Fabricated prefabricated part damage detection method and device, terminal and medium

Abstract

The invention provides an assembly type prefabricated part damage detection method and device, a terminal and a medium, and the method comprises the steps: obtaining an original three-dimensional point cloud data set of an assembly type prefabricated part, and carrying out the preprocessing of the original three-dimensional point cloud data set, and obtaining a target three-dimensional point cloud data set; extracting geometric features of the point cloud from the target three-dimensional point cloud data set, and constructing a point cloud feature map based on the geometric features of the point cloud; converting the point cloud feature map into a CAD image based on a preset capsule network; and detecting a damaged area in the CAD image by using a preset clustering algorithm, and carrying out area calibration on the damaged area on the CAD image. According to the method, on the basis of geometric feature extraction of point cloud data, the original three-dimensional point cloud is converted into high-level CAD image information by introducing deep learning analysis and clustering analysis, so that high-precision automatic damage detection and visual visualization of the prefabricated part are realized; and higher detection efficiency and robustness can be shown in large-scale point cloud data processing and complex scenes.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, terminal and medium for detecting damage of assembled prefabricated components. Background Art

[0002] At present, prefabricated components are widely used in construction engineering, bridge structures and other fields due to their high construction efficiency, stable quality and good environmental performance. However, due to the inevitable mechanical damage or external impact during the production, transportation and installation process, prefabricated components may be damaged or defective. If these problems are not detected and repaired in time, they may lead to a decline in project quality and even structural safety hazards.

[0003] The current methods for detecting damage to prefabricated components mainly include two categories: traditional manual detection methods and automated detection methods based on image processing or three-dimensional point cloud analysis. Manual detection methods rely on experienced professionals to complete the damage determination of component surfaces through visual inspection or simple tools. Its advantage is strong adaptability, but low efficiency and difficulty in meeting the needs of large-scale projects. Automated detection methods use non-contact measurement equipment, such as laser scanners, depth cameras, etc., to collect geometric information on the surface of components, identify cracks or defects through image processing algorithms, or use three-dimensional point clouds to analyze abnormalities in surface geometric features. However, image processing algorithms are usually limited by lighting conditions and background interference, making it difficult to ensure the robustness of detection, while three-dimensional point cloud analysis methods have problems such as insufficient feature extraction and insufficient accuracy in abnormality recognition.

[0004] Although the existing automated detection methods based on 3D point cloud analysis have achieved certain results, they still have the following major defects: First, the point cloud data volume is large and noisy, and traditional noise reduction and feature extraction algorithms are prone to loss of geometric details when processing high-precision data; second, most existing methods remain at the stage of simple extraction and comparison of geometric features, lacking the ability to conduct in-depth analysis of multi-scale and multi-feature fusion, and it is difficult to accurately determine complex damage patterns; third, the detection results are mostly based on data point markings, lacking intuitive and operational visual calibration methods, and it is difficult to meet the needs of rapid positioning and subsequent repair in actual engineering.

[0005] Therefore, the prior art has defects and needs to be improved and developed. Summary of the invention

[0006] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a method, device, terminal and medium for detecting damage of prefabricated components are provided, aiming to solve the problems of insufficient geometric feature extraction, insufficient complex damage pattern recognition ability and unintuitive visualization of detection results in the existing methods for detecting damage of prefabricated components.

[0007] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0008] A method for detecting damage of prefabricated components, wherein the method comprises:

[0009] Acquire an original three-dimensional point cloud data set of the prefabricated component, and preprocess the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set;

[0010] Extracting geometric features of a point cloud from the target three-dimensional point cloud data set, and constructing a point cloud feature map based on the geometric features of the point cloud;

[0011] Converting the point cloud feature map into a corresponding CAD image based on a preset capsule network;

[0012] A preset clustering algorithm is used to detect damaged areas in the CAD image, and the damaged areas are calibrated on the CAD image.

[0013] In one implementation, the preprocessing of the original three-dimensional point cloud dataset to obtain a target three-dimensional point cloud dataset includes:

[0014] Using a bilateral filter to process each point cloud in the original three-dimensional point cloud data set to obtain a processed three-dimensional point cloud data set;

[0015] The processed 3D point cloud dataset is downsampled using voxel gridding to obtain a uniformly distributed target 3D point cloud dataset.

