A lightweight detection method for identifying building surface defects

Abstract

The present invention proposes a lightweight detection method for identifying building surface defects, which belongs to the field of computer vision and intelligent detection technology. The method includes the following steps: standardizing and pre-processing the three-dimensional point cloud data collected by the terahertz radar, performing local scanning with a large-size sliding window, and constructing a local feature vector; constructing a normal material template set through cluster analysis, and performing fine-grained scanning with a small-size sliding window. During the defect identification process, the defect situation of the previous small-size sliding window is prioritized for comparison with the state of the previous small-size sliding window to achieve rapid identification of the defect pattern. The present invention adopts a two-stage process of "coarse modeling and fine detection" to improve the system's ability to distinguish between local small defects and large-area anomalies; and supports efficient deployment of a lightweight processing framework: through dynamic window loading, local cache management, and adjacency traversal strategies, it significantly reduces memory and computing resource requirements, and is suitable for edge computing or embedded deployment scenarios.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and intelligent detection technology, and specifically to a method for detecting building surface defects based on three-dimensional point clouds. The method is a rapid detection method for building surface defects that combines adaptive extraction of point cloud features with a lightweight recognition strategy, and in particular relates to a lightweight detection method for identifying building surface defects. The method is suitable for scenarios such as intelligent building inspections and digital construction management. Background Art

[0002] Detecting building exterior wall defects is a critical component of quality management throughout the building lifecycle, encompassing multiple application scenarios, including quality control during construction, acceptance and evaluation at completion, periodic structural safety testing during operation, and subsequent maintenance and repair. Building an efficient, accurate, and intelligent exterior wall defect detection system is both a key requirement for project quality assurance and a crucial technical support for urban building safety management.

[0003] Traditionally, building exterior wall defects are detected using image-based methods. These methods employ camera arrays to capture high-definition images of the wall surface and combine them with image processing algorithms (such as edge detection, texture analysis, and deep learning) to identify defects. While mature and widely used, this approach has significant limitations. Firstly, images are subject to external factors such as view angle occlusion, lighting variations, and shadow interference, which can easily lead to detection errors and missed detections. Secondly, image information lacks the ability to express three-dimensional spatial information such as wall depth and concave-convex structures, making it difficult to identify defects with complex geometric features or deep structures.

[0004] To address these issues, 3D scanning technology has been introduced into building defect detection processes in recent years. Terahertz radar, a new high-frequency imaging technology, demonstrates significant potential for scanning and identifying building surface structures due to its strong penetration, high resolution, and contactless acquisition capabilities. Compared to traditional lidar or millimeter-wave radar, terahertz systems can acquire high-density, high-precision 3D point cloud information independently of lighting conditions, providing a data foundation for the 3D modeling and quantitative assessment of complex surface defects.

[0005] Despite the breakthroughs brought about by point cloud technology, existing systems still face several challenges in practical applications. First, point cloud data itself is high-dimensional and dense, resulting in large data volumes, which leads to high computational overhead and long response times for subsequent processing, making it difficult to meet real-time requirements. Second, given the wide variety of wall materials and complex surface textures, there is still a lack of effective strategies for adaptively extracting local features sensitive to defects from point clouds. Furthermore, current point cloud recognition algorithms mostly focus on high-precision modeling and geometric reconstruction, leaving gaps in lightweight, robust detection and processing of defect targets. Therefore, there is an urgent need to design a point cloud defect detection method with efficient data processing capabilities, adaptive local feature mining capabilities, and good generalization performance to achieve efficient identification and practical deployment of building exterior wall defects. Summary of the Invention

[0006] To address the existing problems of high computing resource consumption, low recognition accuracy, and inability to adapt to walls of different materials in the process of three-dimensional point cloud defect recognition on building surfaces, the present invention provides a lightweight detection method for building surface defect recognition. During dynamic sampling, this method takes "normal area dominance and abnormal difference extraction" as its core strategy, adopts phased window scanning and adaptive feature learning mechanisms, and achieves efficient recognition of defect areas under different wall material conditions. It has high efficiency, scalability, and migration robustness. This method can dynamically update the normal wall feature representation and identify low-contrast or large-area abnormal areas. At the same time, through modular design, it effectively controls memory consumption and improves the deployability and practicality of the system.

