A prefabricated building construction quality safety management monitoring method and system

Abstract

The application discloses a prefabricated building construction quality safety management monitoring method and system, relates to prefabricated buildings, and comprises the following steps: preprocessing a prefabricated component image; adopting an improved Hessian matrix ridge line detection method to perform linear structure analysis on standardized image data, calculating a second-order partial derivative matrix of each pixel point, extracting linear structure response data through eigenvalue decomposition, and generating a linear crack feature data set containing crack position, direction and intensity; according to the linear crack feature data set, converting discrete crack response point data into crack region data with a continuous direction field by calculating a main direction consistency factor in a local neighborhood; and generating prefabricated building construction quality safety management measures according to the crack region data. The application improves the detection accuracy of linear cracks, and the traditional Hessian matrix method can only rely on eigenvalue size to judge, so that real cracks and pseudo-ridge line responses generated by concrete surface textures, air hole edges and the like cannot be effectively distinguished.

Description

Technical Field

[0001] This application relates to the field of prefabricated buildings, and in particular to a method and system for monitoring and managing the construction quality and safety of prefabricated buildings. Background Technology

[0002] With the accelerated advancement of industrialized construction, prefabricated buildings are widely used due to their advantages such as high construction efficiency, controllable quality, and environmental friendliness. Precast components, as the core components of prefabricated buildings, play a crucial role in the construction industry. However, precast components are prone to defects such as surface cracks during production, transportation, hoisting, and storage. If these defects are not detected and addressed in a timely manner, they will seriously affect the safety and durability of the building structure. Therefore, in actual construction, comprehensive defect detection of precast components is necessary before their use to ensure project quality.

[0003] Currently, the detection of surface cracks in precast components mainly relies on manual visual inspection and image processing technology. Manual inspection is inefficient and highly subjective, making it difficult to meet the needs of large-scale prefabricated building construction. Adaptive threshold segmentation is a commonly used automated detection method, but it faces many challenges in actual construction scenarios: environmental limitations, especially insufficient and uneven lighting during nighttime operations, significantly increase the noise content of the acquired precast component images, severely affecting detection accuracy. Furthermore, the surface of precast components contains complex interferences such as concrete textures, formwork imprints, and stains; the cracks themselves also exhibit complex forms such as discontinuity, bifurcation, and width variations, making it difficult for simple threshold segmentation methods to accurately identify them.

[0004] The Hessian matrix method, a ridge detection technique based on the second derivative, is theoretically suitable for detecting linear crack structures. However, traditional Hessian matrix methods suffer from severe noise robustness issues in crack detection applications of precast components in assembled buildings. This method uses only the second eigenvalue |λ²| as a measure of ridge strength; any structure generating a negative eigenvalue will produce a response, including non-crack structures such as random textures on concrete surfaces and pore edges, leading to a large number of pseudo-ridge responses. Due to the lack of a deep discrimination mechanism for the essential characteristics of linear structures, traditional methods cannot effectively distinguish between genuine linear cracks and pseudo-responses generated by noise, resulting in a large number of false detections in the detection results and seriously affecting the reliability of automated quality inspection.

[0005] Therefore, there is an urgent need to develop an improved detection method that can accurately identify surface cracks in prefabricated components and effectively suppress interference from complex backgrounds, in order to meet the actual needs of quality and safety management in prefabricated building construction. Summary of the Invention

[0006] To address the problem that traditional Hessian matrix methods suffer from poor noise robustness in crack detection of precast components in prefabricated buildings, and that relying solely on eigenvalue magnitude leads to an inability to effectively distinguish between genuine cracks and pseudo-ridge responses generated by concrete surface textures, pore edges, etc., resulting in numerous false detections, this application provides a method and system for monitoring and managing the construction quality and safety of prefabricated buildings. This method introduces an eigenvalue ratio factor into the Hessian matrix ridge detection process. and gradient direction consistency factor A dual discrimination mechanism is used to construct an improved ridge response intensity. This improves the detection accuracy of linear cracks in prefabricated components.

[0007] One aspect of this application provides a method for monitoring and managing the construction quality and safety of prefabricated buildings, comprising: S1, acquiring images of prefabricated components at the construction site of the prefabricated building; S2, preprocessing the images of the prefabricated components to obtain standardized image data; S3, performing linear structure analysis on the standardized image data using an improved Hessian matrix ridge detection method, calculating the second-order partial derivative matrix of each pixel, extracting linear structural response data through eigenvalue decomposition, and generating a linear crack feature dataset containing crack location, direction, and intensity; S4, based on the linear crack feature dataset, converting discrete crack response point data into crack region data with continuous direction field by calculating the principal direction consistency factor in the local neighborhood; and S5, generating construction quality and safety management measures for prefabricated buildings based on the crack region data.

[0008] Further, in step S2, the prefabricated component image is preprocessed to obtain standardized image data, including: converting the prefabricated component image into a single-channel grayscale image; performing Gaussian filtering on the single-channel grayscale image to obtain a denoised image; enhancing the local contrast of the denoised image using an adaptive histogram equalization algorithm to obtain enhanced image data; and normalizing the enhanced image data to map pixel values ​​to the 0-1 range to obtain standardized image data.