[0016] In one implementation, extracting geometric features of a point cloud from the target three-dimensional point cloud data set includes:

[0017] Obtaining neighborhood information of each point cloud in the target three-dimensional point cloud data set, and extracting corresponding curvature features and normal vectors using the neighborhood information of the point cloud;

[0018] Wherein, constructing a point cloud feature map based on the geometric features of the point cloud includes:

[0019] The curvature feature and the normal vector are combined into a feature vector, and a corresponding point cloud feature map is constructed based on the feature vector.

[0020] In one implementation, converting the point cloud feature map into a corresponding CAD image based on a preset capsule network includes:

[0021] Projecting the point cloud feature map onto multiple two-dimensional planes, and calculating the geometric feature value of each projection point using the feature value calculation function of the point;

[0022] The geometric feature values ​​are fused using a multi-level capsule fusion network to generate a corresponding CAD image.

[0023] In one implementation, the detecting the damaged area in the CAD image by using a preset clustering algorithm includes:

[0024] The damaged area in the CAD image is detected using the DBSCAN clustering algorithm.

[0025] In one implementation, the detecting the damaged area in the CAD image using the DBSCAN clustering algorithm includes:

[0026] Determine a feature point set of the CAD image, and calculate the number of neighborhood points of each feature point in the feature point set;

[0027] Determine whether the number of neighborhood points of each feature point is less than a preset neighborhood point threshold;

[0028] If the number of neighborhood points of the feature point is less than the preset neighborhood point threshold, the feature point is determined to be an abnormal point, and a corresponding abnormal point set is obtained;

[0029] Determine the number of outliers in the outlier point set and the total number of point cloud points in the target three-dimensional point cloud data set;

[0030] The degree of damage is calculated based on the number of abnormal points and the total number of points in the point cloud, and the corresponding damaged area is determined according to the degree of damage.

[0031] In one implementation, the performing area calibration on the damaged area on the CAD image includes:

[0032] The damaged area is calibrated on the CAD image using a color coding calibration method.

[0033] The present invention also discloses a prefabricated component damage detection device, wherein the device comprises:

[0034] A point cloud data acquisition module is used to acquire the original three-dimensional point cloud data set of the prefabricated component;

[0035] A point cloud data preprocessing module, used for preprocessing the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set;

[0036] A feature extraction module, used to extract geometric features of point clouds from the target three-dimensional point cloud data set;

[0037] A feature map construction module, used to construct a point cloud feature map based on the geometric features of the point cloud;

[0038] A CAD image generation module, used for converting the point cloud feature map into a corresponding CAD image based on a preset capsule network;

[0039] A damaged area detection module, used to detect damaged areas in the CAD image using a preset clustering algorithm;

[0040] The damaged area calibration module is used to calibrate the damaged area on the CAD image.

[0041] The present invention also discloses a terminal, which includes: a memory, a processor, and an assembled prefabricated component damage detection program stored in the memory and executable on the processor, wherein the assembled prefabricated component damage detection program implements the steps of the assembled prefabricated component damage detection method as described above when executed by the processor.

[0042] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program can be executed to implement the steps of the method for detecting damage of assembled prefabricated components as described above.

[0043] The present invention provides a method, device, terminal and medium for detecting damage to prefabricated components. The method for detecting damage to prefabricated components comprises: obtaining an original three-dimensional point cloud data set of prefabricated components, preprocessing the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set; extracting geometric features of the point cloud from the target three-dimensional point cloud data set, and constructing a point cloud feature map based on the geometric features of the point cloud; converting the point cloud feature map into a corresponding CAD image based on a preset capsule network; detecting a damaged area in the CAD image using a preset clustering algorithm, and calibrating the damaged area on the CAD image. It can be seen that the present invention preprocesses the point cloud data, and then extracts features from the preprocessed point cloud data to construct a feature map, and uses a preset capsule network to perform deep learning analysis on the feature map to convert the feature map into a two-dimensional CAD image, thereby realizing deep encoding and fusion of features, and then detecting the damaged area in the CAD image through a preset clustering algorithm, and visually calibrating the damaged area on the CAD image, thereby improving the accuracy and efficiency of damage detection, and enhancing the intuitiveness and engineering practicality of the detection results, thereby efficiently meeting the damage detection needs of prefabricated components in complex scenarios. That is, the technical solution of the present application is based on the geometric feature extraction of point cloud data, and through deep learning analysis and clustering analysis, the original three-dimensional point cloud is converted into high-level CAD image information, thereby realizing high-precision automated damage detection of prefabricated components, and can show higher detection efficiency and robustness under large-scale point cloud data processing and complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of a preferred embodiment of the method for detecting damage of assembled prefabricated components in the present invention;

