A machine vision-based intelligent inspection method and system for pipe surface quality

Abstract

This application provides a machine vision-based intelligent inspection method and system for pipe surface quality, relating to the fields of industrial automation inspection and machine vision technology. This application establishes a pipe surface feature library and labels abnormal shape features. Images of the pipe are acquired under a combined bright and dark light field. After polarization filtering, grayscale and texture features are extracted. Principal component analysis is used to reduce dimensionality and fuse the feature set with the feature library to reconstruct a three-dimensional point cloud covering the mortar and concrete layers. The point cloud is then converted into a two-dimensional unfolded image via cylindrical projection, segmenting mortar and concrete abnormality areas. In the mortar area, the volume of pores is calculated using point cloud residuals, and internal voids are detected by combining acoustic vibration excitation and thermal response. In the concrete area, defects are located based on point cloud features. According to the process sequence of concrete application preceding mortar application, an association model is constructed to match overlapping defect sites and output the detection results, improving the automation and accuracy of the inspection.

Description

Technical Field

[0001] This application relates to the fields of industrial automation inspection and machine vision technology, and in particular to an intelligent inspection method and system for pipe surface quality based on machine vision. Background Technology

[0002] Concrete pipes are widely used in critical projects such as urban water supply and drainage and water transmission trunk lines. Their mortar lining layer, as a protective structure covering the concrete base layer, is susceptible to damage such as detachment or voids, which directly threaten the structural integrity and service life of the pipeline. Therefore, it is necessary to accurately locate the areas of mortar lining detachment and quantify the volume of voids to assess repair costs.

[0003] Currently, the mainstream solution employs a pipeline inspection robot equipped with a high-definition camera, which collects image data of the pipeline's inner wall through internal movement. The robot travels along the pipeline's axis, acquiring surface images using visible light or multimodal imaging equipment, and optimizing image quality through image preprocessing.

[0004] However, while this solution achieves automated detection, it has significant limitations. First, it relies on two-dimensional image analysis and cannot calculate the three-dimensional volume of the hole, resulting in a lack of material usage basis for the repair plan. Second, image processing is susceptible to uneven lighting, surface texture interference, and water stains, leading to a high false detection rate, especially at the interface between the mortar layer and the concrete base layer, where it is difficult to distinguish the defect source layer. Summary of the Invention

[0005] The purpose of this application is to provide a machine vision-based intelligent inspection method and system for pipe surface quality, in order to solve the problem of high false detection rate in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a machine vision-based intelligent inspection method for pipe surface quality, comprising:

[0007] A pipe surface feature library is pre-established, and abnormal shape features are marked on the detached areas in the surface feature library;

[0008] Images of the pipe surface under a combined light and dark field are acquired, polarization filtering is applied to the images, and principal component analysis is performed on the grayscale and texture features extracted from the processed images to obtain the principal component analysis results.

[0009] A fusion feature set is generated based on the grayscale features, texture features and principal component analysis results, and the fusion feature set is matched with the surface feature library. Based on the matching results, a three-dimensional point cloud covering the mortar layer and concrete layer is reconstructed.

[0010] The three-dimensional point cloud is converted into a two-dimensional unfolded image by cylindrical projection. Based on the shape anomaly features in the surface feature library, the abnormal regions in the two-dimensional unfolded image are segmented to obtain mortar anomaly regions and concrete anomaly regions.

[0011] In the mortar anomaly zone, the cavity volume is calculated based on the residual of the three-dimensional point cloud, and the internal voids are detected by combining acoustic vibration excitation with thermal response to generate mortar defect sites including the cavity volume. In the concrete anomaly zone, the concrete defect sites are obtained based on the feature analysis of the three-dimensional point cloud.

[0012] Based on the process sequence of concrete being applied before mortar, an association matching model is constructed between the mortar defect sites and the concrete defect sites. The association matching model is then used to determine whether the defect sites overlap. If they overlap, it is determined that the preceding process has an impact, and a test result is generated based on the determination result.

[0013] Optionally, based on the process sequence of concrete application preceding mortar application, an association matching model is constructed between the mortar defect sites and the concrete defect sites. This model is then used to determine whether the defect sites overlap. If they overlap, it is determined to be due to the influence of the preceding process. Based on this determination, a detection result is generated, including:

[0014] According to the construction sequence of concrete layer covering mortar layer, a vertical projection relationship of spatial coordinates from the concrete defect site to the mortar defect site is established, and an association matching model is formed based on the vertical projection relationship.

[0015] The correlation matching model is used to compare the coordinate intersection range of the mortar defect site and the concrete defect site. When the area of ​​the coordinate intersection range exceeds a preset threshold, it is marked as a defect affected by the previous process.

[0016] Summarize the spatial distribution information of all defects affected by the preceding processes to generate inspection results that include process responsibility attribution.

[0017] Optionally, the coordinate intersection range of the mortar defect site and the concrete defect site is compared using the correlation matching model. When the area of ​​the coordinate intersection range exceeds a preset threshold, it is marked as a defect affecting the previous process, including:

[0018] The spatial coordinates of the concrete defect sites are mapped to the mortar layer coordinate system plane through coordinate transformation rules to form a set of projected coordinate points. A circular region is constructed based on each projected point in the set of projected coordinate points.

[0019] The circular area is compared with the actual coordinates of the mortar defect sites. The intersection area value of the circular area and the actual mortar defect sites is calculated. The intersection area values ​​corresponding to all projected coordinate points are summed to form the total area value.

[0020] When the total area value exceeds a preset area threshold, the location corresponding to the projected coordinate point is marked as a defect affecting the previous process.

[0021] Optionally, in the mortar anomaly zone, the void volume is calculated based on the residual of the three-dimensional point cloud, and internal voids are detected by combining acoustic vibration excitation with thermal response to generate mortar defect sites including void volumes. In the concrete anomaly zone, concrete defect sites are obtained based on feature analysis of the three-dimensional point cloud, including:

[0022] In the mortar anomaly zone, the positional deviation between the actual surface points of the three-dimensional point cloud and the theoretical model points in the surface feature library is calculated, and all the positional deviations are summed to form a total depth deviation. The hole volume is then derived based on the total depth deviation.

[0023] Mechanical vibration waves are applied to the abnormal mortar area, and the temperature change curve of the infrared thermal imager is recorded simultaneously. When the peak rate of the temperature change curve exceeds the set rate threshold, the existence of voids is confirmed, and the mortar defect site containing the volume of the voids is output.

[0024] In the abnormal concrete area, the difference in the normal vector angle between adjacent points in the three-dimensional point cloud is detected. When the value of the difference in the normal vector angle exceeds a preset angle threshold, the spatial position corresponding to the difference in the normal vector angle is marked as the concrete defect site.

[0025] Optionally, a fused feature set is generated based on the grayscale features, texture features, and principal component analysis results. The fused feature set is then matched with the surface feature library. Based on the matching results, a three-dimensional point cloud covering the mortar and concrete layers is reconstructed, including:

[0026] The grayscale features, texture features, and principal component analysis results are merged according to their spatial locations to generate a fusion feature set containing multidimensional parameters;

[0027] The similarity between each element in the fused feature set and the feature patterns in the surface feature library is calculated to obtain a similarity value, and matching points with similarity values ​​higher than a preset threshold are selected.

[0028] Combine the three-dimensional spatial coordinates of all the matching points to generate a discrete point set, and reconstruct a three-dimensional point cloud covering the mortar layer and the concrete layer based on the discrete point set.

[0029] Optionally, an image of the pipe surface under a combined bright and dark light field is acquired, the image is subjected to polarization filtering, and principal component analysis is performed on the processed image to extract grayscale and texture features, obtaining the principal component analysis results, including:

[0030] Multiple light sources illuminate the surface of the pipe from different directions to form a composite light field of light and dark, and images under the composite light field of light and dark are collected.

[0031] A polarization filter is superimposed on the image to eliminate interference components caused by specular reflection and retain the effective light signal on the pipe surface;

[0032] From the image after polarization filtering, brightness values ​​are extracted pixel by pixel as grayscale features, and directional change information is statistically analyzed within the range of adjacent pixels as texture features.

[0033] The grayscale features and the texture features are combined into an initial dataset, and a linear transformation is performed on the initial dataset to obtain principal component analysis results representing the main direction of variation.

[0034] Optionally, the three-dimensional point cloud is converted into a two-dimensional unfolded diagram via cylindrical projection. Based on the shape anomaly features in the surface feature library, the abnormal regions in the two-dimensional unfolded diagram are segmented to obtain mortar anomaly regions and concrete anomaly regions, including:

[0035] Extract the three-dimensional coordinate points from the three-dimensional point cloud, calculate the arc angle and axial height value of each three-dimensional coordinate point in the circumferential direction of the pipe, map the arc angle and axial height value to two-dimensional plane coordinate points, and connect the two-dimensional plane coordinate points to generate a continuously distributed two-dimensional unfolded diagram;

[0036] Read the contour coordinate data of the shape anomaly features from the surface feature library, and calculate the weighted sum of the distances between each pixel and the contour coordinate data on the two-dimensional unfolded map;

[0037] When the value of the distance-weighted sum is less than a preset similarity threshold, it is marked as an anomaly point, and adjacent anomaly points are connected to form a closed discontinuous region;

[0038] Obtain the boundary line coordinates between the mortar layer and the concrete layer in the two-dimensional unfolded diagram, calculate the difference between the center coordinates of each discontinuous region and the boundary line coordinates, and classify the discontinuous region into the mortar abnormal region when the difference is greater than zero, and classify the discontinuous region into the concrete abnormal region when the difference is less than zero.