[0009] Furthermore, S3 employs an improved Hessian matrix ridge detection method to perform linear structure analysis on standardized image data, including: at multiple scales. Below is a standardization of image data Gaussian smoothing is performed to obtain a scale-space image sequence. ,in, , and These are the minimum and maximum detection scales, respectively. The scale sampling factor; for each scale The image below Calculate the second-order partial derivatives of each pixel and construct a scale-normalized Hessian matrix. : ;in, respectively scale The second-order partial derivatives and mixed partial derivatives of the image below. The scale normalization factor; the Hessian matrix for each pixel at each scale. Perform eigenvalue decomposition to obtain eigenvalues. , and the corresponding feature vectors , ;in, Based on eigenvalues , Calculate the eigenvalue ratio factor at each scale. Based on the image gradient vector and feature vector Calculate the gradient direction consistency factor for each pixel at each scale. ; Calculate the ridgeline response intensity at various scales ,when The response is retained if it is active, otherwise it is set to zero to ensure that only dark linear structures are detected.

[0010] Select the maximum response value and the corresponding optimal scale from all scales: ; Local adaptive thresholding is applied to the maximum response intensity S. deal with: .

[0011] Extract the pixel location coordinates (x, y) that meet the threshold condition and the optimal scale. The principal direction angle θ and response intensity and estimated crack width Construct a linear crack feature dataset .

[0012] In particular, the width of surface cracks in prefabricated components of assembled buildings varies greatly, ranging from 0.1 mm for capillary cracks to 5 mm for structural cracks. Traditional single-scale Hessian matrix detection methods face a fundamental contradiction: while small-scale detection can detect minute cracks, it produces a double-edge response for wide cracks, misidentifying a crack as two parallel lines; large-scale detection can accurately identify the overall structure of wide cracks, but it completely ignores minute cracks. This dilemma of scale selection severely restricts the applicability of crack detection.

[0013] This application constructs a scale-space sequence. This approach computes ridge responses in parallel across multiple scales, fundamentally overcoming the limitations of single-scale computation. Each scale's Gaussian smoothing kernel is matched to cracks within a specific width range, when... At this scale, the Hessian response reaches its optimal level, which can fully capture the linear characteristics of the crack while avoiding edge separation.

[0014] In addition, a scale normalization factor is introduced. Standardize the Hessian matrix: The magnitude of the second derivative is inversely proportional to the scale; the response at large scales is naturally suppressed. Normalization makes the feature values ​​at different scales comparable, allowing wide cracks to produce responses of equal strength to narrow cracks at their appropriate scales. This scale invariance ensures the fairness of the detection algorithm for cracks of various widths.

[0015] Furthermore, based on eigenvalues , Calculate the eigenvalue ratio factor at each scale. This includes: for each scale eigenvalues ​​under , Calculate the eigenvalue ratio According to the eigenvalue ratio The eigenvalue ratio factor at each scale is calculated using a Gaussian decay function: ;in: For scale The eigenvalue ratio factor is defined below, with a value range of [0, 1]. For scale-adaptive linear structure discrimination sensitivity parameters, ,in, As the current testing standard, This is the scaling factor, with a value range of [1.0, 3.0].

[0016] In particular, the fundamental flaw of traditional Hessian matrix crack detection methods lies in their overly simplistic discrimination mechanism—relying solely on the second eigenvalue. The negative value characteristic is used to identify dark linear structures. This single criterion leads to any structure that produces local depressions being identified as a potential crack, including pore edges, random textures, and formwork imprints on concrete surfaces. In the complex surface environment of precast components, this coarse discrimination method generates a large number of spurious responses, seriously affecting the reliability of the detection.

[0017] Real cracks, being elongated linear structures, exhibit a significant but not extreme difference between the two eigenvalues ​​of their Hessian matrix, as shown in the following manner: It is in the middle range, at this time Close to 1; while point noise Approaching 1 (isotropic), patchy texture In cases where the value is very large (extreme anisotropy), Rs will be significantly suppressed by the Gaussian decay function in both situations. Specifically, the discrimination sensitivity parameter... With detection scale Related ( This ensures that the discrimination criteria remain consistent across different scales, avoiding the failure of fixed thresholds in a multi-scale framework.

[0018] Furthermore, based on the image gradient vector and feature vector Calculate the gradient direction consistency factor for each pixel at each scale. Including: computational scale The image below The gradient vector at each pixel (x, y): ;in, Using either the first-order difference operator or the Sobel operator on scaled images The above calculation is obtained; based on the gradient vector and scale The eigenvectors below Calculate the pixel at scale Directional consistency factor : , where ε is a small constant to prevent division by zero.

[0019] In particular, real cracks exhibit a regular grayscale distribution: the center of the crack is the darkest, gradually brightening towards both sides, forming a gradient field perpendicular to the crack's orientation. This gradient direction corresponds to the second eigenvector of the Hessian matrix. (Corresponding to the crack normal) The height is consistent, making The gradient direction is close to 1; however, the gradient direction of random textures and noise points is chaotic and lacks correlation with the local principal direction, resulting in... The value is very small.