[0045] Figure 2 It is a flow chart of a specific method for detecting damage of assembled prefabricated components disclosed in the present invention;

[0046] Figure 3 It is a logic schematic diagram of the method for detecting damage of assembled prefabricated components in the present invention;

[0047] Figure 4 It is a functional principle block diagram of a preferred embodiment of the assembled prefabricated component damage detection device of the present invention;

[0048] Figure 5 It is a functional principle block diagram of a preferred embodiment of the terminal in the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] See also Figure 1 , Figure 1 Flow chart of the method for detecting damage of prefabricated components in the present invention. Figure 1 As shown, the method for detecting damage of prefabricated components according to the embodiment of the present invention includes:

[0051] Step S11, obtaining an original three-dimensional point cloud dataset of the prefabricated component, and preprocessing the original three-dimensional point cloud dataset to obtain a target three-dimensional point cloud dataset.

[0052] In this embodiment, point cloud data is first obtained, that is, the original three-dimensional point cloud data set of the prefabricated component is obtained, that is, the original three-dimensional point cloud data set obtained by discretizing the surface of the prefabricated component is obtained, wherein the point cloud data set is a set composed of a large number of three-dimensional space points, so the original three-dimensional point cloud data set can be expressed as a space coordinate point set, that is: P = {p i |p i ∈R 3 ,i=1,2K,N};

[0053] Among them, the i-th point cloud p i The spatial coordinates of i ,y i ,z i ).

[0054] It should be pointed out that point cloud technology is an accurate means of acquiring three-dimensional spatial data, and provides an important basis for digital modeling and defect detection of prefabricated components. Therefore, the development of point cloud-based prefabricated component damage detection technology can achieve efficient and accurate damage positioning and evaluation, which has important engineering application value.

[0055] In this embodiment, after obtaining the original three-dimensional point cloud data set of the prefabricated component, the obtained original three-dimensional point cloud data set is preprocessed to obtain the target three-dimensional point cloud data set. Specifically, each point cloud in the original three-dimensional point cloud data set is processed by using a bilateral filter to obtain a processed three-dimensional point cloud data set, and then the processed three-dimensional point cloud data set is downsampled by using a voxel gridding method to obtain a uniformly distributed target three-dimensional point cloud data set. It can be understood that since the point cloud usually contains a large amount of noise and redundant data, the noise and redundant data will affect the quality of feature extraction and need to be preprocessed. Therefore, in the preprocessing stage, the noise and non-uniform distribution problems in the original three-dimensional point cloud data are reduced by bilateral filtering to retain geometric edge information. At the same time, the voxel gridding method is used for downsampling, and the dense point cloud data is sampled as a uniform distribution to generate a uniformly distributed point set, so as to reduce the computational complexity and reduce the amount of calculation. That is to say, the bilateral filtering and voxel gridding technology are used in the point cloud data preprocessing stage, which not only reduces noise interference, but also effectively retains key geometric details such as the component edge. Multi-stage preprocessing effectively reduces the false alarm rate and missed alarm rate, which can improve the reliability of subsequent damage detection.

[0056] It should be pointed out that the bilateral filter is used for noise reduction, where the bilateral filter is used to reduce the noise of the point cloud while retaining the edge details, and the spatial distance of the point cloud || p i -p j || and attribute differences || p i -p j ||Adjust the weight of each point cloud, namely:

[0057]

[0058] Among them, G s and G r represents the Gaussian weight function, N i Represents the point cloud p i The neighborhood point set, G s (||p i -p j ||) represents the spatial domain Gaussian weight, G r (|f(p i )-f(p j )| represents the intensity domain Gaussian weight.