[0007] The specific technical solutions of the present invention are as follows:

[0008] A lightweight detection method for identifying building surface defects, the method comprising the following steps:

[0009] Step S11: performing standardized preprocessing on the three-dimensional point cloud data collected by the terahertz radar to obtain a standardized point cloud;

[0010] Step S12: Use a large-size sliding window to perform local scanning on the standardized point cloud, extract three local structural features: roughness, normal vector change rate, and point density, and construct a local feature vector;

[0011] Step S13: constructing a normal material template set through cluster analysis, and performing spatial connectivity analysis during the process of constructing the normal material template set;

[0012] Step S21: Use a small-size sliding window to perform fine-grained scanning on the standardized point cloud, and plan the sliding path of the small-size window according to the spatial adjacency rule, so that any small-size sliding window has spatial overlap or shared boundary points with the previous small-size sliding window;

[0013] Step S22: performing matching of the local feature vector of the first small-size sliding window with the normal material template set to determine defects;

[0014] Step S23: During the defect identification process of subsequent small-size sliding windows, different processes are performed according to the attribution label of the previous small-size sliding window. If the attribution label of the previous small-size sliding window is a defect mode, the current small-size sliding window is preferentially compared with the defect mode of the previous small-size sliding window to achieve rapid identification of the defect mode; if the attribution label of the previous small-size sliding window is a normal template, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window to quickly determine whether there is a defect.

[0015] Furthermore, step S11 includes the following steps:

[0016] The three-dimensional point cloud data collected by the terahertz radar is represented as ;in, Indicates the first 3D sampling points, represents the set of real numbers, Represents the number of three-dimensional sampling points; the standardized preprocessing calculation process is as follows:

[0017]

[0018]

[0019]

[0020] in, After preprocessing, A standardization point, is the geometric center of the point cloud, is the standard deviation of the point cloud spatial distribution, Represents the square of the Euclidean distance.

[0021] Furthermore, step S12 includes the following steps:

[0022] No. The corresponding sampling point set of the sliding window is:

[0023]

[0024] in, represents the center of the sliding window, represents the side length of the sliding window, Indicates the sliding window center is the center and the side length is The area of the cube. Sliding window Include Points, the number of swipes required to obtain a complete scan of the building surface;

[0025] No. Sliding window roughness Calculated as follows:

[0026]

[0027] in, represents transpose, Indicates the A sliding window local least squares fit of the plane normal vector, For the A sliding window plane offset term;

[0028] No. Sliding window normal vector change rate Calculated as follows:

[0029]

[0030] in, Indicates the Standardization points The main direction normal vector of the neighborhood;

[0031] No. Sliding window point density Calculated as follows:

[0032]

[0033]

[0034] in, Indicates the The convex hull volume of the point set in the sliding window, Indicates the The convex hull of the point set in the sliding window is formed. It represents the volume function;

[0035] The three local structural features of roughness, normal vector change rate and point density are integrated into the local feature vector. Sliding window local feature vector It is expressed as follows:

[0036] f j =[ r j , ∆ n j , ρ j ] T .

[0037] Furthermore, in step S13, the normal material template set is constructed in the following manner:

[0038] All sliding window local feature vectors constitute the spatial structure feature set , the spatial structure feature set is input into the clustering module to perform unsupervised learning, extract the main features, automatically cluster the areas that meet high consistency and structural continuity, and construct a normal material template set; the normal material template set is expressed as follows:

[0039]

[0040]

[0041] in, Represents a normal material template set; Indicates the total number of feature clusters; Indicates the A normal template; For the feature clusters, Indicates the The size of the eigenvalues of the feature clusters.

[0042] Furthermore, the determination of defects in matching the normal material template set in step S22 includes the following steps:

[0043] Step S221: Perform fine-grained scanning on the standardized point cloud of the current small-size sliding window to extract the roughness , normal vector change rate and point density Three local structural features to construct the local feature vector of the current small-size sliding window ;

[0044] Step S222: Calculate the defect response value of the current small-sized sliding window. If the defect response value is greater than or equal to the defect threshold, the current small-sized sliding window is considered a normal area, and the attribution label of the current small-sized sliding window is recorded. If the defect response value of the current small-sized sliding window is less than the defect threshold, mark the current small-sized sliding window as a suspected defect window.

[0045] Step S223: reconstructing the local boundary and performing multi-scale verification on the suspected defect window;

[0046] Step S224: Calculate the structural confidence under multi-scale radius and output the defect detection results.

[0047] Furthermore, in step S222, the current small-size sliding window defect response value is calculated as follows:

[0048]

[0049]

[0050]

[0051]

[0052] in, Indicates the current small-size sliding window defect response value, Represents the local feature vector of the current small-size sliding window Minimum Euclidean distance to the normal material template set; is a small positive number that prevents division by zero; represents the Euclidean distance; Indicates the Sliding window local feature vector The minimum Euclidean distance to the normal material template set, Represents the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding window and the normal material template set; Indicates that all Take the maximum value.

