. Spline interpolation is then performed using the direction indicated by the feature vector to generate continuous trajectory curves, transforming isolated image pixel sets into topological descriptions with geometric meaning. This provides a definite geometric basis for subsequently updating the underlying data structure of the digital model, and solves the semantic gap problem between image detection results and design models. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0030] Figure 1 This is a flowchart of the building defect identification and topology model reconstruction process for image analysis in this invention;

[0031] Figure 2 This is a finite element mesh generation and bearing capacity evaluation diagram of the physical damage boundary of the present invention. Detailed Implementation

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0033] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0034] Secondly, an embodiment or embodiment referred to herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. An embodiment appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0035] This invention is described in detail with reference to the schematic diagrams. When describing the embodiments of this invention, for ease of explanation, the cross-sectional views of the device structure will be partially enlarged without adhering to the general scale. Moreover, the schematic diagrams are only examples and should not limit the scope of protection of this invention. In addition, in actual manufacturing, the three-dimensional spatial dimensions of length, width and depth should be included.

[0036] Furthermore, in the description of this invention, it should be noted that the terms such as "upper," "lower," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or component referred to has a specific orientation, or is constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0037] Unless otherwise explicitly specified and limited, the terms installation, connection, and linking in this invention should be interpreted broadly. For example, they can refer to fixed connection, detachable connection, or integrated connection; similarly, they can refer to mechanical connection, electrical connection, or direct connection, or indirect connection through an intermediate medium, or internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0038] A method for inspecting the quality of building construction projects based on image big data analysis includes the following steps:

[0039] Step 101: Acquire digital image data of the building surface and the pose parameters of the sensors;

[0040] Step 102: Extract the frequency domain response of digital image data using a multi-scale logarithmic Gaussian filter bank, and calculate the local energy based on the odd symmetric component and the even symmetric component to generate a phase consistency feature map with edge saliency features.

[0041] Step 103: Calculate the local structure tensor eigenvalues ​​of the phase consistency feature map, quantize the anisotropic feature values ​​of the pixels based on the polarity distribution of the first and second eigenvalues, so as to distinguish the intrinsic texture of the material from the defect boundary in the geometric gradient dimension and identify the defect pixel set.

[0042] Step 104: Based on the sensor's optical intrinsic parameter matrix and pose parameters, perform a spatial projection transformation on the defect pixel set to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system.

[0043] Step 105: Map the three-dimensional geometric parameters to the parameterized space of the three-dimensional structural model. By calculating the intersection constraints between the three-dimensional geometric parameters and the topological elements of the three-dimensional structural model, perform topological description data reconstruction in the database of the three-dimensional structural model to generate geometric entity data containing physical fracture boundaries.

[0044] Preferably, the process of generating the phase consistency feature map in step 102 includes: performing multi-scale convolution processing on the digital image data using a multi-scale log-Gaussian filter bank to obtain the extended complex response of the pixel at each scale; calculating the sum of local energy and the sum of local amplitude of the pixel using the real and imaginary parts of the extended complex response; introducing a preset noise power compensation parameter to perform offset correction on the sum of local energy, and determining the ratio of the corrected sum of local energy to the sum of local amplitude as the response value of the pixel in the phase consistency feature map.

[0045] Preferably, the response value is measured by the following phase consistency measure. Sure: Where W(x) is the frequency expansion weighting coefficient, This represents the local energy value at the nth scale. Let be the local amplitude value at the nth scale, T be the noise power compensation parameter, ϵ be a small constant to prevent the denominator from being zero, and ⌊⋅⌋ be an operator that takes a value of 0 when the value in the parentheses is negative.

[0046] Preferably, the process of identifying the defective pixel set in step 103 includes: constructing a local second-order moment matrix using a phase consistency feature map, obtaining a first eigenvalue representing the local structural evolution direction and a second eigenvalue perpendicular to it through eigenvalue decomposition; calculating the distribution ratio of the first eigenvalue and the second eigenvalue to obtain a linearity index; and when the linearity index exceeds a preset defect judgment threshold and the principal axis direction of its anisotropic feature has spatial continuity, classifying the pixel into the defective pixel set.