[0059] And, Gs (||p i -p j ||) reflects the point cloud p j Distance point cloud p i The influence of is as follows:

[0060]

[0061] Where d = || p i -p j ||,σ s represents the spatial weight scale parameter;

[0062] and G r (|f(p i )-f(p j )|Reflects the influence of attribute differences, such as height differences, as shown in the following formula:

[0063]

[0064] Where r = |f(p i )-f(p j )|,σ r is the attribute weight scale parameter.

[0065] Step S12: extracting geometric features of the point cloud from the target three-dimensional point cloud data set, and constructing a point cloud feature map based on the geometric features of the point cloud.

[0066] In this embodiment, the acquired original 3D point cloud data set is preprocessed to obtain a target 3D point cloud data set, and the geometric features of the point cloud can be extracted from the target 3D point cloud data set. It can be understood that the goal of extracting point cloud geometric features is to extract features that describe local geometric shapes from the target 3D point cloud data set, that is, the extracted point cloud geometric features can be used to describe the surface details of the component, thereby providing a basis for damage detection.

[0067] In this embodiment, after extracting the point cloud geometric features used to describe the surface details of the component, a feature map can be constructed by combining the multi-dimensional geometric features, so as to comprehensively characterize the local and global geometric characteristics of the surface of the prefabricated component, providing more accurate basic data for subsequent analysis.

[0068] Step S13: converting the point cloud feature map into a corresponding CAD image based on a preset capsule network.

[0069] It can be understood that in the geometric feature conversion stage, the two-dimensional CAD (Computer-Aided Design) image generated by the feature map is used, that is, a multi-level capsule fusion network is used to perform deep learning analysis on the feature map to convert the feature map into a two-dimensional CAD image, which realizes the deep encoding and fusion of features, significantly improves the recognition ability of complex damage patterns, and makes up for the defect of insufficient expression of complex geometric information in traditional point cloud analysis.

[0070] It should be pointed out that the expression ability of CAD image features can be improved through the multi-level capsule fusion network, namely:

[0071]

[0072] Among them, s j represents a high-dimensional feature vector, u ij represents the input vector of the multi-level capsule fusion network, W ij represents the weight matrix learned in the network, c ij represents the weight distribution of the feature vector, and g represents the nonlinear activation function.

[0073] It is understandable that in order to further improve the feature representation capability, a multi-level capsule fusion network is introduced. The multi-level capsule network captures the feature relationship between different regions in the high-dimensional feature space through a dynamic routing mechanism. Its core operations include weight allocation and feature combinations This generates a more expressive high-dimensional feature vector.

[0074] In other words, compared with traditional detection methods based on shallow analysis, such as threshold segmentation or simple clustering, the multi-level capsule fusion network can significantly enhance the feature representation ability of complex geometric patterns. Among them, the capsule network uses a dynamic routing mechanism to establish semantic relationships between feature vectors, which can capture the multi-scale and multi-directional characteristics of component damage, and through high-dimensional encoding and fusion of point cloud features, it not only improves the ability to determine complex damage patterns, but also enhances the algorithm's adaptability to different component surface morphologies.

[0075] Specifically, the point cloud feature map is projected onto multiple two-dimensional planes, and the geometric feature value of each projected point is calculated using the feature value calculation function of the point; the geometric feature values ​​are fused using a multi-level capsule fusion network to generate a corresponding CAD image. It can be understood that the feature map of the point cloud is projected onto a two-dimensional plane to generate a CAD image representation.

[0076] For example, the point cloud feature map is projected along the XY, YZ, and XZ planes, that is, by projecting the point cloud, a three-dimensional feature map along the XY, YZ, and XZ planes is obtained, the geometric feature value of each projection point is calculated, and the geometric feature values ​​of the point cloud are fused using a multi-level capsule fusion network to generate a multi-view CAD image, namely:

[0077]

[0078] f(M)=α1k1+α2k2+α3||n||;

[0079] Among them, α1, α2, α3 represent feature weights, CAD xy (x, y) represents the projection image on the XY plane, and f(M) represents the weight function, that is, the eigenvalue calculation function of the projection point.