[0053] Furthermore, in step S23, the defect pattern of the current small-size sliding window is preferentially compared with the defect pattern of the previous small-size sliding window and calculated as follows:

[0054] Previous small sliding window The attribute label is defect mode , the local feature vector of the current small-size sliding window Prioritize defect patterns with the previous small sliding window size Compare and calculate the similarity to obtain the defect response value of the current small-size sliding window defect mode , calculated as follows:

[0055]

[0056]

[0057] in, Indicates the defect response value of the current small-size sliding window defect mode; Indicates the Sliding window local feature vector Minimum Euclidean distance to the defect characterization library; Represents the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding window and the defect characterization library.

[0058] Furthermore, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window and calculated as follows:

[0059] Previous small sliding window The attribute tag is a normal template , the local feature vector of the current small-size sliding window Prioritize the normal template of the previous small sliding window Perform a quick similarity comparison to obtain the local defect response value of the current small-size sliding window , calculated as follows:

[0060]

[0061] in, Indicates the local defect response value of the current small-size sliding window.

[0062] Compared with the prior art, the present invention has the following beneficial technical effects:

[0063] 1) The dynamic modeling of the present invention enhances generalization capabilities: the system automatically learns and updates the normal wall material template set through a large-scale sliding window scanning process, without the need for large-scale manual labeling, and has good cross-building and cross-material migration capabilities.

[0064] 2) The present invention uses phased scanning to improve recognition accuracy: a two-stage process of "coarse modeling-fine detection" is adopted to improve the system's ability to distinguish between small local defects and large-area anomalies, avoiding misjudgments and missed judgments.

[0065] 3) The lightweight processing framework of the present invention supports efficient deployment: through dynamic window loading, local cache management and adjacency traversal strategy, it significantly reduces memory and computing resource requirements, and is suitable for edge computing or embedded deployment scenarios.

[0066] 4) The defect response of the present invention is interpretable and reliable: Defect judgment is based on template matching and feature comparison, and the results are interpretable, which facilitates manual review and maintenance operations; the abnormal feedback mechanism ensures the robust operation of the system.

[0067] 5) The present invention provides efficient and accurate technical support for the quality management of buildings throughout their life cycle through intelligent and lightweight building surface defect detection technology. On the one hand, it greatly improves the efficiency and accuracy of identifying building exterior wall defects through automated detection, effectively reduces the safety risks and cost investment of manual inspections, provides reliable guarantees for urban building safety management, and helps prevent safety accidents caused by structural defects; on the other hand, it promotes the transformation of building inspection from traditional manual modes to intelligent and digital modes, which is in line with the development trend of "new infrastructure" and building industrialization. Its cross-material and cross-scenario generalization capabilities and lightweight deployment characteristics can be widely used in scenarios such as intelligent inspection and construction management, improving the overall technical level of the industry, and has important demonstration significance for promoting the safe upgrade of urban infrastructure and digital construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in describing the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0069] Figure 1 Schematic diagram of an overview of the lightweight inspection process for identifying building surface defects according to the present invention.

[0070] Figure 2 This is a schematic diagram of collecting the normal material template set of the present invention.

[0071] Figure 3 Schematic diagram of lightweight detection for identifying building surface defects according to the present invention. DETAILED DESCRIPTION

[0072] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the technical ideas proposed by the present invention, any extension, improvement or replacement made by ordinary technicians in the relevant fields without creative work should fall within the scope of protection of the present invention.

[0073] This paper proposes a lightweight detection method for identifying building surface defects. The overall solution is based on a dynamic sliding window strategy and performs three steps in stages: wall feature modeling, fine defect identification, and data loading optimization. It can significantly reduce memory and computing resource consumption while ensuring detection accuracy, and has good deployability and generalization capabilities.

[0074] This method is based on the input of original high-density point cloud. Relying on the sliding window mechanism, it first performs dynamic scanning and feature clustering of local areas with a larger window to extract and construct a representative set of normal wall material templates. Then, it performs a second-stage fine scan with a smaller window and compares the current area with the learned normal material template set to identify structural anomalies and surface defects. Finally, the block loading and memory scheduling mechanism of the point cloud area are coordinated throughout the entire process to ensure that the entire recognition system can run stably on resource-constrained platforms.