[0020] This application integrates two factors into the response intensity calculation in the form of a product: Only when reasonable linear geometric characteristics are simultaneously satisfied ( High) and regular gradient distribution ( Only a high-resolution (high-resolution) structure can achieve a strong response. Failure to meet any one of these conditions will suppress the response, effectively filtering out various spurious responses, such as point noise. Low suppression, random texture factor Low suppression means that only genuine cracks can achieve high scores in both dimensions.

[0021] Furthermore, local adaptive thresholding : ,in, represents the mean and standard deviation of the response intensity within the local window; k is the threshold coefficient.

[0022] Furthermore, S4, by calculating the principal direction consistency factor within the local neighborhood, the discrete crack response point data is converted into crack region data with continuous direction field, including: for the linear crack feature dataset Each crack point in Define its neighborhood search window : ;in, Let i be the neighborhood search radius of crack point i. β is the neighborhood coefficient; calculate the principal direction consistency factor among crack points within the neighborhood. ; Calculate the spatial continuity factor between crack points in the neighborhood. ; Calculate the overall connection strength: ;in, Ensure the connection strength does not exceed the response strength of the weakest point; when At that time, crack points i and j are marked as the same crack region; where, A connection threshold is set; a continuous set of crack regions is generated using a region growing algorithm. , of which each region Includes a set of points belonging to the same crack region; for each crack region Calculate regional attributes: centerline coordinate sequence, average width, total length, average direction, and maximum response intensity, and construct crack region data. .

[0023] In particular, the actual output of crack detection is often a discrete set of response points, rather than a continuous crack profile. This discreteness stems from several factors: local contamination or texture changes on the surface of precast components may cause local interruptions in the crack response; the crack itself may have subtle discontinuities or branches; and the setting of the detection threshold may also cause the loss of weak response areas.

[0024] This application considers both the directional continuity and spatial proximity of cracks. Main direction consistency factor. The directional similarity between two crack points was quantified, and the periodicity of the sine function was cleverly used to handle the periodicity of the angle, ensuring that only points with similar directions can be connected. Spatial continuity factor. Instead of simply calculating the Euclidean distance, it calculates the perpendicular distance from point j to the crack direction line where point i is located. This design allows for larger intervals along the crack direction, but strictly limits the deviation in the vertical direction, which fits the slender and linear geometric characteristics of the crack.

[0025] Furthermore, traditional fixed-radius searches cannot adapt to variations in crack width—the search range may be too large for fine cracks, leading to incorrect connections, while the search range may be too small for wide cracks, resulting in connection interruptions. By correlating the search radius with the crack width, an adaptive effect is achieved where "wide cracks allow for larger gaps, while fine cracks require tighter connections." (Comprehensive connection strength) The calculation introduces a constraint on the response strength, using the min function to ensure that the connection strength does not exceed the response strength of the weaker of the two points.

[0026] Furthermore, the principal direction consistency factor among crack points in the neighborhood is calculated. : ;in, and These are the principal direction angles of crack points i and j, respectively; This is the tolerance parameter for angle differences;

[0027] Specifically, real cracks typically exhibit a continuous and consistent direction, while pseudo-crack points, such as concrete surface textures and pore edges, often display a random distribution of directions. The principal direction consistency factor, through an exponential Gaussian decay function, precisely quantifies the degree of directional consistency between adjacent points, resulting in points on the same crack receiving high connection weights, while points on different cracks or noise points receive low connection weights.

[0028] Furthermore, this application adopts Instead of a direct angle difference, this method cleverly solves the problem of angle periodicity, ensuring accurate calculation of angle differences near 0° and 180°. This is crucial for accurate assessment of crack direction in complex environments, avoiding the failure of traditional angle difference calculations under boundary conditions.

[0029] Finally, through parameters An adjustable tolerance for angle differences is introduced, enabling the algorithm to adapt to the natural curvature and variations of different types of cracks while maintaining effective suppression of random directional noise. This soft thresholding mechanism exhibits greater robustness and adaptability compared to hard thresholding.

[0030] Furthermore, the spatial continuity factor between crack points within the neighborhood is calculated. : ,in, Let j be the perpendicular distance from point j to the crack direction line where point i is located. The average width of the two points;

[0031] In particular, this application will use tolerance parameters Setting the value to the average width of relevant points enables adaptive discrimination of cracks of different widths. This solves the problem of insufficient connection of wide cracks or false connection of narrow cracks caused by fixed thresholds in traditional methods, making it particularly suitable for accurate identification of cracks of various scales in prefabricated buildings. In addition, this factor ensures the geometric continuity of the crack area, effectively filters out discrete response points that do not conform to the characteristics of cracks in spatial distribution, and significantly improves the ability to identify pseudo-crack features such as surface scratches, texture edges, and hole edges.