[0080] It should be pointed out that the eigenvalue of each projection point is calculated by the weight function f(M)=α1k1+α2k2+α3||n||, and the feature weights α1, α2, α3 can be optimized and adjusted according to the actual application situation to ensure that the contribution of each feature is reasonable.

[0081] Step S14: Detect the damaged area in the CAD image using a preset clustering algorithm, and perform region calibration on the damaged area on the CAD image.

[0082] In this embodiment, a preset clustering algorithm is used to detect damaged areas in the CAD image. The preset clustering algorithm may be DBSCAN (Density-Based Spatial Clustering of Applications with Noise), that is, the DBSCAN clustering algorithm is used to detect damaged areas in the CAD image. It can be understood that based on the CAD image, the DBSCAN clustering algorithm is used to perform efficient clustering analysis on abnormal areas, which avoids the threshold sensitivity problem in the traditional rule definition method and enhances the robustness and adaptability of the algorithm.

[0083] In this embodiment, the specific process of using the DBSCAN clustering algorithm to detect damaged areas in CAD images is as follows: determine the feature point set of the CAD image, and calculate the number of neighborhood points of each feature point in the feature point set; determine whether the number of neighborhood points of each feature point is less than a preset neighborhood point threshold; if the number of neighborhood points of the feature point is not less than the preset neighborhood point threshold, the feature point is determined to be a core point; if the number of neighborhood points of the feature point is less than the preset neighborhood point threshold, the feature point is determined to be an outlier, and a corresponding outlier point set is obtained; determine the number of outliers in the outlier point set and the total number of point cloud points of the target three-dimensional point cloud data set; calculate the degree of damage based on the number of outliers and the total number of point cloud points, and determine the corresponding damaged area according to the degree of damage.

[0084] For example, DBSCAN clustering is performed on the CAD image features to detect and identify abnormal areas (damaged areas) in the CAD image, that is, to determine the feature point set M of the CAD image. CAD , determine the predefined neighborhood radius ε, which reflects the density of local clustering and the minimum number of points MinPts, and then calculate any point cloud p i The number of neighboring points |N(i)|, if |N(i)|≥MinPts, then p is determined i is the core point. If |N(i)|≥MinPts is not satisfied, then p is determined i is an outlier point, and the outlier point set is output at the same time, that is: A={p i ∈M CAD |||p i -μ||>ε};

[0085] Among them, A represents the set of outliers.

[0086] In addition, the clustering determination of the above-mentioned outliers is based on the following formula:

[0087]

[0088] Then, based on the detected abnormal point set, the damage degree (the proportion of abnormal points to the total number of points) is calculated, and then the damaged area is determined according to the proportion of abnormal points to the total number of points. The calculation of the damage degree is shown in the following formula:

[0089]

[0090] Among them, ||A|| represents the number of abnormal points, and ||P′|| represents the total number of point cloud points in the target 3D point cloud dataset.

[0091] It can be understood that DBSCAN is a density-based clustering algorithm that determines the density of points by defining the neighborhood radius and the minimum number of neighborhood points. Core points satisfy the condition |N(i)|≥MinPts, that is, boundary points fall within the neighborhood of core points, while isolated points that do not satisfy the condition |N(i)|≥MinPts are determined to be noise. For the clustering of feature points in CAD images, the algorithm extracts abnormally distributed point sets based on local density differences, and further determines abnormal areas based on the distribution patterns of these point sets.

[0092] In this embodiment, the damaged area is calibrated on the CAD image, which may specifically include: calibrating the damaged area on the CAD image using a color coding method. It is understandable that the range and severity of the damaged area can be intuitively displayed through color coding.

[0093] For example, the outliers are calibrated on the CAD image, and the region of interest (ROI) is constructed by extracting the neighborhood of each outlier, that is:

[0094]

[0095] Among them, N(p i ) represents the neighborhood of the outlier point.

[0096] It should be pointed out that highlighting the abnormal area on the CAD image, calculating the target area ROI, and expressing it with color coding can not only help users quickly understand the detection results, but also provide a clear reference for subsequent repairs. For example, the damaged area is coded as a red area.