[0075] The present invention proposes a lightweight detection method for identifying building surface defects, such as Figure 1-3 As shown, the method includes the following steps:

[0076] Step S1: Standardize and preprocess the three-dimensional point cloud data collected by the terahertz radar to obtain a standardized point cloud; use a large-size sliding window to perform local scanning on the standardized point cloud, extract features and construct local feature vectors; construct a normal material template set through cluster analysis, and perform spatial connectivity analysis during the construction of the normal material template set.

[0077] Based on the assumption that most building walls are normal structures, this invention first preprocesses the 3D point cloud data collected by terahertz radar, including standardization operations such as geometric center translation and scale normalization. Subsequently, a large sliding window is used to dynamically scan a local area within the global point cloud. Material-related features within the windowed point cloud, such as roughness, normal vector change rate, and point density, are extracted to determine the surface properties and geometric stability of the area. Automatic clustering analysis is performed on regions that demonstrate high consistency and structural continuity during the scanning process, gradually establishing a set of normal material templates representing different wall material types. This normal material template set reflects the typical spatial distribution characteristics of walls under different material conditions and serves as a reference for subsequent defect identification. Furthermore, to prevent large-area defects from being misidentified as normal wall surfaces, the invention also incorporates a spatial connectivity analysis mechanism. When the features of multiple consecutive sliding windows deviate significantly from the current template and form a connected region with an area exceeding a set threshold, the system automatically marks the region as a potential misidentification area and triggers a request for manual feedback to ensure the accuracy and purity of the template set.

[0078] Specifically, step S1 includes the following steps:

[0079] Step S11: performing standardized preprocessing on the three-dimensional point cloud data collected by the terahertz radar to obtain a standardized point cloud.

[0080] Standardized preprocessing ensures the consistency of point cloud data across different building inspection projects. It is the basic operation for regional analysis and structural description, and improves the stability and migration capabilities of the system.

[0081] The three-dimensional point cloud data collected by the terahertz radar is represented as ;in, Indicates the first 3D sampling points, represents the set of real numbers, Indicates the number of 3D sampling points. In order to eliminate the influence of building scale difference, scanning area position offset and density change on subsequent feature analysis, the collected point cloud data needs to be standardized and preprocessed. Standardization preprocessing includes operations such as geometric center translation and scale normalization. The standardized point cloud after preprocessing is recorded as , its standardized preprocessing calculation process is as follows:

[0082]

[0083]

[0084]

[0085] in, After preprocessing, A standardization point, is the geometric center of the point cloud, is the standard deviation of the point cloud spatial distribution, Represents the square of the Euclidean distance. After normalization preprocessing, the point cloud is mapped to a unified scale and position framework, making subsequent sliding window sampling independent of the size of the building object, effectively improving the system's versatility and structural sensitivity.

[0086] Step S12: Use a large-size sliding window to perform local scanning on the standardized point cloud, extract three local structural features of roughness, normal vector change rate and point density, and construct a local feature vector.

[0087] Using a large sliding window to normalize point clouds Perform local scanning to extract statistical features of the building wall surface in terms of roughness, normal vector change rate and point density, and integrate these features into local feature vectors that describe regional properties.

[0088] No. The corresponding sampling point set of the sliding window is:

[0089]

[0090] in, represents the center of the sliding window, represents the side length of the sliding window, Indicates the sliding window center is the center and the side length is The area of the cube. Sliding window Include Points, The number of slides required to completely scan the building surface is calculated for the window sampling point set: roughness, normal vector change rate, and point density.

[0091] Roughness is used to measure the residual variance between the local point cloud and the fitting plane, indicating the degree of surface undulation and measuring whether the surface is smooth. Sliding window roughness Calculated as follows:

[0092]

[0093] in, represents transpose, Indicates the A sliding window local least squares fit of the plane normal vector, For the A sliding window plane offset term.

[0094] The normal vector change rate is used to measure the consistency of the point cloud normal vector direction in a local area, to identify corners or slightly distorted areas, and to measure surface consistency. Sliding window normal vector change rate Calculated as follows:

[0095]

[0096] in, Indicates the Standardization points The main direction normal vector of the neighborhood.

[0097] Point density reflects the spatial distribution density of points in an area, indicating the state of material particles or loss. Sliding window point density Calculated as follows:

[0098]

[0099]

[0100] in, Indicates the The convex hull volume of the point set in the sliding window, Indicates the The convex hull of the point set in the sliding window is formed. Represents the volume function.

[0101] The three local structural features of roughness, normal vector change rate and point density are integrated into the local feature vector. Sliding window local feature vector It is expressed as follows:

[0102] f j =[ r j , ∆ n j , ρ j ] T

[0103] The local feature vector comprehensively reflects the material characteristics of the wall and is the key basis for constructing a normal material template set.