[0039] Secondly, this application provides a machine vision-based intelligent inspection system for pipe surface quality, comprising:

[0040] A module is established to pre-build a pipe surface feature library and mark the abnormal shape features of the detached areas in the surface feature library.

[0041] The analysis module is used to acquire images of the pipe surface under a combined light field of light and dark, perform polarization filtering on the images, extract grayscale features and texture features from the processed images, perform principal component analysis, and obtain principal component analysis results.

[0042] The matching module is used to generate a fusion feature set based on the grayscale features, texture features and principal component analysis results, and to perform feature matching between the fusion feature set and the surface feature library, and to reconstruct the three-dimensional point cloud covering the mortar layer and the concrete layer based on the matching results.

[0043] The segmentation module is used to convert the three-dimensional point cloud into a two-dimensional unfolded diagram through cylindrical projection, and to segment the abnormal areas in the two-dimensional unfolded diagram according to the shape abnormal features in the surface feature library, so as to obtain mortar abnormal areas and concrete abnormal areas.

[0044] The calculation module is used to calculate the volume of pores in the mortar anomaly zone based on the residual of the three-dimensional point cloud, and to detect internal voids by combining acoustic vibration excitation with thermal response to generate mortar defect sites including the volume of pores. The module is also used to obtain concrete defect sites in the concrete anomaly zone based on the feature analysis of the three-dimensional point cloud.

[0045] The generation module is used to construct an association matching model between the mortar defect sites and the concrete defect sites according to the process sequence of concrete being applied before mortar, and to use the association matching model to determine whether the defect sites overlap. If they overlap, it is determined that the preceding process has an impact, and the detection result is generated based on the judgment result.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor is configured to execute the computer program to implement the steps of a machine vision-based intelligent inspection method for pipe surface quality as described in the first aspect above.

[0049] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the intelligent detection method for pipe surface quality based on machine vision as described in the first aspect above.

[0050] This application provides a machine vision-based intelligent inspection method for pipe surface quality. By pre-establishing a pipe surface feature library and annotating the shape anomalies of detached areas, a standardized defect sample database can be constructed, providing a benchmark for subsequent feature matching and improving the targeting of detachment defects. By acquiring images under combined bright and dark light fields and performing polarization filtering, specular reflection interference from metal or highly reflective surfaces can be effectively suppressed, enhancing the extraction accuracy of grayscale and texture features and solving the imaging distortion problem caused by complex lighting environments. By extracting grayscale and texture features from the images and performing principal component analysis, dimensionality reduction and redundant information can be removed, retaining key defect features and improving the discriminative power and computational efficiency of the feature set. By fusing the feature set with the surface feature library and reconstructing a three-dimensional point cloud, it is possible to detect mortar layers and concrete layers. Three-dimensional spatial modeling provides a geometric basis for quantifying volumetric defects, overcoming the limitation of two-dimensional images in representing depth information. By converting 3D point clouds into 2D unfolded maps and segmenting abnormal regions, surface detection can be transformed into planar analysis, simplifying defect localization on complex surfaces. Simultaneously, it distinguishes independent abnormal regions in mortar and concrete layers, solving the problem of layered defect identification. By calculating point cloud residuals in mortar abnormal areas to quantify pore volume and combining acoustic vibration excitation and thermal response to detect voids, it integrates morphological and physical property analysis to achieve accurate measurement of pore volume and non-destructive testing of internal void defects, overcoming the limitations of single vision technology. By constructing a correlation matching model of mortar and concrete defect sites and determining overlap, it is possible to trace the transmission influence of concrete base layer defects on the mortar layer, achieving process-based tracing of defect causes and providing a basis for process optimization. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating a machine vision-based intelligent inspection method for pipe surface quality provided in this application embodiment;

[0053] Figure 2 A schematic diagram illustrating a specific implementation of a machine vision-based intelligent inspection method for pipe surface quality provided in this application embodiment;

[0054] Figure 3 A scene diagram illustrating a machine vision-based intelligent inspection method for pipe surface quality provided in an embodiment of this application;

[0055] Figure 4This is a schematic diagram of a machine vision-based intelligent inspection system for pipe surface quality, provided in an embodiment of this application. Detailed Implementation

[0056] Research has found that while current automated pipe surface quality inspection technologies can achieve basic automation, they have significant limitations. First, they rely on two-dimensional image analysis and cannot calculate the three-dimensional volume of holes, resulting in a lack of material quantity basis for repair plans. Second, image processing is susceptible to uneven lighting, surface texture interference, and water stains, leading to a high false detection rate, especially at the interface between the mortar layer and the concrete substrate, where it is difficult to distinguish the defect source layer. Therefore, there is an urgent need for an intelligent pipe surface quality inspection method based on three-dimensional reconstruction and intelligent correlation matching to improve inspection accuracy, quantify defect volume, and accurately determine the defect source.

[0057] To address the aforementioned issues, this invention proposes an intelligent inspection method for pipe surface quality based on machine vision. Its core lies in achieving three-dimensional quantitative detection and accurate differentiation of defect sources through 3D point cloud reconstruction and multi-feature fusion analysis. Specifically, a pipe surface feature library is pre-established and anomaly features are labeled; images under combined bright and dark light fields are acquired, and after polarization filtering, grayscale and texture features are extracted for principal component analysis; a fusion feature set is generated to match the feature library to reconstruct a 3D point cloud covering the mortar and concrete layers; the point cloud is converted into a two-dimensional unfolded image through cylindrical projection and abnormal regions are segmented; in the mortar abnormal region, the void volume is calculated based on the point cloud residual and hollow areas are detected by combining acoustic vibration excitation thermal response; in the concrete abnormal region, features are analyzed to obtain defect locations; finally, an association matching model is constructed to determine the overlap of defect locations to generate detection results. This method can accurately calculate the three-dimensional volume of voids, significantly reduce the false detection rate caused by uneven illumination and texture interference through polarization filtering and feature fusion, and effectively differentiate the defect sources of mortar and concrete layers using the association matching model.

[0058] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] The core of this application is to provide an intelligent inspection method for pipe surface quality based on machine vision, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0060] S101. Establish a pipe surface feature library in advance, and mark the abnormal shape features of the detached areas in the surface feature library;

[0061] In the above steps, the pipe surface feature library refers to a pre-built database used to store typical images and feature data of various defects on the pipe surface, such as cracks, scratches or pits; the detached area refers to the local defect area on the pipe surface formed by material peeling off, such as the peeling of mortar layer or concrete layer; the abnormal shape features refer to parameters that quantify the geometric shape of the detached area, such as contour irregularity, edge breakage or area ratio, used to distinguish it from normal surface structure.

[0062] In this embodiment, firstly, a pipe surface feature library is established and the shape anomaly features of the detached area are labeled in step S101. The specific process is as follows: First, pipe sample images containing various surface defects are collected, such as bright field, dark field, or polarized light images; then, defect areas are extracted using image segmentation techniques such as thresholding or edge detection. During this process, noise reduction algorithms, such as median filtering, are used to reduce noise interference, and morphological operations, such as erosion and dilation, are combined to separate target defect areas, such as pitting or detached areas. Next, the segmented detached areas are labeled and quantized, with a focus on extracting shape feature parameters, including contour irregularity. The calculation formula for this parameter is: ,in, Indicates the irregularity index. Indicates the perimeter of the defect. This indicates the area of ​​the defect; it also includes edge fragmentation, which is the number of jagged vertices on the defect boundary per unit length, and area percentage, which is the proportion of the defect area to the total area of ​​the inspection area. For example, for a detached area on the surface of a concrete pipe with an area of ​​15 mm², and a total inspection area of ​​100 mm², the area percentage is calculated as 15 / 100 = 0.15 (i.e., 15%). This value is marked as abnormal if it exceeds a preset threshold such as 10%. Simultaneously, the irregularity index I is calculated. Assuming a perimeter P = 10 mm and an area A = 15 mm², then: However, due to the fact that actual abnormal situations usually require For example, if in another example ,but: .when Furthermore, when the value exceeds a threshold of 1.5, the shape anomaly is confirmed. Finally, these quantified feature parameters are associated with and stored with the corresponding defect images to construct a structured feature library. For example, an independent data table is created for mortar layer detachment to store information such as image ID, irregularity, fragmentation, or area ratio. Threshold ranges for shape anomaly features are set to complete the annotation work.

[0063] In practical applications, at an aluminum alloy pipe manufacturing plant, to improve the automatic identification capability of pipe surface defects, technicians first established a pipe surface feature library. The plant collected typical surface defect samples of Class B structural pipes, such as LF6M aluminum alloy cold-drawn pipes, including detachment areas, scratches, and dents. For detachment areas, technicians used image annotation tools to precisely outline their contours and marked them as shape anomalies. For example, a 3D scanner measurement revealed that the edge of a detachment area was irregularly serrated, with a central depression depth of 0.04 mm, which is within the standard allowable defect depth range. The feature library also integrates size rules for different defects, such as stipulating that the allowable defect area on the structural pipe surface should not exceed 3% of the total surface area. If the surface area of ​​a single pipe is 2 square meters, then the upper limit of the total defect area must be controlled within 0.06 square meters. Through repeated iterative annotation, the feature library gradually covered abnormal morphologies under various working conditions, providing benchmark data for subsequent intelligent detection systems.