[0032] Another aspect of this application provides a prefabricated building construction quality and safety management and monitoring system, including: an image acquisition module for acquiring images of prefabricated components at the prefabricated building construction site; an image preprocessing module for preprocessing the prefabricated component images to obtain standardized image data; the image preprocessing module converts the prefabricated component images into single-channel grayscale images, performs Gaussian filtering on the single-channel grayscale images to obtain denoised images, enhances local contrast through an adaptive histogram equalization algorithm to obtain enhanced image data, and normalizes the enhanced image data to map pixel values ​​to the 0 to 1 range to obtain standardized image data;

[0033] The linear structure analysis module performs linear structure analysis on standardized image data using an improved Hessian matrix ridge detection method. It calculates the second-order partial derivative matrix for each pixel, extracts linear structure response data through eigenvalue decomposition, and generates a linear crack feature dataset containing crack location, orientation, and intensity. The linear structure analysis module operates at multiple scales. The standardized image data is Gaussian smoothed, the second-order partial derivatives are calculated to construct a scale-normalized Hessian matrix, eigenvalues ​​and eigenvectors are obtained through eigenvalue decomposition, the eigenvalue ratio factor and gradient direction consistency factor are calculated, the ridge response intensity is calculated and the maximum response value and the corresponding optimal scale are selected, and the linear crack feature dataset is extracted through local adaptive thresholding.

[0034] The crack region analysis module is used to convert discrete crack response point data into crack region data with continuous directional field by calculating the principal direction consistency factor within the local neighborhood, based on the linear crack feature dataset. The module defines a neighborhood search window for each crack point, calculates the principal direction consistency factor and spatial continuity factor between crack points within the neighborhood, determines the comprehensive connection strength, generates a continuous set of crack regions through a region growing algorithm, and calculates the attributes of each crack region. The safety management module is used to generate prefabricated building construction quality and safety management measures based on the crack region data.

[0035] Among them, the eigenvalue ratio factor By calculating the eigenvalue ratio at each scale And a Gaussian decay function is used. Calculations show that For scale-adaptive linear structure discrimination sensitivity parameters, ,in, As the current testing standard, Scale adjustment coefficient; gradient direction consistency factor By calculating the gradient vector of the scaled image With feature vectors The relationship is obtained as follows: Local adaptive threshold ,in, Here, k represents the mean and standard deviation of the response intensity within the local window, and k is the threshold coefficient; the principal direction consistency factor... ,in, and Let i and j be the principal direction angles, respectively. Angular difference tolerance parameter; spatial continuity factor ,in, Let j be the perpendicular distance from point j to the crack direction line where point i is located. The average width between the two points.

[0036] Compared to existing technologies, the advantages of this application are:

[0037] To address the problem that traditional Hessian matrix methods in prefabricated building component crack detection suffer from poor noise robustness and rely solely on eigenvalue magnitude, failing to effectively distinguish between genuine cracks and pseudo-ridge responses generated by concrete surface textures, pore edges, etc., leading to numerous false detections, this application provides a method for monitoring and managing the construction quality and safety of prefabricated buildings. This method introduces an eigenvalue ratio factor into the Hessian matrix ridge detection process. and gradient direction consistency factor A dual discrimination mechanism is used to construct an improved ridge response intensity. This enabled the accurate identification of the essential characteristics of cracks.

[0038] Among them, the eigenvalue ratio factor By evaluating the eigenvalue ratio To identify linear structural features, the thin-plate structure of a real crack makes... Smaller and Approaching 1, point noise or patchy texture greatly led to Approaching 0; Gradient direction consistency factor To verify whether the image gradient is perpendicular to the ridge direction, real cracks exhibit a regular gradient direction, while random textures lack this consistency. By combining the two criteria in a product form, only those that simultaneously satisfy linear feature significance ( (High) and the gradient direction are consistent ( Only a high-strength structure can achieve a strong response, fundamentally suppressing the pseudo-response of non-crack structures such as complex textures and pore edges on the concrete surface.

[0039] Furthermore, this application adopts a multi-scale analysis framework to adapt to the detection of cracks of different widths, improves the detection accuracy of contrast variation areas through local adaptive threshold processing, and connects discrete crack points into complete crack regions by utilizing the dual constraints of principal direction consistency and spatial continuity. Finally, crack region data containing key parameters such as crack location, direction, width and length are generated, which improves the detection accuracy of linear cracks in prefabricated components. Attached Figure Description

[0040] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0041] Figure 1 This is an exemplary flowchart illustrating a method for monitoring and managing the construction quality and safety of prefabricated buildings according to some embodiments of this application;

[0042] Figure 2 This is an exemplary flowchart illustrating the construction of a linear crack feature dataset according to some embodiments of this application;

[0043] Figure 3 These are detection response curves of real cracks and surface textures as shown in some embodiments of this application;

[0044] Figure 4 This is an exemplary flowchart illustrating the construction of crack region data according to some embodiments of this application. Detailed Implementation

[0045] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0046] like Figure 1 As shown, images of prefabricated components at the prefabricated building construction site are acquired; the prefabricated component images are preprocessed to obtain standardized image data; an improved Hessian matrix ridge detection method is used to perform linear structure analysis on the standardized image data, calculating the second-order partial derivative matrix of each pixel, and extracting linear structural response data through eigenvalue decomposition to generate a linear crack feature dataset containing crack location, direction, and intensity; based on the linear crack feature dataset, discrete crack response point data are converted into crack region data with continuous direction field by calculating the principal direction consistency factor in the local neighborhood; S5, based on the crack region data, prefabricated building construction quality and safety management measures are generated.