[0104] Step S13: constructing a normal material template set through cluster analysis, and performing spatial connectivity analysis during the process of constructing the normal material template set.

[0105] Step S131: constructing a normal material template set through cluster analysis.

[0106] All sliding window local feature vectors constitute the spatial structure feature set , input the spatial structure feature set into the clustering module to perform unsupervised learning, extract the main features, automatically cluster the areas that meet high consistency and structural continuity, and construct a normal material template set. The clustering method can be selected -means or density-aware clustering (such as DBSCAN) to cope with scene complexity. The normal material template set is represented as follows:

[0107]

[0108]

[0109] in, Represents a normal material template set; the normal material template set is a template set composed of feature cluster center vectors, which serves as a structural comparison reference library for the subsequent defect identification stage; Indicates the total number of feature clusters; Indicates the A normal template is the center vector of the feature cluster, representing a typical wall representation structure; For the feature clusters, Indicates the The size of the eigenvalues of the feature clusters.

[0110] Step S132: performing spatial connectivity analysis, spatial anomaly detection and connectivity elimination in the process of constructing the normal material template set.

[0111] In order to avoid template contamination caused by scanning coverage of abnormal areas, spatial connectivity analysis is introduced in the process of constructing the normal material template set. That is, when it is detected that the features of multiple consecutive sliding windows deviate significantly from the current template and form a connected area in space with an area exceeding the set threshold, the system will automatically mark the area as a potential misjudgment area and trigger a manual feedback request to ensure the accuracy and purity of the template set.

[0112] Spatial connectivity analysis is achieved through the "structural deviation judgment + connectivity detection" mechanism. The "structural deviation judgment + connectivity detection" mechanism is as follows:

[0113] For each local feature vector of the sliding window, the minimum Euclidean distance between the local feature vector and the normal material template set is calculated. Taking the sliding window as an example, we introduce the calculation of Sliding window local feature vector Minimum Euclidean distance to the normal material template set , calculated as follows:

[0114]

[0115] in, represents the Euclidean distance. The decision threshold is expressed as , which is generally 1e-3; if the minimum Euclidean distance between the local feature vector of the sliding window and the normal material template set is greater than the judgment threshold, that is , the sliding window is considered to deviate significantly from the normal template and is marked as a “suspected abnormal” area.

[0116] Calculate the maximum value of the minimum Euclidean distance between all sliding window local feature vectors and the normal material template set. The calculation method is as follows:

[0117]

[0118] in, Represents the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding window and the normal material template set; Indicates that all Take the maximum value.

[0119] Satisfy all The sliding window of the abnormal window set , a three-dimensional connected graph is constructed based on the spatial adjacency relationship, which is expressed as follows:

[0120]

[0121] in, Represents a three-dimensional connected graph; Indicates finding connected components.

[0122] For each connected subset, its spatial coverage area is calculated based on the adjacent relationship of the windows, and the relationship between the spatial coverage area and the spatial coverage area threshold is judged. If the spatial coverage area is greater than the spatial coverage area threshold, the connected subset is judged as a "high-risk area for abnormal template contamination", and the features of this area are removed from the template candidate set, and the manual verification mechanism is triggered to ensure the purity and representativeness of the template.

[0123] First connected subsets As an example, we introduce the calculation of spatial coverage area. connected subsets Include Sliding window, connected subsets The spatial coverage area is The sum of the sliding window areas is expressed as , Indicates the spatial coverage area. The spatial coverage area threshold is expressed as ,like , then determine the connected subsets It is a "high-risk area for abnormal template contamination."

[0124] The "structural deviation judgment + connectivity detection" mechanism effectively prevents large damaged areas from being learned as "normal material templates" by judging the threshold. Control the deviation of single sliding window features and the spatial coverage area threshold Control the impact range of regional anomalies and improve the safety and practicality of normal material template set construction.

[0125] Step S2: Use a small-size sliding window to perform fine-grained scanning of the standardized point cloud, and plan the sliding path of the small-size window according to the spatial adjacency rule, so that any small-size sliding window has spatial overlap or shared boundary points with the previous small-size sliding window; match the local feature vector of the first small-size sliding window with the full normal material template set to determine defects; in the defect identification process of subsequent small-size sliding windows, different processes are performed according to the attribution label of the previous small-size sliding window. If the attribution label of the previous small-size sliding window is a defect mode, the current small-size sliding window is preferentially compared with the defect mode of the previous small-size sliding window to achieve rapid identification of the defect mode; if the attribution label of the previous small-size sliding window is a normal template, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window to quickly determine whether there is a defect.