[0047] Preferably, the process of obtaining the three-dimensional geometric parameters of the defect boundary in the object coordinate system in step 104 includes: obtaining the 6-DOF pose parameters of the image acquisition sensor during sampling, constructing a homography mapping matrix from the pixel coordinate system to the object coordinate system; inputting the contour coordinates of the defect pixel set into the homography mapping matrix, and combining the preset surface depth constraints of the building component to output a defect boundary point cloud sequence with physical scale dimension.

[0048] Preferably, the process of performing topological description data reconstruction in step 105 includes: retrieving the target facets in the three-dimensional structural model that correspond to the three-dimensional geometric parameters, using an interpolation algorithm to fit the three-dimensional geometric parameters into an analytical curve in the surface parameter space; locating the intersection points of the analytical curves and the existing topological edges in the target facets, and dynamically inserting new topological vertices and topological half-edges into the database for the target facets.

[0049] Preferably, after dynamically inserting new topological vertices and topological half-edges into the database, the following steps are also included: Step 701, based on the newly inserted topological vertices and topological half-edges, performing sub-face segmentation on the target facet to generate defective sub-faces; Step 702, according to the depth components corresponding to each pixel in the defective pixel set, calculating the normal displacement deviation value of the defective sub-face relative to the target facet, and adjusting the spatial position of the defective sub-face according to the normal displacement deviation value, so as to reconstruct the geometry of the damaged area in the three-dimensional structural model.

[0050] Preferably, after generating geometric entity data containing physical fracture boundaries, the method further includes the following steps: calling the finite element analysis module to perform mesh generation on the three-dimensional structural model, defining the physical fracture boundaries as geometric discontinuities; applying preset building load conditions to the three-dimensional structural model, and outputting the remaining bearing capacity index characterizing the safety state of the building structure.

[0051] Preferably, when performing mesh generation, the finite element analysis module increases the mesh distribution density within a preset neighborhood of the physical fracture boundary and performs material physical parameter correction on the mesh elements containing the physical fracture boundary through a stiffness reduction factor. The process of acquiring digital image data of the building surface in step 101 is performed by an unmanned aerial vehicle (UAV) system. The UAV system uses an image sensor to perform overlay image acquisition on the building surface and synchronously records the real-time three-dimensional pose data corresponding to each frame of image.

[0052] A construction engineering quality inspection system based on image big data analysis includes:

[0053] The data acquisition module acquires digital image data of the building surface and the pose parameters of the sensors;

[0054] The feature map generation module uses a multi-scale logarithmic Gaussian filter bank to extract the frequency domain response of digital image data, and calculates the local energy based on odd symmetric components and even symmetric components to generate a phase consistency feature map with edge saliency features.

[0055] The defect identification module calculates the local structure tensor feature value of the phase consistency feature map, quantizes the anisotropic feature quantity of the pixel based on the polarity distribution of the first feature value and the second feature value, so as to distinguish the intrinsic texture of the material and the defect boundary in the geometric gradient dimension and identify the defect pixel set.

[0056] The projection transformation module performs spatial projection transformation on the defect pixel set based on the sensor's optical intrinsic parameter matrix and pose parameters to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system.

[0057] The model reconstruction module maps 3D geometric parameters to the parameterized space of the 3D structural model. By calculating the intersection constraints between the 3D geometric parameters and the topological elements of the 3D structural model, it performs topological description data reconstruction in the database of the 3D structural model to generate geometric entity data containing physical fracture boundaries.

[0058] Example 1: During the inspection of concrete pouring surfaces, the dynamic fluctuation of the natural light projection angle and the high-frequency gradient signals generated by the concrete surface pores and hydration patches in the gray-scale matrix of the image spatial domain output by the acquisition device are spatially overlapping with the gradient signals of structural cracks. Conventional threshold determination methods based on spatial brightness gradients cannot separate the high-frequency response of the material's intrinsic texture from the structural edges of physical cracks at the pixel level.