[0047] Acquire images of prefabricated components at the prefabricated building construction site; use a high-resolution digital camera or a dedicated industrial camera to photograph the prefabricated components at the construction site, acquiring surface images of the components. The camera resolution should be no less than 12 megapixels to ensure that even minute crack features can be captured. During shooting, the camera should be kept perpendicular to the surface of the prefabricated component, with the distance controlled within 0.5-2 meters to obtain clear surface details. Use uniform diffused lighting conditions, avoiding highlights and shadows caused by strong direct light sources to reduce interference with subsequent crack detection.

[0048] The original images of precast components often suffer from uneven lighting, insufficient contrast, and noise interference. A series of preprocessing steps are required to improve image quality and enhance the identifiability of crack features:

[0049] The RGB color precast component image was converted into a single-channel grayscale image I using a weighted average method: I = 0.299 × R + 0.587 × G + 0.114 × B; where R, G, and B are the red, green, and blue channel values ​​of the original image, respectively. Grayscale conversion simplifies subsequent processing while preserving the main characteristic information of the cracks.

[0050] A Gaussian filter is applied to the single-channel grayscale image I for denoising to obtain denoised image data. : Where G is a two-dimensional Gaussian kernel function: The standard deviation σ of the Gaussian filter kernel function is adaptively determined based on the surface texture characteristics of the prefabricated components. Where k is an empirical coefficient, with a value range of [0.5, 1.5]. This represents the average spacing (in pixels) of texture features in the image. For prefabricated component surfaces with high smoothness, k is taken as a smaller value; for components with rough surface texture, k is taken as a larger value. In practical applications, k is automatically estimated by analyzing the statistical properties of the image gradient histogram. value.

[0051] Based on denoised image data The contrast-limited adaptive histogram equalization algorithm (CLAHE) is used to enhance local contrast, resulting in enhanced image data. Divide the image into M×N sub-regions (typically 8×8 or 16×16), calculate and apply histogram equalization independently to each sub-region, and set a contrast limit threshold. (Typical value is 3.0-4.0) Prevents excessive noise amplification; for noise exceeding the threshold The histogram portion is cropped and redistributed to other gray levels; bilinear interpolation is used to smoothly stitch together the equalization results of each sub-region to avoid artificial artifacts at the region boundaries.

[0052] Enhanced image data Normalization is performed to linearly map pixel values ​​to the [0, 1] interval, resulting in standardized image data. : ,in, For the original enhanced image pixel values, and These are the minimum and maximum values ​​of the enhanced image data, respectively. These are the normalized pixel values. Numerical normalization unifies the range of image values, facilitating consistent parameter settings in subsequent Hessian matrix analysis.

[0053] like Figure 2 As shown, the detection scale parameter is set as follows: minimum detection scale. : Typically set to 1.0~2.0 pixels, corresponding to the finest crack structure in the image; maximum detection scale The scale factor ρ is typically set to 8.0~12.0 pixels, corresponding to the widest crack structure in the image; the scale sampling factor ρ is typically set to 1.2~1.5, controlling the growth ratio between adjacent scales; the number of scales n is typically 6~10 scales, determined by the formula... .

[0054] Constructing a scale-space sequence: Scale set: For each scale Normalized image data is processed by Gaussian convolution. Perform smoothing: ,in, For the scale Two-dimensional Gaussian kernel function: The size of the Gaussian kernel is set to... (To ensure the truncation error is less than 0.01), convolution is implemented using Fast Fourier Transform (FFT) to improve computational efficiency.

[0055] For each scale The image below Calculate the second-order partial derivatives for each pixel: Calculate the second-order partial derivatives in the x-direction. : Through the With Gaussian second-order partial derivative kernel Perform convolution on x and y; calculate the second-order partial derivative in the y-direction. : Through the With Gaussian second-order partial derivative kernel Perform convolution on yy; calculate the mixed partial derivatives. : Through the Mixed partial derivative kernel with Gaussian Convolution is performed on x and y.

[0056] Construct a scale-normalized Hessian matrix :

[0057] ,in, respectively scale The second-order partial derivatives and mixed partial derivatives of the image below, where σs² is the scale normalization factor;

[0058] For each pixel at each scale, the Hessian matrix Perform eigenvalue decomposition to obtain eigenvalues. , and the corresponding feature vectors , ;in, ; Corresponding to eigenvalues This indicates the direction of the greatest change in the local structure of the image. Corresponding to eigenvalues , representing the direction of the linear structure (crack). For eigenvalues... : For eigenvalues : .