[0126] After completing the normal material template set After the construction of the point cloud, the defect recognition stage begins. In this stage, a smaller sliding window is used to identify the standard point cloud. Fine-grained scanning is performed with the goal of extracting the structural features of each local area with high resolution, and performing structural comparison with a normal material template set, performing traversal and discrimination with higher spatial resolution.

[0127] After completing the construction of the normal material template set in step S1, the point cloud is traversed again with a smaller sliding window. For each local area, the system extracts three local structural features: roughness, normal vector change rate, and point density, and compares them with the normal material template set constructed in step S1. The defect response value is obtained by calculating the similarity. When the defect response value is lower than the defect threshold, the area is judged as a suspected defect area, and further strategies such as local boundary reconstruction, multi-scale comparison, and structural confidence estimation are used to refine the identification and improve the accuracy and robustness of the detection. This stage particularly enhances the detection capability of defect areas with blurred boundaries or subtle structural changes such as tiny cracks, shedding, and expansion.

[0128] To ensure the system can cope with the computational resource pressures posed by large-scale architectural point clouds, this paper designs an efficient data loading and resource scheduling mechanism. This divides the overall point cloud space into spatial voxels and only loads on-demand point cloud blocks within the area covered by the current sliding window. A traversal strategy based on spatial adjacency is employed along the sliding path to ensure data continuity and processing consistency between adjacent windows. Furthermore, the system maintains a fixed number of active windows and, in conjunction with cache invalidation and memory recycling strategies, dynamically releases data in invalid areas, effectively avoiding memory overflows caused by full map loading and significantly improving the detection system's scalability and field deployment capabilities.

[0129] In order to further improve the execution efficiency of the overall system, avoid repeated full template comparison in wall areas with strong structural continuity, and avoid the risk of resource overflow in large-scale point cloud scenarios, the present invention introduces a set of sliding scheduling and knowledge reference mechanisms based on spatial neighborhood continuity and context reuse to achieve a lightweight and efficient dynamic point cloud recognition strategy. Its core idea is "recognizing neighbors by neighbors", that is, using the material ownership information of adjacent windows in space to guide the current window to make quick decisions, thereby significantly reducing unnecessary computational overhead.

[0130] Specifically, step S2 includes the following steps:

[0131] Step S21: Use a small-size sliding window to perform fine-grained scanning on the standardized point cloud, and plan the small-size window sliding path according to the spatial adjacency rule so that any small-size sliding window has spatial overlap or shared boundary points with the previous small-size sliding window.

[0132] Specifically, the small-size sliding window path is continuous, so that the small-size sliding window slides continuously in one direction (such as the X axis) first, ensuring that the current small-size sliding window A small sliding window on it There is spatial overlap or at least shared boundary points.

[0133] Step S22: performing matching of the local feature vector of the first small-size sliding window with the normal material template set to determine defects.

[0134] The following steps are involved in determining the defect of matching a normal material template set:

[0135] Step S221: Perform fine-grained scanning on the standardized point cloud of the current small-size sliding window, extract three local structural features of roughness, normal vector change rate and point density, and construct the local feature vector of the current small-size sliding window.

[0136] Specifically, the current small-size sliding window , its coverage area is:

[0137]

[0138] in, Indicates the normalized point under the current small-size sliding window, , represents the side length of the small sliding window, The center of the small sliding window; Indicates the center of a small sliding window is the center and the side length is The current small size sliding window Include points.

[0139] Three local structural features, namely small-size sliding window roughness, normal vector change rate, and point density, are extracted.

[0140] Current small sliding window size Roughness Calculated as follows:

[0141]

[0142] in, Represents the local least squares fitting plane normal vector of the current small-size sliding window, Indicates the current small-size sliding window plane bias item.

[0143] Current small sliding window size The rate of change of the normal vector Calculated as follows:

[0144]

[0145] in, For the Standardization points The main direction normal vector of the neighborhood.

[0146] Current small sliding window size Point density Calculated as follows:

[0147]

[0148]

[0149] in, Indicates the current small-size sliding window The volume of the convex hull of the midpoint set.

[0150] The three local structural features extracted from the small-size sliding window roughness, normal vector change rate, and point density are integrated into the small-size sliding window local feature vector. The current small-size sliding window local feature vector It is expressed as follows:

[0151] f i ' =[ r i ' , ∆n i ' , ρ i ' ] T

[0152] This step extracts the local feature vector of each small-size sliding window at a higher spatial resolution. , its definition is consistent with step S1 to ensure structural consistency and comparability. This local feature vector serves as the basis for subsequent comparison with the normal material template set.