[0097] It can be understood that the DBSCAN clustering algorithm is used to identify damaged areas and make damage judgments, and at the same time, the damaged areas are visually calibrated on the CAD image. That is to say, after the geometric feature map is converted into a two-dimensional CAD image, the DBSCAN clustering algorithm is used to detect damaged areas in the CAD image. The algorithm identifies densely distributed areas and marks noise points through clustering, and visually calibrates the damaged areas on the CAD image, thereby obtaining a calibration map that can be used to quickly locate damaged areas. This visualization result makes it easy for engineers to intuitively understand the detection results and provides a direct basis for subsequent repair operations.

[0098] It can be seen that in the embodiment of the present invention, by preprocessing the point cloud data, and then extracting features from the preprocessed point cloud data, a feature map is constructed, and a preset capsule network is used to perform deep learning analysis on the feature map to convert the feature map into a two-dimensional CAD image, thereby achieving deep encoding and fusion of features, and then detecting the damaged area in the CAD image through a preset clustering algorithm, and visually calibrating the damaged area on the CAD image, thereby improving the accuracy and efficiency of damage detection, and enhancing the intuitiveness and engineering practicality of the detection results, thereby efficiently meeting the damage detection needs of prefabricated components in complex scenarios. That is, the technical solution of the present application is based on the geometric feature extraction of point cloud data, and through deep learning analysis and clustering analysis, the original three-dimensional point cloud is converted into high-level CAD image information, thereby achieving high-precision automated damage detection of prefabricated components, and can exhibit higher detection efficiency and robustness under large-scale point cloud data processing and complex scenarios.

[0099] See also Figure 2 As shown, the embodiment of the present invention discloses a specific method for detecting damage of assembled prefabricated components. Compared with the previous embodiment, this embodiment further illustrates and optimizes the technical solution.

[0100] Step S21, obtaining an original three-dimensional point cloud dataset of the prefabricated component, and preprocessing the original three-dimensional point cloud dataset to obtain a target three-dimensional point cloud dataset.

[0101] Step S22: Obtain neighborhood information of each point cloud in the target three-dimensional point cloud data set, and use the neighborhood information of the point cloud to extract corresponding curvature features and normal vectors.

[0102] It can be understood that by preprocessing the point cloud data, the neighborhood information of each point cloud in the preprocessed target three-dimensional point cloud data set is obtained, and then the neighborhood information of the point cloud is used to extract the curvature feature and the normal vector.

[0103] It should be noted that the curvature feature can represent the local surface shape through the principal curvatures k1 and k2, namely:

[0104]

[0105] Among them, λ max represents the maximum eigenvalue of the covariance matrix, λ min represents the minimum eigenvalue of the covariance matrix, which reflects the degree of change of the local surface main direction, and ∑λ represents the sum of all eigenvalues, which is used to normalize the curvature.

[0106] And, the normal vector represents the unit vector n calculated based on the spatial position of the point i , which represents the local surface direction of the point, that is:

[0107]

[0108] Among them, ω ij Represents the weight based on Euclidean distance, emphasizing the contribution of neighboring points to the calculation of normal vectors, P j -P i Represents the vectors between neighborhood points.

[0109] Step S23: combining the curvature feature and the normal vector into a feature vector, and constructing a corresponding point cloud feature map based on the feature vector.

[0110] In this embodiment, the curvature feature and the normal vector are combined into a feature vector, and a corresponding point cloud feature map is constructed based on the feature vector.

[0111] For example, on the preprocessed target 3D point cloud dataset, local features such as curvature features, such as the principal curvatures k1, k2 and the normal vector n are calculated based on the neighborhood of the point cloud. i , and further combine them into feature vectors, namely:

[0112] F i =[k1,k2,K,n];

[0113] Then, the corresponding point cloud feature map is constructed based on the feature vector, namely:

[0114] M={F i |i=1,2,K,N};

[0115] It should be pointed out that these local features comprehensively characterize the local geometric information of the point cloud and are subsequently processed in the form of feature maps.

[0116] Step S24: converting the point cloud feature map into a corresponding CAD image based on a preset capsule network.