[0064] In the overall solution of step S101 above, by systematically integrating the morphological data of typical surface defects, a standardized feature library is constructed, and the abnormal shape features of the detached area are marked, which significantly improves the accuracy and efficiency of defect identification. This feature library provides a benchmark reference for automated detection systems. Combined with machine vision or artificial intelligence algorithms, it can quickly match abnormal areas in real-time detection images, reduce the risk of missed or misjudged detection by visual inspection, and at the same time, by accurately marking the geometric abnormal features of the detached area, it helps to quantitatively assess the potential impact of defects on the strength, sealing performance and corrosion resistance of the pipe, thereby optimizing the quality judgment criteria and providing key parameter support for subsequent repair processes, ultimately realizing the intelligent upgrade of the quality control of the entire life cycle of the pipe.

[0065] S102. Acquire an image of the pipe surface under a combined light and dark light field, perform polarization filtering on the image, and extract grayscale features and texture features from the processed image to perform principal component analysis and obtain the principal component analysis results.

[0066] Optionally, step S102 may specifically include the following steps:

[0067] S1021. Illuminate the surface of the pipe with multiple light sources from different directions to form a composite light field of light and dark, and collect an image under the composite light field of light and dark.

[0068] S1022. A polarization filter is superimposed on the image to eliminate interference components caused by specular reflection and retain the effective light signal on the surface of the pipe.

[0069] S1023. From the image after polarization filtering, extract the brightness value as grayscale feature pixel by pixel, and statistically analyze the direction change information within the range of adjacent pixels as texture feature.

[0070] S1024. Combine the grayscale features and the texture features into an initial dataset, perform a linear transformation on the initial dataset, and obtain principal component analysis results representing the main direction of variation.

[0071] In the above steps, the bright and dark composite light field refers to the lighting environment formed by combining light sources from different directions. The high-angle bright field highlights the surface texture details, while the low-angle dark field enhances the shadow contrast of the recessed areas, thus simultaneously presenting the macroscopic and microscopic features of the pipe surface. Polarization filtering refers to superimposing a polarization filter device in front of the camera lens to eliminate specular reflection interference from the metal substrate by blocking reflected light in a specific polarization direction, while retaining the diffuse reflection signal of non-metallic layers such as mortar and concrete to improve image clarity. Grayscale features refer to the brightness value sequence of each pixel in the image, reflecting the brightness and darkness distribution of a local area. Texture features refer to the pattern regularity described by statistically analyzing the brightness variation amplitude and direction of adjacent pixels, such as gradient values ​​or roughness, used to quantify the uniformity or anomalies of the surface structure. Principal component analysis results refer to the dimensionality-reduced feature vectors representing the main variation directions of the data extracted by eigenvalue decomposition after standardizing and calculating the covariance matrix of the combined grayscale and texture feature datasets, used to compress redundant information and retain core difference features.

[0072] In this embodiment, firstly, in step S1021, a ring-shaped high-angle light source, such as an LED array ring light, is symmetrically arranged on both sides of the pipe along its axial direction to illuminate the area at a 60° incident angle, ensuring uniform coverage of the detection area with bright field light to clearly reveal the rough texture of the mortar layer. Simultaneously, a low-angle light source, such as a strip light source, is arranged around the pipe at a 10° incident angle to obliquely illuminate the edges of defects such as holes in the concrete layer, creating significant shadows. By adjusting the position and angle of the light sources, the bright and dark fields of light are superimposed to form a composite lighting environment. Finally, an industrial camera acquires images under synchronous triggering conditions. For example, for a 100mm diameter concrete pipe, the inner diameter of the ring light is set to 120mm, the distance of the strip light source from the pipe surface is 50mm, and the camera exposure time is 5ms to capture a high-contrast image without motion blur.

[0073] Secondly, in step S1022, a 0-degree linear polarizer is installed in front of the light source to unify the polarization direction of the incident light. Simultaneously, a 90-degree linear polarizer is superimposed in front of the camera lens, with the transmission axis perpendicular to the polarization direction of the light source. When light shines on the pipe surface, the specular reflection light from the metal substrate is completely filtered out by the camera polarizer because its polarization direction remains unchanged. Meanwhile, the diffuse reflection light from non-metallic surfaces such as the mortar layer passes through the 90-degree polarizer due to its random polarization, thus eliminating overexposure interference and improving defect visibility. The effect of this process is verified by calculating the image contrast: the contrast calculation formula is: Where C represents contrast. This represents the average grayscale value of the defect area. This represents the average grayscale value of the background area. For example, when inspecting galvanized steel pipes, the original average grayscale value of the scratched area... 210, background area If the value is 180, then the contrast ratio is as follows: Grayscale of the scratch area after polarization processing Reduced to 120, background Increased to 185, the contrast ratio becomes: The actual measured value is improved; therefore, polarization simultaneously suppresses background interference and enhances the significance of defects.

[0074] Next, in step S1023, the polarized RGB image is converted into a single-channel grayscale image. Brightness values ​​ranging from 0 to 255 are read pixel by pixel to generate a grayscale feature matrix. Simultaneously, texture features are calculated within a local window. Roughness is calculated using a 3x3 pixel neighborhood; the brightness difference between the center pixel and its 8 neighboring pixels is calculated, and the maximum difference is taken as the roughness. : ,in The grayscale of the center pixel, The grayscale value is the value of the neighboring pixels. The directional gradient is calculated using the Sobel operator to compute the horizontal gradient. and vertical gradient :

[0075]

[0076] in, A 3x3 pixel block represents the convolution operation, and the gradient direction is... Depend on Calculation. For example, the grayscale of the center pixel in the pitted area of ​​a concrete pipe. The value is 120, and the neighboring grayscale value is... Maximum difference The horizontal gradient is |120-110-|=10. The vertical gradient is 8. -3, direction atan2(-3,8) is approximately -20.6 degrees.

[0077] Finally, in step S1024, the grayscale feature matrix (e.g., 1000x1 dimension) and the texture feature matrix (e.g., 1000x4 dimension) are merged column-wise to form an initial dataset of 1000x5 dimensions. The dataset is then standardized so that the mean of each feature dimension is 0 and the standard deviation is 1. The standardization formula is: ,in, Let the j-th dimension feature value be denoted as . Let j be the mean of the feature in dimension j. Let be the standard deviation of the j-th feature dimension. For example, the grayscale feature column. for The original value of a certain pixel is 20. If the value is 145, then the standardized value is: 。 Calculate the covariance matrix C of the standardized data: , where matrix elements This represents the degree of linear correlation between feature k and feature m, where m is the number of samples. The eigenvalues ​​are obtained by eigenvalue decomposition of the covariance matrix C. And the eigenvector v, according to Sort the eigenvectors from largest to smallest, and select the first k eigenvectors to form the projection matrix. Projecting the original data into a new space: ， The projected m-k dimensional matrix Y is the result of principal component analysis. For example, with m = 1000 pixels and k = 3, the variance retention rate is 95%, improving the accuracy of defect classification.

[0078] In a practical application, during the inspection of aluminum alloy pipes, technicians used three sets of light sources to illuminate the pipe surface from different directions, creating a composite light field environment of light and shadow. The high-angle light source illuminated the surface at a 60° tilt angle to highlight its unevenness, the low-angle light source at a 30° tilt angle to enhance the contrast of scratches, and the coaxial light source illuminated perpendicularly to capture color differences. After acquiring images of the pipe surface using a black-and-white line scan camera at a scanning speed of 2000 lines per second, a linear polarizing filter with a 0° polarization direction was added in front of the lens to eliminate strong light interference caused by specular reflection on the metal surface. For example, the intensity of the original reflected light was reduced from 180 Lux to 40 Lux, while retaining the scattered light signal to clearly present defect features such as oxide white spots. When extracting grayscale features pixel by pixel from the processed image, taking a suspected white spot area as an example: a 5×5 pixel window is selected, and the brightness value of the center pixel P(128,79) is calculated to be 85 (grayscale range 0-255). The brightness values ​​of its eight adjacent pixels are [72, 80, 91, 68, 89, 75, 82,94], and the average brightness μ is 81.375. The absolute difference between the center pixel and the mean is |85-81.375|=3.625. In texture feature calculation, the brightness gradient change of adjacent pixels in the horizontal direction within the same window is statistically analyzed: the sum of squares of the brightness difference between adjacent pixel pairs along the X-axis is 1482, and the average gradient magnitude is 13.6, indicating strong local directional changes. The grayscale feature (85) and texture feature (13.6) are combined into an initial dataset. Principal component analysis is performed on the dataset containing 100,000 sample points. First, the grayscale feature column is centered. For example, the grayscale value of a sample is 85, and the mean of the entire column is 120, resulting in -35. The texture feature column is 13.6, and the mean of the entire column is 8.2, resulting in 5.4. When calculating the covariance matrix, the covariance between grayscale and texture features is -28.7. After eigenvalue decomposition, the eigenvector [0.87, -0.49] corresponding to the largest eigenvalue λ1=12.3 represents the main direction of variation, indicating that the joint feature of grayscale reduction and texture enhancement can explain 82% of the data variation. The principal component analysis results are used for defect classification.

[0079] In the overall scheme of step 102 above, macroscopic contour features are analyzed by bright field illumination, and the contrast of scattered signals of micro-defects such as scratches and dents is enhanced by dark field low-angle illumination, which effectively overcomes the problem of uneven illumination on the surface of curved pipes. At the same time, polarization filter devices are superimposed to suppress specular reflection glare on the metal surface, and retain the diffuse reflection texture details of oxide scale peeling and micro-cracks, which greatly improves the image signal-to-noise ratio. On this basis, the brightness distribution is extracted pixel by pixel as grayscale features, and a multi-dimensional feature space is constructed by combining texture direction change information such as neighborhood edge gradient and local grayscale variance. Then, linear dimensionality reduction is performed by principal component analysis to extract feature vectors that represent core abnormal components such as discontinuous boundaries and thickness abrupt changes in the detachment area. While compressing data redundancy, the robustness of defect feature expression is enhanced, and finally, core technical support is provided for the accurate classification, in-depth quantification and efficient detection of pipe defects.