[0153] Step S222: Calculate the defect response value of the current small-size sliding window. If the defect response value is greater than or equal to the defect threshold, the current small-size sliding window is a normal area, and the attribution label of the current small-size sliding window is recorded. If the defect response value of the current small-size sliding window is less than the defect threshold, mark the current small-size sliding window as a suspected defect window.

[0154] The local feature vector extracted from the current small-size sliding window is compared with the normal material template set The template features in the image are compared to measure the degree of structural deviation and output the defect response value.

[0155] The local feature vector of the current small-size sliding window is calculated as the minimum Euclidean distance between the local feature vector and the normal material template set as the closest structural degree. Minimum Euclidean distance to the normal material template set is calculated as follows:

[0156]

[0157] Then perform similarity calculation to obtain the current small-size sliding window defect response value , so that the range is standardized to [0,1] , the similarity calculation formula is as follows:

[0158]

[0159] in, Represents the maximum value of the minimum Euclidean distance between all sliding window local feature vectors and the normal material template set, It is a small positive number to prevent division by zero, and is generally set to 1e-6. Current small-size sliding window defect response value The smaller it is, the greater the deviation between the current area and any normal material template, that is, the more likely it is a defective area.

[0160] Setting defect thresholds , perform material characterization similarity threshold screening, that is, if the current small-size sliding window defect response value Greater than or equal to the defect threshold ,Right now , then the current small-size sliding window is the normal area, and the local feature vector of the current small-size sliding window is The normal template with the smallest Euclidean distance to the normal material template set , recorded as the current small-size sliding window If the current small-size sliding window defect response value Less than the defect threshold ,Right now , then mark the current small-size sliding window It is a suspected defect window and is included in the next stage of refined identification.

[0161] This step uses the structural similarity metric to generate an interpretable "defect response value" for a small sliding window, taking the normal material template set as a reference. , quickly screen out potential abnormal areas and avoid generalization bias caused by erroneous learning.

[0162] Step S223: Perform local boundary reconstruction and multi-scale verification on the suspected defect window.

[0163] In order to further improve the robustness of defect recognition and suppress false detection, the defect response value is less than the defect threshold, that is, The suspected defect window is used to perform local boundary reconstruction, multi-scale verification and structural confidence assessment to ensure the stability and structural consistency of the output results.

[0164] For suspected defect windows, construct multiple resampling areas of different neighborhood scales ,in represents the scale radius, , Represents three different radii, reflecting multi-scale; resampling area It is expressed as follows:

[0165]

[0166] Resample area Represents the center of the sliding window with small size The distance is less than the normalized point of the scale radius;

[0167] For each scale radius Resampling area Extract local feature vectors , local eigenvector The extraction method is the same as the extraction method of local feature vectors in the sliding window. First, the resampling area is calculated The three local structural features of the internal point set, namely, roughness, normal vector change rate, and point density, are integrated into a local feature vector. Calculate the radius of each scale The local eigenvector of the resampled region The similarity with the normal material template set is obtained for each scale radius Defect response value , the similarity calculation formula is as follows:

[0168]

[0169] in, Indicates scale radius Defect response value, Indicates scale radius Local eigenvectors of the resampled region Minimum Euclidean distance from the normal material template set.

[0170] Step S224: Calculate the structural confidence under multi-scale radius and output the defect detection results.

[0171] Statistics show that the defect response value is lower than the defect threshold at multiple scale radii. The frequency of , defining the structural confidence for:

[0172]

[0173] in, is an indicator function, if but ;if but ; is the scale number.

[0174] The structural confidence threshold is expressed as ; If the structural confidence is greater than or equal to the structural confidence threshold, that is , then the current small-size sliding window is identified as a structural defect area and the detection result is output; and the local feature vector of the current small-size sliding window is Included in the defect characterization library as a defect candidate set , defect characterization library It is expressed as follows:

[0175]

[0176] in, represents the total number of defect feature clusters, Indicates the defect mode.

[0177] when When the local feature vector of the current small-size sliding window is directly As the 0th defect mode , and record it as the current small-size sliding window The attribution label.

[0178] when When the local feature vector of the current small-size sliding window is calculated The minimum Euclidean distance to the defect characterization library is calculated as follows:

[0179]

[0180] Current small-size sliding window local feature vector The defect pattern with the smallest Euclidean distance to the defect characterization library , recorded as the current small-size sliding window The attribution label.