[0059] For each scale eigenvalues ​​under , Calculate the eigenvalue ratio For an ideal linear structure, The absolute value is much greater than The absolute value, that is The values ​​are relatively large; for noise points or surface textures, the absolute values ​​of the two feature values ​​are close, i.e. Close to 1.

[0060] Based on the eigenvalue ratio The eigenvalue ratio factor at each scale is calculated using a Gaussian decay function: ;in: For scale The eigenvalue ratio factor is defined below, with a value range of [0, 1]. For scale-adaptive linear structure discrimination sensitivity parameters, ,in, The current detection scale is α, which is the scale adjustment coefficient, ranging from [1.0, 3.0]. hour, The structure was determined to be at scale The following exhibits significant linear characteristics; With detection scale Correlation ensures consistency in the discrimination of linear structures across different scales.

[0061] Computational scale The image below The gradient vector at each pixel (x, y): ;in, Using either the first-order difference operator or the Sobel operator on scaled images The above calculation is used to obtain the magnitude of the gradient vector. ; Calculate the gradient vector With scale The second eigenvector below dot product: ;in, For scale The lower Hessian matrix corresponds to smaller eigenvalues. Normalized eigenvectors; computational scale Gradient direction consistency factor: ;in: The value range of is [0, 1], indicating that at scale The degree of alignment between the downgradient direction and the direction perpendicular to the ridge line; ε is a small constant to prevent division by zero, with a value of [value missing]. By calculating the gradient at the corresponding scale, we ensure that the gradient direction matches the linear structural features detected at that scale.

[0062] Calculate the ridge response intensity at various scales ,when Retain the response if it is valid, otherwise set it to zero. The conditions ensure that only dark linear structures are detected (cracks are typically darker than the background). exist and When the numerical value is close to 0 (for nonlinear structures), it approaches 0, thus suppressing spurious responses; When the gradient direction is not perpendicular to the crack direction, it approaches 0, further filtering out spurious responses.

[0063] like Figure 3 As shown in the figure, the response intensity comparison between the traditional method and the present application is illustrated. It can be seen from the figure that the improved Hessian matrix method, by introducing an eigenvalue ratio factor... and gradient direction consistency factor The dual discrimination mechanism has significant advantages in crack detection:

[0064] In the actual crack region: both methods produce strong responses, but the improved method produces a smoother response curve, a more stable peak value, and greater adaptability to changes in crack width. Observing the AB segment of the curve, it can be seen that the improved method maintains a high response intensity to the actual crack, while the traditional method shows a significant decrease in response at the crack edge (near points A and B).

[0065] Concrete surface texture area: Traditional methods generate a large number of false high responses in the CD segment, making it difficult to distinguish from real cracks; while the improved method effectively suppresses the response in this area, reducing the response value by about 85%, significantly reducing the false detection rate.

[0066] Pore ​​edge region: Near point E, the traditional method produces a high response to the pore edge that is close to that of a real crack, while the improved method... The factor identifies the circular features of the pore edges (λ1≈λ2), effectively suppressing the response intensity.

[0067] Improved signal-to-noise ratio: The improved method increases the ratio of the response intensity of the real crack to the background texture from about 2:1 in the traditional method to about 10:1, which greatly enhances the robustness and reliability of the detection algorithm.

[0068] In summary, the improved Hessian matrix ridge detection method, by introducing a dual discrimination mechanism, achieves high-precision detection of cracks in prefabricated components of prefabricated buildings, effectively overcoming the problem that traditional methods cannot distinguish between real cracks and pseudo-ridge features such as concrete surface texture and pore edges.

[0069] Select the maximum response value and the corresponding optimal scale from all scales:

[0070] ; Local adaptive thresholding is applied to the maximum response intensity S:

[0071] For each pixel (x, y) in the image, its local neighborhood window W(x, y) is defined: Where w is the side length of the square window. , This is the largest scale in multi-scale detection. Indicates rounding up;

[0072] Calculate the mean response intensity within the local window W(x, y). : The summation range is (i, j) ∈ W(x, y). The total number of pixels within the window. Let be the maximum response intensity at pixel (i, j);

[0073] Calculate the standard deviation of the response intensity within the local window W(x,y). : The summation range is (i, j) ∈ W(x, y).

[0074] Calculate the local adaptive threshold: ;in: is the local adaptive threshold at pixel (x, y); k is the threshold coefficient, ranging from [1.5, 3.0], adjusted according to the surface texture complexity of the prefabricated component: smooth surface k = 1.5~2.0; medium texture surface k = 2.0~2.5; rough texture surface k = 2.5~3.0; when When, set ,in, This avoids generating excessively low thresholds in uniform regions.

[0075] Extract the pixel location coordinates (x, y) that meet the threshold condition and the optimal scale. The principal direction angle θ and response intensity and estimated crack width Construct a linear crack feature dataset .in: Represents the pixel coordinates of the crack point; Indicates the direction angle of the crack at that point, ranging from [0, π). The value represents the crack response intensity at that point, reflecting the salience of the crack; the value represents the estimated crack width at that point, in pixels.