[0080] S103. Generate a fusion feature set based on the grayscale features, texture features and principal component analysis results, and perform feature matching between the fusion feature set and the surface feature library, and reconstruct the three-dimensional point cloud covering the mortar layer and concrete layer based on the matching results.

[0081] Optionally, step S103 may specifically include the following steps:

[0082] S1031. Merge the grayscale features, texture features, and principal component analysis results according to their spatial locations to generate a fusion feature set containing multidimensional parameters;

[0083] S1032. Calculate the similarity between each element in the fused feature set and the feature patterns in the surface feature library to obtain a similarity value, and filter out matching points whose similarity value is higher than a preset threshold.

[0084] S1033. Combine the three-dimensional spatial coordinates of all the matching points to generate a discrete point set, and reconstruct a three-dimensional point cloud covering the mortar layer and the concrete layer based on the discrete point set.

[0085] In the above steps, the fusion feature set refers to a multidimensional data set formed by merging grayscale features, texture features, and principal component analysis results at the same spatial location; the surface feature library is a pre-stored reference database of standard surface features of mortar and concrete layers, such as the brightness distribution of defect-free areas and texture patterns; similarity calculation refers to the quantification of the degree of matching between each element in the fusion feature set and the feature patterns in the surface feature library using mathematical methods; matching point refers to a spatial location point where the similarity calculation result is higher than a preset threshold, indicating that the feature at that location is consistent with the standard feature; the 3D point cloud refers to a discrete point set composed of the three-dimensional coordinates of all matching points, which generates a three-dimensional surface model covering the mortar and concrete layers through a spatial reconstruction algorithm.

[0086] In this embodiment, firstly, in step S1031, the grayscale feature matrix of each pixel location, such as the single-channel brightness value 120, the texture feature matrix, such as the directional gradient value 35, the roughness value 10, and the principal component analysis result, such as the dimensionality-reduced 3D vector [0.8, -0.2, 1.1], are aligned according to spatial coordinates and merged into a fusion feature set containing multi-dimensional parameters. Before merging, the numerical ranges of the three types of features need to be standardized: grayscale features are divided by 255 to scale to [0,1], texture features are divided by the maximum gradient value 100 to normalize, and the principal component results are kept in their original values. Finally, a 6-dimensional feature vector, such as [0.47, 0.35, 0.1, 0.8, -0.2, 1.1], is generated. For example, the original grayscale value of a pixel on the surface of a concrete pipe is 120, which is 120 / 255≈0.47 after standardization; the texture roughness is normalized to 10 / 100=0.1; the principal component vector is directly retained and merged to form a fused feature vector.

[0087] Secondly, step S1032 calculates the similarity between each element in the fused feature set and the standard features in the surface feature library. The cosine similarity formula is used to quantify the degree of matching. Where A is the fused feature vector, such as B is a standard feature vector in the surface feature library, such as [ The specific calculation process is as follows: First, the dot product is... ; Modulus Similarity If the preset threshold is 0.85, then a point with a similarity of 0.92 meets the requirement and is selected as a matching point. After filtering, a list of matching point coordinates is generated, for example, pixel positions. The corresponding similarity is 0.9267.

[0088] Finally, step S1033 converts the two-dimensional pixel coordinates of the matching points into three-dimensional spatial coordinates. Based on the camera imaging model, the three-dimensional coordinates are calculated through back projection:

[0089]

[0090] in For the camera intrinsic parameter matrix, For pixel coordinates, This is the depth value at that point (calculated from the results of principal component analysis). For example, the camera focal length. Main point Then the intrinsic parameter matrix for:

[0091]

[0092] Calculation process: Assuming depth Pixel (156, 278) backprojection; normalized coordinates: ; Three-dimensional coordinates: After calculating all matching points, a discrete point set is formed, which is then fitted to a continuous surface using the Poisson reconstruction algorithm. For example, the matching point set at the edge of a hole in a concrete layer is fitted to the surface to generate a 3D point cloud model of the concave region.

[0093] In practical applications, at a concrete pipe pile quality inspection center, technicians perform surface feature fusion and 3D reconstruction on type B concrete pipe piles. First, the grayscale features extracted in step S102, such as a brightness value of 85 at a certain detection point, and texture features such as the average gradient amplitude of 13.6, are aligned with the principal component analysis results, such as the first principal component value of -35, and merged into a fusion feature set. This feature set contains multi-dimensional parameters for each detection point; for example, the fusion feature vector at coordinates (128, 79) is [85, 13.6, -35]. Then, the fusion feature set is matched with a pre-established surface feature library: the Euclidean distance between the feature patterns of typical detached areas in the feature library, such as the preset detached area feature vector [90, 15, -30], and the fusion feature is calculated. Taking point (128, 79) as an example, the similarity is calculated as follows: Matching points with a similarity below a threshold (e.g., a threshold of 8.0) are retained. The three-dimensional coordinates of all matching points are combined, and the two-dimensional coordinates are superimposed on the pile radius data to convert it into a three-dimensional structure, generating a discrete point set. For example, a matching point set on the surface of a pile contains 12,000 points, and its three-dimensional coordinates are calculated using the cylindrical projection formula: The radius r = 0.5 meters, the height h is scaled by 0.1 millimeters per pixel, and the angle θ is mapped from the pixel's horizontal coordinate, such as 2π radians corresponding to 2000 pixel columns. Based on a discrete point set, the Delaunay triangulation algorithm is used to reconstruct the 3D point cloud covering the mortar and concrete layers. For example, 18,000 triangular patches are constructed from 12,000 points. Points with gradient values ​​> 10 in the mortar area and points with gradient values ​​≤ 10 in the concrete area are distinguished by gradient differences, forming a complete surface topology. Finally, the point cloud is smoothed using Laplacian smoothing to eliminate local distortions caused by noise, achieving 3D visualization reconstruction of mortar spalling and concrete cracks.

[0094] In the overall scheme of step S103 above, by merging grayscale features, texture features, and principal component analysis results according to spatial location, a multi-dimensional fusion feature set is generated, which integrates spectral reflectance intensity, local texture direction changes, and core abnormal components. This significantly enhances the characterization ability of defects such as mortar layer peeling and concrete cracks. Subsequently, each element in the fusion feature set is matched with a pre-built surface feature library through similarity calculation. The feature weights are adaptively optimized using a channel attention mechanism to select matching points with similarity higher than a preset threshold, effectively suppressing noise interference and improving registration accuracy. Finally, a discrete point set is generated by combining the three-dimensional spatial coordinates of all matching points. The multi-view spatial coordinates are aligned and redundancy is eliminated by combining an elastic registration algorithm, and a high-precision three-dimensional point cloud model covering the mortar layer and concrete layer is reconstructed. This completely restores the geometric topology of the surface morphology and provides a three-dimensional data foundation for the quantitative analysis of structural damage.

[0095] S104. The three-dimensional point cloud is converted into a two-dimensional unfolded diagram by cylindrical projection. Based on the shape anomaly features in the surface feature library, the abnormal regions in the two-dimensional unfolded diagram are segmented to obtain mortar anomaly regions and concrete anomaly regions.

[0096] Optionally, step S104 may specifically include the following steps:

[0097] S1041. Extract the three-dimensional coordinate points from the three-dimensional point cloud, calculate the arc angle and axial height value of each three-dimensional coordinate point in the circumferential direction of the pipe, map the arc angle and axial height value to two-dimensional plane coordinate points, and connect the two-dimensional plane coordinate points to generate a continuously distributed two-dimensional unfolded diagram.

[0098] S1042. Read the contour coordinate data of the shape anomaly features from the surface feature library, and calculate the weighted sum of the distances between each pixel and the contour coordinate data on the two-dimensional unfolded diagram;

[0099] S1043. When the value of the distance-weighted sum is less than a preset similarity threshold, it is marked as an anomaly point, and adjacent anomaly points are connected to form a closed discontinuous region.

[0100] S1044. Obtain the boundary line coordinates between the mortar layer and the concrete layer in the two-dimensional unfolded diagram, calculate the difference between the center coordinates of each discontinuous region and the boundary line coordinates, and classify the discontinuous region into the mortar abnormal region when the difference is greater than zero, and classify the discontinuous region into the concrete abnormal region when the difference is less than zero.

[0101] In the above steps, cylindrical projection refers to a method of mapping each point in the 3D point cloud to two-dimensional plane coordinates by calculating its radian angle and axial height around the pipe axis, thus forming a unfolded image of the pipe surface. The two-dimensional unfolded image is a planar image generated by cylindrical projection, with the horizontal axis representing the circumferential unfolded length of the pipe and the vertical axis representing the axial height, realizing the two-dimensional visualization of the 3D curved surface. Shape anomaly features refer to the defect contour data pre-stored in the surface feature library, such as standard geometric patterns of honeycomb and cracks. Distance weighted sum is an indicator that quantifies the spatial similarity between each pixel on the two-dimensional unfolded image and the shape anomaly feature contour. The distances from each pixel to the contour points are calculated and weighted to obtain the similarity threshold, which is a preset judgment boundary value. When the weighted distance sum is lower than this value, it is marked as an anomaly. The discontinuous region refers to the closed defect region formed by connecting adjacent anomaly points, reflecting the interruption of the continuity of the local surface structure. The boundary line coordinates are the junction positions of the mortar layer and the concrete layer in the two-dimensional unfolded diagram, which are extracted by point cloud material features (such as gray-scale gradient abrupt changes). The mortar anomaly area and the concrete anomaly area refer to the defect areas where the center of the discontinuous region is located above or below the boundary line, respectively, corresponding to defects such as pitting and peeling of the mortar layer or pores and honeycombing in the concrete layer.