[0117] Step S25: Detect the damaged area in the CAD image using a preset clustering algorithm, and perform region calibration on the damaged area on the CAD image.

[0118] For the specific contents of the above-mentioned step S21 and steps S24 to S25, reference may be made to the corresponding contents disclosed in the above-mentioned embodiments, which will not be repeated here.

[0119] It can be seen that in the embodiment of the present invention, by preprocessing the point cloud data, and then extracting features from the preprocessed point cloud data, a feature map is constructed, and a preset capsule network is used to perform deep learning analysis on the feature map to convert the feature map into a two-dimensional CAD image, thereby achieving deep encoding and fusion of features, and then detecting the damaged area in the CAD image through a preset clustering algorithm, and visually calibrating the damaged area on the CAD image, thereby improving the accuracy and efficiency of damage detection, and enhancing the intuitiveness and engineering practicality of the detection results, thereby efficiently meeting the damage detection needs of prefabricated components in complex scenarios. That is, the technical solution of the present application is based on the geometric feature extraction of point cloud data, and through deep learning analysis and clustering analysis, the original three-dimensional point cloud is converted into high-level CAD image information, thereby achieving high-precision automated damage detection of prefabricated components, and can exhibit higher detection efficiency and robustness under large-scale point cloud data processing and complex scenarios.

[0120] For example, see Figure 3 As shown, the original three-dimensional point cloud data of the prefabricated component is preprocessed by bilateral filtering, noise reduction and downsampling, and then geometric features are extracted from the preprocessed target point cloud, and the extracted geometric features are converted into CAD images. The CAD image conversion and feature fusion are realized through a multi-level capsule fusion network, and finally the damage detection judgment and calibration are completed through the DBSCAN clustering algorithm. That is, the technical solution of this application converts the original point cloud data into high-level information through multi-stage processing of point cloud geometric feature extraction, deep feature fusion and anomaly detection, and integrates the advantages of geometric analysis, deep learning and clustering technology. It has high detection accuracy and applicability, and finally realizes the automatic detection and intuitive visualization of prefabricated component damage. It can be seen that the technical solution of this application generates a two-dimensional CAD image from three-dimensional point cloud data by optimizing the point cloud preprocessing process, using point cloud geometric feature extraction technology combined with efficient deep learning model and density clustering algorithm, so that the point cloud feature analysis is combined with CAD tools, which provides an intuitive and efficient implementation path for the rapid calibration of damaged areas. This visualization technology facilitates the understanding of the results and subsequent engineering applications.

[0121] In one embodiment, if Figure 4 As shown, based on the above-mentioned prefabricated component damage detection method, the present invention also provides a prefabricated component damage detection device, including:

[0122] The point cloud data acquisition module 11 is used to acquire the original three-dimensional point cloud data set of the prefabricated component;

[0123] A point cloud data preprocessing module 12 is used to preprocess the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set;

[0124] A feature extraction module 13, used to extract geometric features of point clouds from the target three-dimensional point cloud data set;

[0125] A feature map construction module 14, configured to construct a point cloud feature map based on the geometric features of the point cloud;

[0126] A CAD image generation module 15, configured to convert the point cloud feature map into a corresponding CAD image based on a preset capsule network;

[0127] A damaged area detection module 16, configured to detect damaged areas in the CAD image using a preset clustering algorithm;

[0128] The damaged area calibration module 17 is used to calibrate the damaged area on the CAD image.

[0129] Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of the present application. The terminal may include:

[0130] A memory 501 , a processor 502 , and a computer program stored in the memory 501 and executable on the processor 502 .

[0131] When the processor 502 executes the program, the method for detecting damage of prefabricated components provided in the above embodiment is implemented.

[0132] Furthermore, the terminal further includes:

[0133] The communication interface 503 is used for communication between the memory 501 and the processor 502 .

[0134] The memory 501 is used to store computer programs that can be executed on the processor 502 .

[0135] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0136] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0137] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0138] The processor 502 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0139] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the above-mentioned method for detecting damage of prefabricated components is implemented.

[0140] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this application. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the claims.

[0141] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can read instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0143] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA, Programmable Gate Array), a field programmable gate array (FPGA, Field-Programmable Gate Array), etc.