[0059] The data acquisition module acquires digital image data of the building surface and the pose parameters of the sensors. The feature map generation module uses a multi-scale logarithmic Gaussian filter bank to perform multi-scale convolution processing on the digital image data to obtain the extended complex response of the pixels at each scale. It then uses the real and imaginary parts of the extended complex response to calculate the sum of local energy and the sum of local amplitude of the pixels. Based on this, the feature map generation module introduces a preset noise power compensation parameter T to perform offset correction on the sum of local energy, and combines the frequency expansion weighting coefficient W(x) with the local energy value at the nth scale. and the local amplitude value at the nth scale The response value is determined by calculating the phase consistency metric PC(x) based on the small constant ϵ that prevents the denominator from being zero, and the operator that takes a value of 0 when the internal value is negative. This generates a phase consistency feature map with edge saliency characteristics. This frequency domain calculation process converts the pixel gray intensity directly affected by illumination into a scalar spectrum that characterizes the structural coherence, providing input parameters independent of physical illumination fluctuations for subsequent topological defect identification.

[0060] The defect identification module calculates the local structural tensor eigenvalues ​​of the phase consistency feature map, constructs a local second-order moment matrix using the phase consistency feature map, and extracts the first eigenvalue representing the local structural evolution direction and the second eigenvalue perpendicular to it through eigenvalue decomposition. The defect identification module further calculates the distribution ratio of the first and second eigenvalues ​​to obtain a linearity index. Based on the polarity distribution of the first and second eigenvalues, the anisotropic feature of each pixel is quantified. When the linearity index of a specific pixel exceeds a preset defect judgment threshold, and the principal axis of its anisotropic feature has spatial continuity, the defect identification module classifies that pixel into the defect pixel set. This geometric tensor analysis step directly separates isotropic intrinsic porosity from highly anisotropic physical crack boundaries using the differences in the distribution characteristics of structural gradients. The projection transformation module performs a spatial projection transformation on the defect pixel set based on the sensor's optical intrinsic parameter matrix and pose parameters to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system. The model reconstruction module maps the three-dimensional geometric parameters to the parameterized space of the three-dimensional structural model. The target facets corresponding to the 3D geometric parameters are retrieved from the 3D structural model. An interpolation algorithm is used to fit the 3D geometric parameters into an analytical curve in the surface parameter space. The model reconstruction module locates the intersection points of the analytical curves and existing topological edges in the target facets. By calculating the intersection constraints between the 3D geometric parameters and the topological elements of the 3D structural model, new topological vertices and topological half-edges are dynamically inserted into the database of the 3D structural model for the target facets. This underlying operation transforms the 2D optical detection results into geometric topological variations of the computer-aided design model boundary description data, ultimately generating geometric entity data containing physical fracture boundaries. In the 3D structural model database, the target facets are limited to half-edge data structures. After locating the intersection points of the analytical curves and existing topological edges, a new topological vertex is created at the intersection coordinates, and the original topological half-edge is split into two directed half-edges sharing the same vertex. The predecessor and successor pointers of adjacent half-edges are updated synchronously to maintain the Euler characteristic number conservation of the facets, ensuring that the segmented defective sub-faces satisfy the topological consistency of the manifold mesh. Based on the normal displacement deviation value, the spatial coordinates of the defective sub-face vertices are adjusted along the normal vector direction of the target facets to reconstruct the geometric shape of the damaged area.

[0061] Example 2: During the inspection of concrete pouring surfaces, the spatiotemporal non-uniform fluctuations of ambient light and the local spatial gradient formed by the superposition of hydration patches on the concrete surface are prone to spectral aliasing with the boundaries of microcracks. To verify the reliability and effectiveness of the present invention, a closed optical darkroom test platform was built, in which an array-type adjustable illumination simulation system was deployed, maintaining the output illuminance range between 50 lx and 1000 lx, and generating a non-uniform illumination field in the spatial coordinate domain according to a preset gradient function. The test selected standard precast concrete components with physical cracks on the surface with widths between 0.15 mm and 0.55 mm, and used an industrial camera array with a calibrated resolution of 0.1 mm to collect raw digital image data. To simulate real engineering environmental disturbances, the test injected local Gaussian white noise with a variance of 15.2 into the raw digital image data, and superimposed a light spot interference signal with an illuminance contrast fluctuation rate of 62.5%.