[0102] In this embodiment of the application, firstly, the three-dimensional coordinates of each point in the stereo point cloud are extracted through step S1041. A cylindrical coordinate system is established with the pipe axis as the reference. First, the radian angle of each point in the circumferential direction is calculated. The formula is Convert Cartesian coordinates to polar coordinates (unit: radians), and directly take... Axial value as axial height Subsequently, Mapped to two-dimensional plane coordinates Horizontal coordinate Through formula calculate( (For pipe radius), realizing the conversion from arc to circumferential straight-line distance; longitudinal coordinate Take directly The height remains constant. Finally, connecting all 2D points generates a continuous unfolded diagram, such as a concrete pipe with a diameter of 1 meter. radians are mapped to rice, Meter mapping Meters, forming a non-overlapping two-dimensional surface unfolded image.

[0103] Secondly, in step S1042, predefined shape anomaly contour coordinate data, such as the set of concave and convex boundary points of honeycomb defects, are read from the surface feature library. For each pixel in the unfolded image Calculate the weighted sum of its distances to all points in the contour point set. The formula is: ,in, For the set of concave and convex boundary points of the honeycomb defect, The weights are the weighted sum of the distances between each pixel in the unfolded image and all points in the contour set. The importance of contour points is assigned, with edge points having a weight of 0.7 and interior points having a weight of 0.3. This step involves quantization. Geometric similarity between a point and a typical defect, such as a honeycomb outline containing 20 boundary points, a certain pixel. The Euclidean distance to each contour point is After weighting This indicates a high degree of match with cellular features.

[0104] Next, a similarity threshold is set in step S1043. For example, 0.1 pixels, when the distance weighted sum of pixels... When an outlier occurs, it is marked as an anomaly. Then, adjacent outliers are merged based on spatial continuity: first, using... The morphological closing operation on pixels involves first dilation followed by erosion, connecting discrete points with a spacing of less than 2 pixels. Then, a boundary tracing algorithm, such as the Moore-Neighbor algorithm, is used to extract the closed contour, forming a discontinuous region. For example, after the closing operation on the honeycomb area of ​​a concrete pipe surface, the 50 scattered outliers are merged into a closed region with an area of ​​120 square millimeters, and the boundary is composed of 12 consecutive pixels.

[0105] Finally, the boundary function between the mortar layer and the concrete layer in the two-dimensional unfolded diagram is obtained through step S1044. This function generates values ​​by analyzing abrupt gradient changes in point cloud depth data, such as locations where depth values ​​jump by more than 5 millimeters. It then calculates the center coordinates of each discontinuous region. Take the average x-coordinate of the region boundary. Take the mean of the ordinate. The classification criterion is: if... This indicates that the center is located above the dividing line, falling within the mortar abnormality zone, such as pitting defects. rice, (meters, difference +0.3 meters); if If it is a concrete defect, it is classified as an abnormal area, such as a cavity or defect. rice, Meters, difference -0.3 meters. Finally, output the defect type and location, for example, the area of ​​the pitted area in the mortar layer is statistically calculated as 0.15 square meters, and the depth of the honeycomb area in the concrete layer is greater than 5 millimeters.

[0106] In practical applications, at a concrete pipe pile quality inspection center, technicians perform two-dimensional unfolding and defect segmentation on the reconstructed 3D point cloud. First, the 3D point cloud, containing 12,000 three-dimensional points including mortar and concrete layers, is converted into a two-dimensional unfolded diagram via cylindrical projection. Using the pipe pile axis as a reference, a projection radius r = 0.5 meters is set. For each 3D point Q(x, y, y), its radian angle θ = arctan(y / x) is calculated. For example, for a point with coordinates (0.35, 0.20, 1.80), θ = arctan(0.20 / 0.35) ≈ 0.52 radians. The axial height h is directly taken as the z-coordinate of 1.80 meters. The radian angle θ is proportionally mapped to the abscissa u = θ·r = 0.52 × 500 ≈ 260 mm, and the height h is mapped to the ordinate v = z = 1800 mm, generating a continuous two-dimensional unfolded diagram. Subsequently, segmentation is performed based on shape anomaly features in the surface feature library: Pre-stored detachment feature contour data is read, such as a 10×10 pixel jagged polygon with vertex coordinates [(50,30),(55,35),…]. For each pixel P in the unfolded image, the weighted sum of its distances to the contour vertices is calculated. For example, the distance d1 from pixel P(48,28) to the first vertex (50,30) is ≈2.83. The weighting coefficient is taken as the contour point importance weight w1=0.3, so the partial weighted sum S1=2.83×0.3≈0.85. All contour points are traversed and accumulated to obtain a total weighted sum ≈7.2. If the preset similarity threshold is 15.0, since 7.2<15.0, point P is marked as an anomaly. Adjacent anomalies are connected using a region growing algorithm to form closed discontinuous regions, such as an elliptical region containing 120 pixels. Finally, based on the boundary line between the mortar layer and the concrete layer, such as the y=60 pixel height line in the 2D diagram, the anomaly types are classified: the center coordinates of each discontinuous area are calculated. If the center of a certain area is (55, 65), and the difference between it and the boundary line is Δy=65-60=5>0, then it is classified as a mortar anomaly area; the center of another area is (55, 55), and Δy=55-60=-5<0, then it is classified as a concrete anomaly area. Through this process, the precise location of surface defects in both types of materials is ultimately achieved.

[0107] In the overall scheme of step S104 above, based on the cylindrical projection and abnormal region segmentation technology of the pipe's three-dimensional point cloud, the three-dimensional coordinates of the point cloud are mapped into a two-dimensional unfolded image through radian angle and axial height to eliminate surface distortion. At the same time, combined with the predefined shape abnormal feature contour data of the surface feature library, the distance weighted sum is calculated for each pixel of the unfolded image, and the Gaussian kernel function is used to enhance the sensitivity to edge burrs and local depressions. When the weighted sum value is lower than the preset threshold, it is marked as an abnormal point. Then, the adjacent abnormal points are connected by the region growing algorithm to form a closed discontinuous region. Furthermore, according to the boundary line coordinates between the mortar layer and the concrete layer, the vertical distance difference between the center of each abnormal region and the boundary line is calculated. When the difference is greater than zero, it is classified as a mortar abnormal region, and when the difference is less than zero, it is classified as a concrete abnormal region. This realizes the accurate identification and spatial positioning of composite material delamination defects, and provides a zoning basis for quantitative evaluation.

[0108] S105. Calculate the cavity volume based on the residual of the three-dimensional point cloud in the mortar anomaly zone, and detect internal voids by combining acoustic vibration excitation with thermal response to generate mortar defect sites including cavity volume. Obtain concrete defect sites in the concrete anomaly zone based on feature analysis of the three-dimensional point cloud.

[0109] Optionally, step S105 may specifically include the following steps:

[0110] S1051. In the mortar abnormality zone, calculate the positional deviation between the actual surface points of the three-dimensional point cloud and the theoretical model points in the surface feature library, summarize all the positional deviations to form a total depth deviation, and derive the hole volume based on the total depth deviation.

[0111] S1052. Apply mechanical vibration waves to the mortar abnormal area and simultaneously record the temperature change curve of the infrared thermal imager. When the peak rate of the temperature change curve exceeds the set rate threshold, confirm the existence of voids and output the mortar defect sites containing the volume of the voids.

[0112] S1053. Detect the normal vector angle difference between adjacent points in the three-dimensional point cloud in the abnormal concrete area. When the value of the normal vector angle difference exceeds a preset angle threshold, mark the spatial position corresponding to the normal vector angle difference as the concrete defect site.

[0113] In the above steps, the positional deviation refers to the difference in three-dimensional spatial distance between each actual surface point in the 3D point cloud and the corresponding theoretical model point in the surface feature library, used to quantify the depression depth of the cavity area; the sum of depth deviations is the cumulative value of all positional deviations, serving as the basis for calculating the cavity volume; the mechanical vibration wave is a low-frequency periodic mechanical vibration signal applied to the mortar surface through an excitation device such as an electromagnetic vibrator, used to excite the frictional heating effect in the hollow area; the temperature change curve is a function of the surface temperature of the abnormal area recorded by the infrared thermal imager over time, and its peak rate of rise reflects the thermal response intensity of the hollow defect; the normal vector angle difference is the angle between the normal directions of two adjacent points in the 3D point cloud, used to quantify the local curvature abrupt changes of the concrete surface, such as geometric discontinuities caused by cracks or spalling.

[0114] In this embodiment of the application, firstly, the coordinates of each surface point of the three-dimensional point cloud are extracted in the mortar anomaly area through step S1051. The coordinates of the theoretical model points at the corresponding locations in the surface feature library. Perform spatial matching. Calculate the positional deviation of each point. ,in, This represents the depth of the actual point along the Z-axis. This represents the depth value of the theoretical point along the Z-axis. That is, the depth difference between the actual point and the theoretical point along the Z-axis, summed up from all points. The total depth deviation is obtained. The volume of the cavity is derived based on this sum. The formula is: ,in, This refers to the unit area of ​​a single point cloud projection region. For example, a mortar-textured area has 1000 points, and the total depth deviation... millimeters, unit area Square millimeters, then the volume of the hole cubic millimeters. This volumetric value reflects the amount of material missing from the mortar layer due to voids.