[0076] like Figure 4 As shown, for the linear crack feature dataset Each crack point in Define its neighborhood search window : ;in, Let i be the neighborhood search radius of crack point i. β is the neighborhood coefficient, ranging from [2.0, 4.0], selected based on crack continuity requirements; for cracks expected to be relatively continuous (such as structural cracks in precast wall panels), β is taken as 2.0~2.5; for cracks that may be discontinuous (such as shrinkage cracks), β is taken as 3.0~4.0. A KD-tree data structure is used for efficient neighborhood point search, avoiding the computational overhead of brute-force search. For each crack point, it is used as a query point, and the radius is searched in the KD-tree. All points within the range.

[0077] Calculate the principal direction consistency factor among crack points in the neighborhood. : ;in, and are the principal direction angles of crack points i and j, respectively, with values ​​ranging from [0, π]. This is the tolerance parameter for angle differences. Corresponding to an angle tolerance of 15°; using Instead of a direct angle difference, this solves the problem of angle periodicity, ensuring accurate calculation of angle differences near 0° and 180°. This is especially true when the two points are in the same direction. When the directional difference increases, Gradually decrease; when the directional difference is greater than approximately 45°... A value close to 0 indicates that the two points are unlikely to belong to the same crack.

[0078] Calculate the spatial continuity factor between crack points in the neighborhood. : ;in: , is the perpendicular distance from point j to the crack direction line where point i is located; The average width at two points is used to control the tolerance for spatial continuity; the tolerance is correlated with the crack width, with a wider crack allowing for greater spatial deviation. When point j is located on the crack direction line of point i, , As the vertical distance increases, Rapidly decreases; when the vertical distance is greater than 2 to 3 times the average crack width, Close to 0.

[0079] Calculate the overall connection strength: ;in, Ensure that the connection strength does not exceed the response strength of the weakest point;

[0080] when At that time, crack points i and j are marked as the same crack region, where the connection threshold is: γ is the connection coefficient, with a value range of [0.3, 0.5]. Smaller γ values ​​(such as 0.3) result in more lenient connection conditions, suitable for cracks that may have discontinuities; larger γ values ​​(such as 0.5) emphasize higher connection reliability, suitable for clear cracks that require precise identification.

[0081] By using a region growing algorithm, adjacent crack points with high connectivity are iteratively merged to generate a continuous set of crack regions. Each region Includes a set of points belonging to the same crack; for each formed crack region If it contains fewer points than the threshold (Typically set to 5~10), then it is considered noise and removed from the region set. This helps filter out false positives caused by local texture or image noise.

[0082] For each crack region Calculate the region attributes:

[0083] Centerline coordinate sequence Fit the centerline of points within the region using the least squares method;

[0084] Average width ,in, For the region The number of points within;

[0085] Total length Along the center line The cumulative arc length;

[0086] Average direction The main direction obtained through principal component analysis;

[0087] Maximum response strength ;

[0088] Constructing crack region data .

[0089] Based on the crack area data, construction quality and safety management measures for prefabricated buildings are generated. Cracks are categorized according to location, shape, and cause as follows: Structural cracks: related to load-bearing capacity, potentially affecting structural safety; Non-structural cracks: primarily affecting aesthetics and functionality; Connection cracks: appearing at the joints of prefabricated components; Internal component cracks: appearing within the prefabricated component itself.

[0090] Calculate the crack severity index: ,in: Crack area Severity index; These represent the average width, total length, and maximum response intensity of the crack region, respectively. These are the corresponding threshold parameters (determined based on standards and experience); Let be the weighting coefficient, satisfying .

[0091] Classifying safety risk levels:

[0092] Low risk (Level I): SI < 0.3, minor cracks, no safety hazards;

[0093] Low to medium risk (Level II): 0.3 ≤ SI < 0.5, minor cracks, requiring regular observation;

[0094] Medium risk (Level III): 0.5 ≤ SI < 0.7, obvious cracks, treatment plan needs to be developed;

[0095] High risk (Level IV): 0.7 ≤ SI < 0.9, severe cracks, requiring immediate treatment;

[0096] Extremely high risk (Level V): SI ≥ 0.9, dangerous crack, work must be stopped immediately.

[0097] Level I (Low Risk): Record kept, regular checks conducted;

[0098] Level II (Low to Medium Risk): Increase the frequency of inspections and develop an observation plan;

[0099] Level III (Medium Risk): Professional personnel conduct on-site assessment and develop a repair plan;

[0100] Level IV (High Risk): Immediately implement reinforcement and repair measures, and suspend construction in the relevant areas;

[0101] Level V (Extremely High Risk): Immediately halt work, evacuate personnel, urgently reinforce the structure, and reassess structural safety.