[0062] The system determines the scale and orientation parameters of a multi-scale logarithmic Gaussian filter bank. The scale parameters are set to balance the coverage integrity of low-frequency structural features with high-frequency noise sensitivity. Based on the local spectral bandwidth evolution law, when the pixel distribution width of the crack under test is in a narrow band state, in order to avoid signal aliasing and ensure edge positioning accuracy, the scale parameters tend to the lower limit of the value range. The experimental group determines the number of filter scales to be 4 and the number of orientations to be 6. Under the boundary condition constraints of a center frequency of 0.24 and a bandwidth coefficient of 0.56, the system extracts the frequency domain response of digital image data and obtains the extended complex response. This parameter combination enables the system to take into account both the high-frequency singularity of microcracks and the low-frequency slow variation of large-scale patches. The experiment sets up a control group and an experimental group. The control group uses a spatial brightness gradient-dependent method. Traditional edge extraction operators are used to obtain feature signals. The experimental group uses the aforementioned multi-scale logarithmic Gaussian filter bank to extract the frequency domain response. The sum of local energy and the sum of local amplitude are calculated using the real and imaginary parts of the extended complex response. The experimental group introduces a noise power compensation parameter T to correct the sum of local energy. Combined with the sum of local amplitude, the phase consistency metric PC(x) is calculated to generate a phase consistency feature map. The experimental group uses this phase consistency feature map to construct a local second-order moment matrix. The first and second eigenvalues ​​are extracted through eigenvalue decomposition, and the linearity index is calculated. The experiment constructs three test gradients with low, medium, and high perturbations, respectively, by adjusting the illuminance contrast fluctuation rate of the illumination simulation system, with contrast fluctuation rates of 25.5%, 45.2%, and 62.5%.

[0063] Under the aforementioned highly perturbated gradient, the raw digital image data containing noise was acquired. In the gradient response matrix output by the control group, the amplitude at the crack center was 125.4, while the amplitude at the hydration patch boundary reached 142.7. This inverted signal-to-noise ratio phenomenon resulted in dense false edge pixels in the control group output. The experimental group processed the same set of raw digital image data and calculated that the sum of local energies of the crack pixels was 85.6, and the sum of local amplitudes was 92.1. The experimental group applied a noise power compensation parameter T of 12.5 to correct this sum of local energies. After correction, the phase consistency metric PC(x) of the crack pixels reached 0.82, while the phase consistency metric PC(x) of the hydration patch region... To verify the engineering rationality of the numerical boundary, the noise power compensation parameter T was gradually increased from 5.0 to 30.0. Test data showed that when the parameter was below 10.0, the high-frequency noise suppression of the material's intrinsic texture was insufficient, leading to discrete artifacts in the phase consistency feature map. When the parameter exceeded 20.0, the response values ​​of crack pixels exhibited a non-linear exponential decay, causing breaks in the feature map. The above data empirically demonstrated that the optimal working window for the noise power compensation parameter T was 12.0 to 18.0. The experimental group constructed a local second-order moment matrix for the phase consistency feature map and performed local structural tensor eigenvalue decomposition. Under the aforementioned highly perturbed gradient, the experimental group extracted the cracks. The first eigenvalue of the region is 4.52, and the second eigenvalue perpendicular to it is 0.18. The calculated linearity index reaches 0.96, which quantifies a strong anisotropic characteristic, and its principal axis remains continuous in geometric space. Conversely, the calculated first eigenvalue of the residual point-like pore region is 1.25, and the second eigenvalue is 1.12, with a linearity index of only 0.05, exhibiting obvious isotropic characteristics. The experimental group set the defect judgment threshold to 0.8. Based on this threshold, the system filters out unqualified pixels, separates hydration patches and pore interference, and classifies the qualified pixels into the defect pixel set. In the three test gradients from low to high, the defect pixel set output by the experimental group and The overlap of the real crack topology remained at 98.2%, 97.5%, and 96.8%, respectively, without any sharp performance degradation due to increased interference intensity. Experimental data show that by extracting the frequency domain response of digital image data and calculating the phase consistency metric PC(x), and combining it with the eigenvalue decomposition of the local structure tensor to quantify anisotropic features, the system can suppress spatial high-frequency gradient interference in environments with drastic fluctuations in illumination contrast. The frequency domain mapping combined with the geometric structure tensor decomposition technique separates illumination contrast fluctuations from intrinsic material porosity at the bottom pixel level. The output defect pixel set is freed from the constraint of absolute brightness, maintaining the determinism and accuracy of real physical fracture boundary identification.