[0115] Secondly, in step S1052, a mechanical vibration wave with a frequency of 20 kHz is applied to the mortar abnormal area. For example, an electromagnetic vibrator generates a simple harmonic vibration with an amplitude of 0.1 mm. Simultaneously, an infrared thermal imager records surface temperature changes at a rate of 100 frames per second. When the vibration wave reaches the hollow area, the air gap generates heat due to friction, causing a local temperature increase. The instantaneous rate of rise of the temperature change curve is calculated; if its peak value exceeds a set threshold, a hollow area is determined to exist. For example, the temperature of a hollow area in a certain tile increases from [value missing] within 3 seconds after the vibration begins. Rise to The rate of increase is After a certain number of seconds, an alarm is triggered when the threshold is reached. Finally, the output shows the mortar defect locations, including the volume of the pores and the coordinates of the voids.

[0116] Finally, in step S1053, adjacent point pairs in the 3D point cloud within the concrete anomaly zone are traversed, such as points within a 1 mm interval, and the normal vector of each point is calculated. and It is generated by fitting a local surface of the point cloud. If the angle difference between the normal vectors of two points... If the angle exceeds a preset threshold, the location is marked as a concrete defect site. For example, a point adjacent to the edge of a concrete crack. and dot product , angle difference The degree was far beyond the 30-degree threshold and was judged to be a crack defect.

[0117] In practical applications, at a precast concrete pipe pile quality inspection site, technicians performed void volume calculations and hollow detection on identified mortar anomaly areas. First, based on the reconstructed 3D point cloud containing 12,000 three-dimensional points, the mortar anomaly area was divided into 10cm × 10cm grid units. The deviation between the actual surface of the point cloud and the theoretical model was calculated for each unit. For example, the depth deviation Δd of the detection point P (x=1.2m, y=0.8m, z=0.05m) within a certain unit compared to the theoretical model point P' (x=1.2m, y=0.8m, z=0.065m) is 0.015m, which is |0.05-0.065|. The total depth deviation of the 150 grid units within this anomaly area, ΣΔd, is 2.25m. Using calculus, the void volume V = 2.25m × 0.01m² = 0.0225m³ is derived. Subsequently, a hollow area verification was performed: a mechanical vibration wave at a frequency of 200Hz was applied to the area using an acoustic vibration excitation device, while an infrared thermal imager with a sampling rate of 100Hz recorded the surface temperature change. When the vibration lasted for 3 seconds, the temperature change rate at a certain coordinate point (1.5m, 1.0m) suddenly increased from 0.3℃ / s to 1.2℃ / s, exceeding the set threshold of 0.8℃ / s, indicating that a local hot spot was generated due to vibration friction, confirming the existence of a hollow defect with a diameter of 15cm. Finally, mortar defect site information was generated, including the cavity volume of 0.0225m³ and the coordinates of the hollow area. For the concrete anomaly zone, defects were located by analyzing the difference in the normal vector directions of adjacent points in the 3D point cloud: Point Q (x=2.1m, y=1.3m) and its adjacent points Q1 and Q2 were selected. The angle θ ≈ 51.2° between the normal vector n=(0.34, -0.12, 0.93) of point Q and the normal vector n1=(0.89, 0.01, 0.45) of Q1 was calculated. When the angle difference exceeded a preset threshold of 15°, the location was marked as a concrete crack defect site. By traversing 12,000 points, a total of 37 crack defects were located.

[0118] In the overall scheme of step S105 above, the three-dimensional volume parameters of the cavity are directly derived by summarizing the depth position deviation between the actual surface points and the theoretical model points. Simultaneously, combining acoustic vibration excitation and infrared thermal response joint detection mechanisms, the interface frictional heating effect induced by cavity resonance is captured after applying mechanical vibration waves. Based on the characteristic that the peak value of the infrared thermal imaging temperature change rate exceeds a set threshold, the non-destructive location and verification of hidden hollow defects are achieved. In the concrete anomaly zone, by analyzing the abrupt change characteristics of the normal vector angle between adjacent points in the three-dimensional point cloud, surface discontinuities such as crack edges and spalling pits are accurately identified, completing the spatial marking and classification of concrete defect sites. This technology significantly improves the detection rate and quantification accuracy of hidden defects in composite structures, providing multi-dimensional defect parameter support for the formulation of repair schemes.

[0119] This application provides a schematic diagram illustrating a specific implementation of a machine vision-based intelligent inspection method for pipe surface quality. Figure 2 As shown, it includes the following:

[0120] S106. Based on the process sequence of concrete being applied before mortar, construct an association matching model between the mortar defect sites and the concrete defect sites, and use the association matching model to determine whether the defect sites overlap. If they overlap, it is determined to be due to the influence of the preceding process, and a detection result is generated based on the judgment result.

[0121] Optionally, step S106 may specifically include the following steps:

[0122] S1061. According to the process order of concrete layer covering mortar layer, establish the vertical projection relationship of the spatial coordinates of the concrete defect site to the mortar defect site, and form an association matching model based on the vertical projection relationship.

[0123] S1062. Using the correlation matching model, compare the coordinate intersection range of the mortar defect site and the concrete defect site. When the area of ​​the coordinate intersection range exceeds a preset threshold, it is marked as a defect affecting the previous process.

[0124] Step S1062 may specifically include the following process: mapping the spatial coordinates of the concrete defect site to the mortar layer coordinate system plane through coordinate transformation rules to form a set of projected coordinate points; constructing a circular area range based on each projected point in the set of projected coordinate points; comparing the circular area range with the actual coordinate positions of the mortar defect sites, calculating the intersection area value of the circular area range and the actual mortar defect sites, accumulating the intersection area values ​​corresponding to all projected coordinate points to form a total area value; when the total area value exceeds a preset area threshold, marking the position corresponding to the projected coordinate point as a defect affected by the previous process.

[0125] S1063. Summarize the spatial distribution information of all defects affected by the previous process and generate a test result that includes the attribution of process responsibility.

[0126] In the above steps, the association matching model refers to the vertical projection relationship model established based on the process coverage order of the concrete layer before the mortar layer. It maps the three-dimensional coordinates of the concrete defect site to the mortar layer plane coordinate system, forming a projection point set to analyze the spatial correlation between the two types of defects. The vertical projection relationship is a mathematical mapping method that maps the concrete defect site to the mortar layer plane through coordinate transformation rules, such as ignoring the height axis and retaining the horizontal coordinate. The coordinate intersection range refers to the area where the projection point set overlaps with the mortar defect site. The degree of correlation is quantified by calculating the overlapping area within the preset range of the projection points. The projection coordinate point set is the two-dimensional coordinate set generated on the mortar layer plane after the concrete defect site is vertically projected. The circular area range is a circular analysis area constructed with the projection point as the center and a preset radius, used to quantify the local overlapping area. The defects affected by the previous process refer to the responsibility attribution results for defects in the corresponding position of the mortar layer caused by defects in the concrete base surface. The spatial distribution information summarizes the statistical results of the location coordinates, area, and responsibility attribution of all defects affected by the previous process.

[0127] In this embodiment of the application, firstly, in step S1061, according to the construction sequence of the concrete layer first and the mortar layer later, the concrete defect sites, such as coordinate points, are established. The vertical projection relationship to the mortar layer plane. The specific mapping rule is: ignore the vertical height coordinates of concrete defect points. , its horizontal coordinate ( The projection points directly mapped to the mortar layer plane This forms a set of projected coordinate points. For example, crack locations in concrete layers. Millimeters, the projected plane coordinates of the mortar layer are Millimeters, this mapping relationship constitutes the core logic of the association matching model.

[0128] Secondly, each projection point is processed through step S1062. Construction radius is The circular region is defined, and its overlapping area with the actual coordinates of the mortar defect site is calculated. The specific process is as follows: First, the overlapping area is calculated for each projection point. Detect points within the circular area of ​​the mortar defect site and count the number of mortar defect points in that area. Then the local overlapping area ,in, This represents the unit area of ​​a single point, such as 1 square millimeter. Then, the overlapping areas of all projected points are summed to obtain the total area value. .like If a preset area threshold, such as 1000 square millimeters, is set, then the area corresponding to the projection point is marked as a defect affected by the previous process. For example, the projection point of a concrete hole. The circular area contains 15 mortar defect points, each with a unit area of ​​1 square millimeter and a local overlap area of ​​15 square millimeters. If the total area of ​​all projected points reaches 1200 square millimeters, exceeding the threshold of 1000 square millimeters, then the area is determined to be a defect caused by uneven concrete substrate in the previous process.

[0129] Finally, in step S1063, all spatial coordinates marked as defects affecting the previous process are summarized as shown in the list. The system uses the overlapping area data to generate a test result report that includes attribution of responsibility. The report logic is as follows: overlapping defects are attributed to the concrete process, defects in the base surface cause defects after the mortar layer is applied, and non-overlapping defects are attributed to the mortar construction. For example, in the inspection of a bridge pier, 12 overlapping defects with a total area of ​​3500 square millimeters were summarized. The report clearly marked "honeycombing of the concrete base surface caused mortar hollowing," providing a basis for the subsequent division of responsibility for rework.