[0144] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for detecting damage to prefabricated components, characterized in that: The method comprises: Acquire an original three-dimensional point cloud data set of the prefabricated component, and preprocess the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set; Extracting geometric features of a point cloud from the target three-dimensional point cloud data set, and constructing a point cloud feature map based on the geometric features of the point cloud; Converting the point cloud feature map into a corresponding CAD image based on a preset capsule network; A preset clustering algorithm is used to detect damaged areas in the CAD image, and the damaged areas are calibrated on the CAD image.

2. The method for detecting damage of prefabricated components according to claim 1, characterized in that: The preprocessing of the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set includes: Using a bilateral filter to process each point cloud in the original three-dimensional point cloud data set to obtain a processed three-dimensional point cloud data set; The processed 3D point cloud dataset is downsampled using voxel gridding to obtain a uniformly distributed target 3D point cloud dataset.

3. The method for detecting damage of prefabricated components according to claim 1, characterized in that: The step of extracting geometric features of point clouds from the target three-dimensional point cloud data set includes: Obtaining neighborhood information of each point cloud in the target three-dimensional point cloud data set, and extracting corresponding curvature features and normal vectors using the neighborhood information of the point cloud; Wherein, constructing a point cloud feature map based on the geometric features of the point cloud includes: The curvature feature and the normal vector are combined into a feature vector, and a corresponding point cloud feature map is constructed based on the feature vector.

4. The method for detecting damage of prefabricated components according to claim 1, characterized in that: The step of converting the point cloud feature map into a corresponding CAD image based on a preset capsule network includes: Projecting the point cloud feature map onto multiple two-dimensional planes, and calculating the geometric feature value of each projection point using the feature value calculation function of the point; The geometric feature values ​​are fused using a multi-level capsule fusion network to generate a corresponding CAD image.

5. The method for detecting damage of prefabricated components according to claim 1, characterized in that: The detecting the damaged area in the CAD image by using a preset clustering algorithm includes: The damaged area in the CAD image is detected using the DBSCAN clustering algorithm.

6. The method for detecting damage of prefabricated components according to claim 5, characterized in that: The method of detecting the damaged area in the CAD image by using the DBSCAN clustering algorithm includes: Determine a feature point set of the CAD image, and calculate the number of neighborhood points of each feature point in the feature point set; Determine whether the number of neighborhood points of each feature point is less than a preset neighborhood point threshold; If the number of neighborhood points of the feature point is less than the preset neighborhood point threshold, the feature point is determined to be an abnormal point, and a corresponding abnormal point set is obtained; Determine the number of outliers in the outlier point set and the total number of point cloud points in the target three-dimensional point cloud data set; The degree of damage is calculated based on the number of abnormal points and the total number of points in the point cloud, and the corresponding damaged area is determined according to the degree of damage.

7. The method for detecting damage of prefabricated components according to any one of claims 1 to 6, characterized in that: The step of performing regional calibration on the damaged area on the CAD image includes: The damaged area is calibrated on the CAD image using a color coding calibration method.

8. An assembled prefabricated component damage detection device, characterized in that: The device comprises: A point cloud data acquisition module is used to acquire the original three-dimensional point cloud data set of the prefabricated component; A point cloud data preprocessing module, used for preprocessing the original three-dimensional point cloud data set to obtain a target three-dimensional point cloud data set; A feature extraction module, used to extract geometric features of point clouds from the target three-dimensional point cloud data set; A feature map construction module, used to construct a point cloud feature map based on the geometric features of the point cloud; A CAD image generation module, used for converting the point cloud feature map into a corresponding CAD image based on a preset capsule network; A damaged area detection module, used to detect damaged areas in the CAD image using a preset clustering algorithm; The damaged area calibration module is used to calibrate the damaged area on the CAD image.

9. A terminal, characterized in that: include: A memory, a processor, and a prefabricated component damage detection program stored in the memory and executable on the processor, wherein the prefabricated component damage detection program, when executed by the processor, implements the steps of the prefabricated component damage detection method according to 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, and the computer program can be executed to implement the steps of the method for detecting damage of prefabricated components as claimed in any one of claims 1 to 7.

Images