[0064] Example 3: In automated visual inspection of building component surfaces, fluctuations in dark current of the image sensor caused by changes in ambient temperature and time-varying intensity of external light sources result in a non-stationary signal-to-noise ratio distribution in the underlying digital image data. If a static noise power compensation parameter T is set during the calculation of the phase consistency metric PC(x), the static parameter cannot suppress the increased high-frequency components when the local ambient noise level increases; conversely, when the ambient noise level decreases, an excessively large static parameter causes excessive cancellation of signals from microcrack structures. The conventional fixed parameter setting logic reduces the adaptability of the feature extraction operator to dynamic physical environments. When extracting the frequency domain response using a multi-scale logarithmic Gaussian filter bank, the system initiates a dynamic calibration procedure for noise parameters. The system extracts the filter channel corresponding to the smallest spatial scale in the multi-scale logarithmic Gaussian filter bank, inputs the digital image data into this smallest spatial scale filter channel to extract the local amplitude set of all pixels, sorts the values ​​in this local amplitude set to obtain the median, and uses this median as the noise amplitude benchmark. The system combines the preset statistical confidence coefficient and the noise amplitude benchmark, according to the formula... Calculate the noise power compensation parameter T, where T is the noise power compensation parameter and k is the dimensionless statistical confidence coefficient. To obtain the median of the local amplitude set, this calibration procedure utilizes the physical response characteristics of minimum-scale filters, which exhibit high sensitivity to high-frequency noise and low sensitivity to smooth structural features. It transforms the noise distribution characteristics in the mixed signal into a quantization benchmark. The noise power compensation parameter T is selected from the minimum spatial scale channel of the multi-scale logarithmic Gaussian filter bank to obtain the median of the local amplitude set of the pixel. Determined by combining the preset statistical confidence coefficient k Where k takes values ​​from 1.5 to 3.0; if discrete high-frequency artifacts appear in the phase consistency feature map, k is increased in increments of 0.1 until the intrinsic texture response value of the material is lower than the preset noise threshold.

[0065] The system constructs a data-driven quantization path for selecting the defect judgment threshold, generates a phase consistency feature map and calculates a linearity index set containing all pixels. It then uses a kernel density estimation algorithm to fit the probability density distribution curve of the linearity index set and calculates the first derivative of this curve. Within the probability density distribution curve, it locates the first peak interval representing the aggregation of intrinsic material texture and the second peak interval representing the aggregation of structural cracks. Local minima located between the first and second peak intervals are extracted. The system sets the linearity index value corresponding to these local minima as the defect judgment threshold. When the linearity index of a specific pixel exceeds this threshold and its anisotropic feature has spatial continuity along its principal axis, the system classifies that pixel into the defect pixel set. The above calculation model and threshold calibration path construct a closed parameter adaptive boundary in the image's underlying data domain. Based on the signal-to-noise ratio of the input image data, the system adjusts the cancellation strength of the phase feature and the cutting threshold of the defect boundary. This mechanism maintains a steady-state output of the polarity decomposition of the local structural tensor eigenvalues, ensuring the determinism of the physical fracture boundary extraction process under complex conditions.