[0130] In practical applications, at the site inspection of precast concrete pipe piles, technicians from Quality Inspection Center A conducted a full-process quality inspection of Type C pipe piles produced by Factory B. First, they verified the consistency between the pipe pile specifications and the design drawings, and checked the product certificate, end plate thickness, and pile body markings item by item. For example, when sampling 3 pile sections (2% of the total batch), they found that the end plate thickness of one section was 14.8mm, lower than the design requirement of 16mm. They then doubled the sampling to 6 sections, and the re-inspection still showed 2 sections failing, leading to the conclusion that this batch of pipe piles was prohibited from use. Next, they conducted a pile body integrity test: using the low-strain reflected wave method to scan the pile body, they analyzed the stress wave propagation time. If the time difference between the incident and reflected waves was Δt = 1.2ms, they determined that there were no internal cracks. Simultaneously, they tapped the pile body; if a hollow sound was emitted, with a frequency deviation ≥200Hz, they initially determined that segregation existed. They then retested using the ultrasonic transmission method, recording a sound velocity of 3800m / s, lower than the healthy pile benchmark value of 4200m / s, confirming the location of the defect. In pile position accuracy testing, a total station is used to measure the verticality of the pile body. The vertical deviation is calculated as offset / pile length × 100%. For example, the verticality of a 6m pile with an offset of 30mm is 0.5%, and the control error is ≤0.5%. The horizontal displacement deviation is determined by coordinate comparison. The design coordinates [102.35, 58.17] and the actual coordinates [102.37, 58.19] yield Δx=0.02m and Δy=0.02m, which meets the ≤0.5% limit. In the bearing capacity verification stage, three engineering piles are selected for static load tests: when the load is applied in stages to the design value of 800kN, the settlement is 4.8mm. After further applying pressure to 1200kN, the settlement suddenly increases to 15mm, exceeding the allowable value of 12mm. The ultimate bearing capacity of a single pile is determined to be 1100kN, which is lower than the design requirement of 1200kN, requiring additional piles. Finally, the material report was compiled, showing that the tensile strength of the steel bars was ≥1420MPa, the welding record showed that the cooling time of the pile splice was ≥5 minutes, and the environmental monitoring data showed that the soil squeezing effect caused the displacement of the adjacent building to be 3mm. The test results were generated and the responsibility was marked. The insufficient bearing capacity was attributed to the defect in the concrete mix design.

[0131] In the overall scheme of step S106 above, an associated matching model is constructed by establishing the vertical projection relationship between concrete defect sites and mortar defect sites. The coordinate transformation rules are used to map the concrete defect sites to the mortar layer coordinate system plane to form a set of projected coordinate points, and a circular area is constructed with each projection point as the center. By calculating the intersection area value between the circular area and the actual mortar defect sites, the total area is accumulated and marked as a defect affecting the previous process when it exceeds a preset threshold, so as to accurately identify the chain damage caused by the concrete layer defects leading to the hollowing and peeling of the mortar layer. Finally, the spatial distribution information of all defects affecting the previous process is summarized to generate a detection result that includes the process responsibility attribution, providing a data-driven decision-making basis for construction quality traceability and process improvement.

[0132] The following is a complete embodiment for steps S101 to S106:

[0133] like Figure 3 As shown, in the quality inspection workshop of a precast concrete pipe pile manufacturing plant, technicians conducted intelligent surface quality inspection on a batch of C80 type pipe piles. Before the inspection began, the system had established a pipe surface feature library containing typical defects, including marking irregular sawtooth contours and other abnormal shape features for areas where the mortar layer had detached. In actual operation, a combination of ring light source and side parallel light source was first used to form a bright and dark composite light field. When 12 sets of LED light sources illuminated the surface of the pipe pile at different angles, a 50-megapixel industrial camera simultaneously acquired images. To eliminate the interference of metal mold reflections, a linear polarizing filter was added in front of the lens to reduce the intensity of reflected light from 180 Lux to 40 Lux, significantly improving the contrast of hole shadows and crack edges.

[0134] The processed image enters the feature analysis stage: taking a 10×10 pixel area as an example, the system calculates the brightness value of the center point to be 85, and the standard deviation of the brightness of its adjacent pixels reaches 13.6. At the same time, the texture entropy value of 1.8 is extracted through the gray-level co-occurrence matrix. After dimensionality reduction by principal component analysis, these gray-level and texture features generate a principal component value of -35 reflecting the surface roughness. Subsequently, the multidimensional features are fused according to spatial coordinates to form a feature vector [85, 13.6, -35], which is matched with the preset detached template vector [90, 15, -30] in the feature library. When the calculated Euclidean distance ≈ 7.21 is lower than the threshold of 8.0, the point is marked as a matching point. Based on the spatial coordinates of 12,000 matching points, the system performs three-dimensional reconstruction using the cylindrical projection formula: for example, pixel coordinates (260, 1800) are converted into three-dimensional points (435mm, 250mm, 1800mm), and then triangulation is performed to generate a three-dimensional point cloud covering the mortar and concrete layers. After the point cloud is unfolded into a two-dimensional image by cylindrical projection, the system segments the abnormal region according to the jagged contour template in the feature library: the weighted distance sum from a certain pixel (48,28) to the contour vertex is 7.2, which is lower than the threshold of 15.0 and is therefore marked as an abnormal point, ultimately forming an elliptical abnormal region of 120 pixels. When classifying the defect attribution by the mortar-concrete boundary line, the area with center coordinates (55,65) is determined to be a mortar abnormal region due to a height difference of +5mm, while the area with center coordinates (55,55) is classified as a concrete abnormal region due to a height difference of -5mm.

[0135] In the mortar anomaly zone, the system was divided into 10cm×10cm grids to calculate the point cloud deviation: the measured height of a certain grid point was 0.05m, which was 0.015m lower than the theoretical value of 0.065m. The cumulative deviation of 2.25m from 150 grids led to a cavity volume of 0.0225m³. Simultaneously, 200Hz mechanical vibration was applied to verify the presence of voids. When the infrared thermal imager detected a temperature change rate jump from 0.3℃ / s to 1.2℃ / s (exceeding the threshold of 0.8℃ / s), a void with a diameter of 15cm was confirmed. In the concrete anomaly zone, cracks were located through normal vector analysis: the angle between the normal vector of a certain point (0.34, -0.12, 0.93) and its neighboring point (0.89, 0.01, 0.45) reached 51.2°, far exceeding the 15° threshold. Based on the process characteristic of concrete preceding mortar coverage, the center point of the concrete crack (1.2m, 0.3m) was vertically projected onto the mortar layer, forming coordinates (1.2m, 0.3m), and a 0.1m radius influence zone was constructed centered on this coordinate. When a mortar hole center (1.19m, 0.32m) was detected only 0.022m from the projection point, the overlapping area reached 0.00196m³, and the total overlapping area of ​​0.00392m³ exceeded the threshold of 0.003m³, indicating that the hole was induced by the concrete crack. The inspection report clearly stated that "3 mortar holes originated from defects in the preceding concrete process," guiding the maintenance team to prioritize repairing the cracks before retesting, resulting in fewer new defects.

[0136] The intelligent inspection method for pipe surface quality based on machine vision provided in this application establishes a pipe surface feature library with labeled shape anomalies. It then acquires high-contrast images using a combination of bright and dark composite light field imaging and polarization filtering technology, extracts grayscale and texture features, and optimizes feature representation through principal component analysis. Furthermore, it integrates a multi-dimensional feature set and matches it with the feature library to reconstruct a three-dimensional point cloud covering the mortar and concrete layers. This cloud is then converted into a two-dimensional unfolded image via cylindrical projection to segment the mortar and concrete anomaly areas. In the mortar anomaly area, the point cloud residual is used to calculate the void volume, and hollow defects are verified based on joint detection using acoustic vibration excitation and infrared thermal response, generating mortar defect sites with quantified parameters. In the concrete anomaly area, structural defect sites are located using the point cloud normal vector mutation feature. Finally, based on the process logic of concrete preceding mortar coverage, an association matching model is constructed. Defects influenced by previous processes are identified through coordinate projection and cross-area calculation, generating inspection results that include responsibility attribution. This achieves intelligent diagnosis across the entire chain, from defect identification and quantification to process responsibility tracing.

[0137] Figure 4 This is a schematic diagram illustrating a specific embodiment of a machine vision-based intelligent inspection system for pipe surface quality provided in this application. (Refer to...) Figure 4 The system may include:

[0138] Module 41 is used to pre-establish a pipe surface feature library and mark the abnormal shape features of the detached areas in the surface feature library;

[0139] Analysis module 42 is used to acquire images of the pipe surface under a combined light field of light and dark, perform polarization filtering on the images, extract grayscale features and texture features from the processed images, perform principal component analysis, and obtain principal component analysis results.

[0140] Matching module 43 is used to generate a fusion feature set based on the grayscale features, texture features and principal component analysis results, and to perform feature matching between the fusion feature set and the surface feature library, and to reconstruct a three-dimensional point cloud covering the mortar layer and the concrete layer based on the matching results;

[0141] The segmentation module 44 is used to convert the three-dimensional point cloud into a two-dimensional unfolded diagram through cylindrical projection, and to segment the abnormal areas in the two-dimensional unfolded diagram according to the shape abnormal features in the surface feature library to obtain mortar abnormal areas and concrete abnormal areas.

[0142] The calculation module 45 is used to calculate the volume of pores in the mortar anomaly zone based on the residual of the three-dimensional point cloud, and to detect internal voids by combining acoustic vibration excitation with thermal response to generate mortar defect sites including the volume of pores, and to obtain concrete defect sites in the concrete anomaly zone based on the feature analysis of the three-dimensional point cloud.