[0102] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A prefabricated building construction quality safety management monitoring method, characterized in that, The method comprises the following steps: acquiring a prefabricated component image of a prefabricated building construction site; preprocessing the prefabricated component image to obtain standardized image data; performing linear structure analysis on the standardized image data by using an improved Hessian matrix ridge line detection method, calculating a second-order partial derivative matrix of each pixel point, extracting linear structure response data through eigenvalue decomposition, and generating a linear crack feature data set containing crack position, direction and intensity; according to the linear crack feature data set, converting discrete crack response point data into crack region data with continuous direction field by calculating a main direction consistency factor in a local neighborhood; generating prefabricated building construction quality and safety management measures according to the crack region data. Wherein, performing linear structure analysis on the standardized image data by using an improved Hessian matrix ridge line detection method, comprising: At multiple scales The normalized image data Gaussian smoothing is performed to obtain a sequence of scale space images wherein, , and are the minimum and maximum detection scales, respectively, is the scale sampling factor; For each scale of the image The second-order partial derivatives of each pixel are calculated to construct the scale-normalized Hessian matrix : ; where are the second-order partial derivatives and the mixed partial derivatives of the image at scale , respectively, is the scale normalization factor; Hessian matrix of each pixel point under each scale Eigenvalue decomposition is performed to obtain eigenvalues , and corresponding eigenvectors , ; wherein ; According to the eigenvalue , , the eigenvalue ratio factor at each scale is calculated ; According to the image gradient vector and the feature vector , a gradient direction consistency factor of each pixel point under each scale is calculated ; The ridge line response intensity under each scale is calculated When the response is reserved, otherwise it is zero, ensuring that only dark linear structures are detected; Select the maximum response value and the corresponding optimal scale among all scales: ; ; locally adaptive thresholding of the maximum response intensity S processing: ; Extracting pixel point position coordinates (x, y) satisfying threshold conditions, main direction angle θ under the optimal scale, response intensity and estimated crack width, constructing a linear crack feature data set ； ​​​ Wherein, According to the eigenvalue , , the eigenvalue ratio factor at each scale is calculated , comprising: For each scale under the eigenvalue , , the eigenvalue ratio is calculated; According to the eigenvalue ratio The eigenvalue ratio factor at each scale is calculated by a Gaussian-type attenuation function: ; wherein: is the eigenvalue ratio factor at scale , and the value range is [0, 1]; is a scale-adaptive linear structure discrimination sensitivity parameter, , wherein, is the current detection scale, is a scale adjustment coefficient, and the value range is [1.0, 3.0]; Wherein, According to the image gradient vector and the feature vector , a gradient direction consistency factor of each pixel point under each scale is calculated , comprising: Computing the scale The image under the scale The gradient vector at each pixel point (x, y): ; wherein, are respectively calculated on the scale image by a first-order difference operator or a Sobel operator. According to the gradient vector and the scale vector , the direction consistency factor of the pixel point at the scale is calculated : wherein ε is a small constant to prevent division by zero. Wherein, Local adaptive threshold : wherein, is the mean and standard deviation of the response intensity within the local window; k is a threshold coefficient; converting discrete crack response point data into crack region data with continuous direction field, comprising: linear crack feature dataset Each crack point in Define its neighborhood search window : ;in, Let i be the neighborhood search radius of crack point i. β is the neighborhood coefficient; Computing a main direction consistency factor between fracture points within a neighborhood ; Computing a spatial continuity factor between fracture points within a neighborhood ; The calculated integrated connection strength is: ; wherein, Ensuring that the connection strength does not exceed the response strength of the weaker point. When the crack points i and j are marked as the same crack region; wherein, is a connection threshold value; A set of continuous crack regions is generated by a region growing algorithm wherein each region contains a set of points belonging to the same crack region; For each fracture zone , calculate zone properties: centerline coordinate sequence, average width, total length, average direction, and maximum response intensity, construct fracture zone data ； Computing a main direction consistency factor between fracture points within a neighborhood : ; wherein, and are the main direction angles of fracture points i and j, respectively; is an angle difference tolerance parameter; Computing a spatial continuity factor between fracture points within a neighborhood : , where, is the perpendicular distance from point j to the fracture direction line on which point i lies, is the average width of the two points.

2. The prefabricated building construction quality and safety management monitoring method according to claim 1, characterized in that: the preprocessing of the prefabricated component image comprises: converting the prefabricated component image into a single-channel grayscale image; performing Gaussian filtering processing on the single-channel grayscale image to obtain a denoising image; enhancing the local contrast by using an adaptive histogram equalization algorithm according to the denoising image to obtain enhanced image data; performing normalization processing on the enhanced image data to map the pixel value to the interval of 0 to 1, and obtaining standardized image data.

3. A prefabricated building construction quality safety management monitoring system for performing the prefabricated building construction quality safety management monitoring method according to any one of claims 1-2, characterized in that, The method comprises the following steps: an image acquisition module acquires a prefabricated component image of a prefabricated building construction site; and converts the collected image into standardized image data; a linear analysis module performs linear structure analysis on the standardized image data by using an improved Hessian matrix ridge line detection method, calculates a second-order partial derivative matrix of each pixel point, extracts linear structure response data through eigenvalue decomposition, and generates a linear crack feature data set containing crack position, direction and intensity; a crack analysis module converts discrete crack response point data into crack region data with continuous direction field by calculating a main direction consistency factor in a local neighborhood according to the linear crack feature data set; a safety management module generates prefabricated building construction quality and safety management measures according to the crack region data.

Images