[0066] Example 4: Before deploying the image acquisition sensor in the inspection of new building components, the system initiates a pre-calibration procedure for optical and spatial references. The system controls the inspection equipment carrying the image acquisition sensor to move along an orthogonal trajectory within the physical space to acquire reference digital images from multiple perspectives, including a standard physical calibration plate. The system extracts feature corner points distributed in a grid within the calibration plate from the reference digital images of each perspective, establishes a two-dimensional pixel coordinate set, and constructs a perspective image model by combining the known three-dimensional physical coordinates of the standard physical calibration plate. The focal length, principal point coordinates, and radial distortion coefficient of the image acquisition sensor are separated using the nonlinear least squares method to generate an optical intrinsic parameter matrix. Based on the perspective image model, the mapping matrix from the spatial coordinate system to the pixel coordinate system is calculated to establish the 6-DOF pose parameters of the sensor. This measurement process utilizes the law of minimizing the reprojection deviation of the physical feature point array at different spatial perspectives to transform the inherent physical deviation of the underlying optical hardware into quantized matrix parameters.

[0067] The system uses the optical intrinsic parameter matrix and 6-DOF pose parameters obtained from the above calibration as the basic parameters for spatial transformation to generate a preset depth constraint on the surface of the building component. It then drives the single-point ranging component to emit a pulse sequence towards the surface of the building component to obtain the normal distance measurement value of the discrete physical surface reflection point, based on the formula... Calculate the depth parameter from the optical center of the sensor to the surface of the building component. ,in, For depth parameters, The distance is measured in the normal direction, where θ is the angle between the sensor's optical axis and the normal direction of the building component's surface. The system uses this depth parameter... Along with setting the local coordinates of the 3D structural model to the preset depth constraints of the building component surface, a homography mapping matrix is ​​constructed using the optical intrinsic parameter matrix and 6-DOF pose parameters, combined with the feedback of the normal distance measurement value from the single-point ranging component. Calculate the depth parameter by the angle θ between the sensor's optical axis and the building surface. Convert the two-dimensional coordinates (u,v) of the defect pixel set into object coordinates. The system generates a 3D geometric point cloud sequence and fits an analytical curve. During the spatial projection transformation, the system extracts the contour coordinates of the defect pixel set and inputs them into the homography mapping matrix, combining the depth parameter. The two-dimensional pixel coordinates are back-projected to the object coordinate system to calculate and output three-dimensional geometric parameters. This pre-quantization measurement procedure establishes a scale benchmark for the two-dimensional optical grid sequence to the three-dimensional physical space coordinate system, so that the reconstructed three-dimensional geometric parameters of the defect boundary have spatial geometric dimensions aligned with the real physical structure.

[0068] Example 5: In the offline optimization parameter search scenario for visual recognition of surface defects in building structures, the integral scale of a fixed local structural tensor is prone to causing divergence of the edge response function and rank reduction of eigenvalues. To obtain the optimal parameter combination, the system initiates an adaptive calibration procedure for the integral window before eigenvalue decomposition before deployment, obtains the previously generated phase consistency feature map, and extracts the center wavelength parameter within the main lobe region of the frequency domain response of the phase consistency feature map. The objective function is set as the maximization index of the anisotropic feature quantity. Based on the center wavelength parameter and the preset scale correlation coefficient, the Gaussian smoothing scale used to construct the local second-order moment matrix is ​​determined through multiplication. The system is on this Gaussian smoothing scale Under constraints, the squared terms of the first-order partial derivatives in the horizontal direction and the squared terms of the first-order partial derivatives in the vertical direction of the image are used to generate a local second-order moment matrix. The first and second eigenvalues ​​are extracted by solving the characteristic equation of this local second-order moment matrix. This calibration procedure optimizes the objective function through iteration and uses the sampling receptive field of the underlying frequency domain wavelength dynamic matching geometric tensor analysis to suppress eigenvalue distortion caused by cross-scale physical porosity, providing a numerically stable parameter basis for the calculation of anisotropic eigenvalues.