[0143] The generation module 46 is used to construct an association matching model between the mortar defect sites and the concrete defect sites according to the process sequence of concrete being covered before mortar, and to use the association matching model to determine whether the defect sites overlap. When they overlap, it is determined that the preceding process has an impact, and the detection result is generated based on the judgment result.

[0144] The intelligent pipe surface quality detection system based on machine vision in this application is used to implement the aforementioned intelligent pipe surface quality detection method based on machine vision. Therefore, the specific implementation of the intelligent pipe surface quality detection system based on machine vision can be found in the embodiment section of the intelligent pipe surface quality detection method based on machine vision above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0145] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the machine vision-based intelligent inspection method for pipe surface quality described above.

[0146] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described intelligent inspection methods for pipe surface quality based on machine vision.

[0147] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0148] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the intelligent detection method for pipe surface quality based on machine vision.

[0149] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0150] The above provides a detailed description of the intelligent inspection method and system for pipe surface quality based on machine vision provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A machine vision-based intelligent inspection method for pipe surface quality, characterized in that, include: A pipe surface feature library is pre-established, and abnormal shape features are marked on the detached areas in the surface feature library; Images of the pipe surface under a combined light and dark field are acquired, polarization filtering is applied to the images, and principal component analysis is performed on the grayscale and texture features extracted from the processed images to obtain the principal component analysis results. A fusion feature set is generated based on the grayscale features, texture features and principal component analysis results, and the fusion feature set is matched with the surface feature library. Based on the matching results, a three-dimensional point cloud covering the mortar layer and concrete layer is reconstructed. The three-dimensional point cloud is converted into a two-dimensional unfolded image by cylindrical projection. Based on the shape anomaly features in the surface feature library, the abnormal regions in the two-dimensional unfolded image are segmented to obtain mortar anomaly regions and concrete anomaly regions. In the mortar anomaly zone, the cavity volume is calculated based on the residual of the three-dimensional point cloud, and the internal voids are detected by combining acoustic vibration excitation with thermal response to generate mortar defect sites including the cavity volume. In the concrete anomaly zone, the concrete defect sites are obtained based on the feature analysis of the three-dimensional point cloud. Based on the process sequence of concrete being applied before mortar, an association matching model is constructed between the mortar defect sites and the concrete defect sites. The association matching model is then used to determine whether the defect sites overlap. If they overlap, it is determined that the preceding process has an impact, and a test result is generated based on the determination result.

2. The method according to claim 1, characterized in that, Based on the sequence of concrete application preceding mortar application, an association matching model is constructed between the mortar defect sites and the concrete defect sites. This model is then used to determine whether the defect sites overlap. If they overlap, it is determined to be due to the influence of the preceding process. Based on this determination, detection results are generated, including: According to the construction sequence of concrete layer covering mortar layer, a vertical projection relationship of spatial coordinates from the concrete defect site to the mortar defect site is established, and an association matching model is formed based on the vertical projection relationship. The correlation matching model is used to compare the coordinate intersection range of the mortar defect site and the concrete defect site. When the area of ​​the coordinate intersection range exceeds a preset threshold, it is marked as a defect affected by the previous process. Summarize the spatial distribution information of all defects affected by the preceding processes to generate inspection results that include process responsibility attribution.

3. The method according to claim 2, characterized in that, The correlation matching model compares the coordinate intersection range of the mortar defect site and the concrete defect site. When the area of ​​the coordinate intersection range exceeds a preset threshold, it is marked as a defect affecting the previous process, including: The spatial coordinates of the concrete defect sites are mapped to the mortar layer coordinate system plane through coordinate transformation rules to form a set of projected coordinate points. A circular region is constructed based on each projected point in the set of projected coordinate points. The circular area is compared with the actual coordinates of the mortar defect sites. The intersection area value of the circular area and the actual mortar defect sites is calculated. The intersection area values ​​corresponding to all projected coordinate points are summed to form the total area value. When the total area value exceeds a preset area threshold, the location corresponding to the projected coordinate point is marked as a defect affecting the previous process.

4. The method according to claim 1, characterized in that, In the mortar anomaly zone, the void volume is calculated based on the residual of the three-dimensional point cloud, and internal voids are detected by combining acoustic vibration excitation with thermal response to generate mortar defect sites including void volumes. In the concrete anomaly zone, concrete defect sites are obtained based on feature analysis of the three-dimensional point cloud, including: In the mortar anomaly zone, the positional deviation between the actual surface points of the three-dimensional point cloud and the theoretical model points in the surface feature library is calculated, and all the positional deviations are summed to form a total depth deviation. The hole volume is then derived based on the total depth deviation. Mechanical vibration waves are applied to the abnormal mortar area, and the temperature change curve of the infrared thermal imager is recorded simultaneously. When the peak rate of the temperature change curve exceeds the set rate threshold, the existence of voids is confirmed, and the mortar defect site containing the volume of the voids is output. In the abnormal concrete area, the difference in the normal vector angle between adjacent points in the three-dimensional point cloud is detected. When the value of the difference in the normal vector angle exceeds a preset angle threshold, the spatial position corresponding to the difference in the normal vector angle is marked as the concrete defect site.

5. The method according to claim 1, characterized in that, A fused feature set is generated based on the grayscale features, texture features, and principal component analysis results. The fused feature set is then matched with the surface feature library. Based on the matching results, a three-dimensional point cloud covering the mortar and concrete layers is reconstructed, including: The grayscale features, texture features, and principal component analysis results are merged according to their spatial locations to generate a fusion feature set containing multidimensional parameters; The similarity between each element in the fused feature set and the feature patterns in the surface feature library is calculated to obtain a similarity value, and matching points with similarity values ​​higher than a preset threshold are selected. Combine the three-dimensional spatial coordinates of all the matching points to generate a discrete point set, and reconstruct a three-dimensional point cloud covering the mortar layer and the concrete layer based on the discrete point set.

6. The method according to claim 1, characterized in that, Images of the pipe surface under a combined bright and dark light field are acquired. These images are then subjected to polarization filtering. Principal component analysis (PCA) is performed on the processed images, extracting grayscale and texture features, to obtain the PCA results, including: Multiple light sources illuminate the surface of the pipe from different directions to form a composite light field of light and dark, and images under the composite light field of light and dark are collected. A polarization filter is superimposed on the image to eliminate interference components caused by specular reflection and retain the effective light signal on the pipe surface; From the image after polarization filtering, brightness values ​​are extracted pixel by pixel as grayscale features, and directional change information is statistically analyzed within the range of adjacent pixels as texture features. The grayscale features and the texture features are combined into an initial dataset, and a linear transformation is performed on the initial dataset to obtain principal component analysis results representing the main direction of variation.

7. The method according to claim 1, characterized in that, The three-dimensional point cloud is converted into a two-dimensional unfolded image via cylindrical projection. Based on the shape anomaly features in the surface feature library, the abnormal regions in the two-dimensional unfolded image are segmented to obtain mortar anomaly regions and concrete anomaly regions, including: Extract the three-dimensional coordinate points from the three-dimensional point cloud, calculate the arc angle and axial height value of each three-dimensional coordinate point in the circumferential direction of the pipe, map the arc angle and axial height value to two-dimensional plane coordinate points, and connect the two-dimensional plane coordinate points to generate a continuously distributed two-dimensional unfolded diagram; Read the contour coordinate data of the shape anomaly features from the surface feature library, and calculate the weighted sum of the distances between each pixel and the contour coordinate data on the two-dimensional unfolded map; When the value of the distance-weighted sum is less than a preset similarity threshold, it is marked as an anomaly point, and adjacent anomaly points are connected to form a closed discontinuous region; Obtain the boundary line coordinates between the mortar layer and the concrete layer in the two-dimensional unfolded diagram, calculate the difference between the center coordinates of each discontinuous region and the boundary line coordinates, and classify the discontinuous region into the mortar abnormal region when the difference is greater than zero, and classify the discontinuous region into the concrete abnormal region when the difference is less than zero.

8. A machine vision-based intelligent inspection system for pipe surface quality, characterized in that, include: A module is established to pre-build a pipe surface feature library and mark the abnormal shape features of the detached areas in the surface feature library. The analysis module is used to acquire images of the pipe surface under a combined light field of light and dark, perform polarization filtering on the images, extract grayscale features and texture features from the processed images, perform principal component analysis, and obtain principal component analysis results. The matching module is used to generate a fusion feature set based on the grayscale features, texture features and principal component analysis results, and to perform feature matching between the fusion feature set and the surface feature library, and to reconstruct the three-dimensional point cloud covering the mortar layer and the concrete layer based on the matching results. The segmentation module is used to convert the three-dimensional point cloud into a two-dimensional unfolded diagram through cylindrical projection, and to segment the abnormal areas in the two-dimensional unfolded diagram according to the shape abnormal features in the surface feature library, so as to obtain mortar abnormal areas and concrete abnormal areas. The calculation module is used to calculate the volume of pores in the mortar anomaly zone based on the residual of the three-dimensional point cloud, and to detect internal voids by combining acoustic vibration excitation with thermal response to generate mortar defect sites including the volume of pores. The module is also used to obtain concrete defect sites in the concrete anomaly zone based on the feature analysis of the three-dimensional point cloud. The generation module is used to construct an association matching model between the mortar defect sites and the concrete defect sites according to the process sequence of concrete being applied before mortar, and to use the association matching model to determine whether the defect sites overlap. If they overlap, it is determined that the preceding process has an impact, and the detection result is generated based on the judgment result.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the machine vision-based intelligent inspection method for pipe surface quality as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the intelligent inspection method for pipe surface quality based on machine vision as described in any one of claims 1 to 7.

Images