[0069] Before reconstructing the topology description data, the system performs a pre-calibration procedure for the geometric compliance of the underlying data structure of the boundary representation method. The model reconstruction module uses an interpolation algorithm to fit the three-dimensional geometric parameters into an analytical curve in the parameter space of the target patch surface. The system extracts the tangent vector at discrete nodes at equal intervals along the parameterized direction of the analytical curve. By calculating the cross product vector of the tangent vectors of adjacent nodes, the polarity data of the cross product vector projected onto the normal of the target patch is extracted. When the system detects that the polarity data has a sign reversal, it establishes that there is a self-intersecting anomalous node at that point. At this node, the analytical curve is truncated into multiple monotonic sub-curves. The truncated monotonic sub-curve sequence is mapped to the database of the three-dimensional structural model. Non-intersecting topological vertices and topological half-edges are inserted into the target patch in sequence. This pre-calibration mechanism actively trims the generation path of non-manifold geometric topology in the underlying spatial domain, standardizes the parameter boundary of the process of dividing a single patch into defective sub-surfaces, and establishes that the generated geometric entity data meets the compatibility requirements of cross-platform mesh generation.

[0070] 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 it. 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 spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for inspecting the quality of building construction projects based on image big data analysis, characterized in that, Includes the following steps: Step 101: Acquire digital image data of the building surface and the pose parameters of the sensors; Step 102: Extract the frequency domain response of digital image data using a multi-scale logarithmic Gaussian filter bank, and calculate the local energy based on the odd symmetric component and the even symmetric component to generate a phase consistency feature map with edge saliency features. Step 103: Calculate the local structure tensor eigenvalues ​​of the phase consistency feature map, quantize the anisotropic feature values ​​of the pixels based on the polarity distribution of the first and second eigenvalues, so as to distinguish the intrinsic texture of the material from the defect boundary in the geometric gradient dimension and identify the defect pixel set. Step 104: Based on the sensor's optical intrinsic parameter matrix and pose parameters, perform a spatial projection transformation on the defect pixel set to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system. Step 105: Map the three-dimensional geometric parameters to the parameterized space of the three-dimensional structural model. By calculating the intersection constraints between the three-dimensional geometric parameters and the topological elements of the three-dimensional structural model, perform topological description data reconstruction in the database of the three-dimensional structural model to generate geometric entity data containing physical fracture boundaries.

2. The method for inspecting the quality of building construction projects based on image big data analysis according to claim 1, characterized in that, The process of generating the phase consistency feature map in step 102 includes: performing multi-scale convolution processing on the digital image data using a multi-scale log-Gaussian filter bank to obtain the extended complex response of the pixel at each scale; calculating the sum of local energy and the sum of local amplitude of the pixel using the real and imaginary parts of the extended complex response; introducing a preset noise power compensation parameter to perform offset correction on the sum of local energy, and determining the ratio of the corrected sum of local energy to the sum of local amplitude as the response value of the pixel in the phase consistency feature map.

3. The method for inspecting the quality of building construction projects based on image big data analysis according to claim 2, characterized in that, The response value is determined using the following phase consistency metric, PC(x): Where W(x) is the frequency expansion weighting coefficient, This represents the local energy value at the nth scale. Let be the local amplitude value at the nth scale, T be the noise power compensation parameter, ϵ be a small constant to prevent the denominator from being zero, and ⌊⋅⌋ be an operator that takes a value of 0 when the value in the parentheses is negative.

4. The method for inspecting the quality of building construction projects based on image big data analysis according to claim 1, characterized in that, The process of identifying the defective pixel set in step 103 includes: constructing a local second-order moment matrix using the phase consistency feature map; obtaining a first eigenvalue representing the local structural evolution direction and a second eigenvalue perpendicular to it through eigenvalue decomposition; calculating the distribution ratio of the first eigenvalue and the second eigenvalue to obtain a linearity index; and when the linearity index exceeds the preset defect judgment threshold and the principal axis direction of its anisotropic feature has spatial continuity, the pixel is included in the defective pixel set.

5. The method for inspecting the quality of building engineering based on image big data analysis according to claim 1, characterized in that, The process of obtaining the three-dimensional geometric parameters of the defect boundary in the object coordinate system in step 104 includes: obtaining the 6-DOF pose parameters of the image acquisition sensor during sampling, constructing a homography mapping matrix from the pixel coordinate system to the object coordinate system; inputting the contour coordinates of the defect pixel set into the homography mapping matrix, and combining the preset surface depth constraints of the building components to output a defect boundary point cloud sequence with physical scale dimensions.

6. The method for inspecting the quality of building construction projects based on image big data analysis according to claim 1, characterized in that, Step 105 involves reconstructing the topological description data, which includes: retrieving the target facets in the 3D structural model that correspond to the 3D geometric parameters; using an interpolation algorithm to fit the 3D geometric parameters into an analytical curve in the surface parameter space; locating the intersection points of the analytical curves and the existing topological edges in the target facets; and dynamically inserting new topological vertices and topological half-edges into the database for the target facets.

7. The method for inspecting the quality of building engineering based on image big data analysis according to claim 6, characterized in that, After dynamically inserting new topological vertices and topological half-edges into the database, the following steps are also included: Step 701, based on the newly inserted topological vertices and topological half-edges, perform sub-face segmentation on the target facet to generate defective sub-faces; Step 702, according to the depth components corresponding to each pixel in the defective pixel set, calculate the normal displacement deviation value of the defective sub-face relative to the target facet, and adjust the spatial position of the defective sub-face according to the normal displacement deviation value, so as to reconstruct the geometry of the damaged area in the three-dimensional structural model.

8. The method for inspecting the quality of building construction projects based on image big data analysis according to claim 1, characterized in that, After generating geometric entity data containing physical fracture boundaries, the following steps are also included: calling the finite element analysis module to perform mesh generation on the three-dimensional structural model, defining the physical fracture boundaries as geometric discontinuities; applying preset building load conditions to the three-dimensional structural model, and outputting the remaining bearing capacity index characterizing the safety state of the building structure.

9. A method for inspecting the quality of building construction based on image big data analysis according to claim 8, characterized in that, When performing mesh generation, the finite element analysis module increases the mesh distribution density within a preset neighborhood of the physical fracture boundary and performs material physical parameter correction on the mesh elements containing the physical fracture boundary through a stiffness reduction factor. The process of acquiring digital image data of the building surface in step 101 is performed by the UAV system. The UAV system uses image sensors to perform overlay image acquisition on the building surface and synchronously records the real-time three-dimensional pose data corresponding to each frame of image.

10. A construction engineering quality inspection system based on image big data analysis, used to implement the construction engineering quality inspection method based on image big data analysis as described in claim 1, characterized in that, include: The data acquisition module acquires digital image data of the building surface and the pose parameters of the sensors; The feature map generation module uses a multi-scale logarithmic Gaussian filter bank to extract the frequency domain response of digital image data, and calculates the local energy based on odd symmetric components and even symmetric components to generate a phase consistency feature map with edge saliency features. The defect identification module calculates the local structure tensor feature value of the phase consistency feature map, quantizes the anisotropic feature quantity of the pixel based on the polarity distribution of the first feature value and the second feature value, so as to distinguish the intrinsic texture of the material and the defect boundary in the geometric gradient dimension and identify the defect pixel set. The projection transformation module performs spatial projection transformation on the defect pixel set based on the sensor's optical intrinsic parameter matrix and pose parameters to obtain the three-dimensional geometric parameters of the defect boundary in the object coordinate system. The model reconstruction module maps 3D geometric parameters to the parameterized space of the 3D structural model. By calculating the intersection constraints between the 3D geometric parameters and the topological elements of the 3D structural model, it performs topological description data reconstruction in the database of the 3D structural model to generate geometric entity data containing physical fracture boundaries.

Images