[0181] Regularly perform cluster analysis on defect candidate sets to extract representative defect patterns , which is expressed as follows:

[0182]

[0183] in, Indicates the Defect clustering; Indicates the The size of the defect cluster eigenvalue. The newly extracted representative defect pattern Will be dynamically added to the defect characterization library , used for priority matching in the subsequent recognition process and building the system's structural anomaly recognition memory.

[0184] If the structural confidence is less than the structural confidence threshold, , which is regarded as a low-confidence noise area and should be ignored or re-inspected.

[0185] Confidence judgment based on the consistency of multi-scale structural responses can effectively solve the misleading effects of edge blur, smoothness and convexity on single-scale results, and improve the stability and generalization ability of the system under complex building structures.

[0186] Step S23: During the defect identification process of subsequent small-size sliding windows, different processes are performed according to the attribution label of the previous small-size sliding window. If the attribution label of the previous small-size sliding window is a defect mode, the current small-size sliding window is preferentially compared with the defect mode of the previous small-size sliding window; if the attribution label of the previous small-size sliding window is a normal template, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window to quickly determine whether there is a defect.

[0187] This invention uses the structural continuity assumption to greatly reduce the comparison frequency of model reasoning. It is particularly suitable for scenarios with extremely high structural continuity, such as building exterior walls, which not only ensures the integrity of regional coverage but also improves the system operation efficiency.

[0188] Step S231: If the attribution label of the previous small-size sliding window is a defect pattern, the current small-size sliding window is preferentially compared with the defect pattern of the previous small-size sliding window to achieve rapid identification of the defect pattern.

[0189] The method of step S221 is used to perform fine-grained scanning on the standardized point cloud of the current small-size sliding window, extract the three local structural features of roughness, normal vector change rate and point density, and construct the local feature vector of the current small-size sliding window. .

[0190] If the previous small sliding window The attribute label is defect mode , the local feature vector of the current small-size sliding window Prioritize defect patterns with the previous small sliding window size Compare and calculate the similarity to obtain the defect response value of the current small-size sliding window defect mode , calculated as follows:

[0191]

[0192]

[0193] in, Indicates the defect response value of the current small-size sliding window defect mode; Indicates the first Sliding window local feature vector Minimum Euclidean distance to the defect characterization library; Indicates the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding windows and the defect characterization library in step S132.

[0194] Setting the continuous defect threshold , if the defect response value of the current small-size sliding window defect mode is greater than or equal to the continuous defect threshold, that is When , it is considered that there are continuous defects in the current small-size sliding window, the detection results are output, and the defect pattern of the previous small-size sliding window is Recorded as the current small size sliding window The attribution label.

[0195] If the defect response value of the current small-size sliding window defect mode is less than the continuous defect threshold, that is, When the defect is detected, execute steps S222 to S224 to perform matching of all normal material template sets to determine the defect.

[0196] Step S232: If the attribution label of the previous small-size sliding window is a normal template, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window to quickly determine whether there is a defect.

[0197] The method of step S221 is used to perform fine-grained scanning on the standardized point cloud of the current small-size sliding window, extract the three local structural features of roughness, normal vector change rate and point density, and construct the local feature vector of the current small-size sliding window. .

[0198] If the previous small sliding window The attribute tag is a normal template , the local feature vector of the current small-size sliding window Prioritize the normal template of the previous small sliding window Perform a quick similarity comparison to obtain the local defect response value of the current small-size sliding window , calculated as follows:

[0199]

[0200] in, Indicates the local defect response value of the current small-size sliding window.

[0201] If the local defect response value of the current small-size sliding window is greater than or equal to the defect threshold, that is , the current small-size sliding window is considered to be a normal area, and the normal template of the previous small-size sliding window is Recorded as the current small size sliding window The attribution label.

[0202] If the local defect response value of the current small-size sliding window is less than the defect threshold, that is, When the defect is detected, execute steps S222 to S224 to perform matching of all normal material template sets to determine the defect.

[0203] This invention leverages spatial continuity to skip most redundant template matching processes, improving speed and further enhancing its ability to quickly respond to "material-dominated continuous segments." By "observing, recording, and reusing" defect structures, the system gradually accumulates experience in identifying defect types during the recognition process, achieving self-growth from a cold start to stable recognition performance.

[0204] This invention develops a lightweight building surface defect recognition process based on three phases: normal material template modeling, small-scale sliding window defect identification, and neighborhood-aware scheduling and loading. It demonstrates structural learnability, responsive judgment, and computational controllability. The normal material template set in step S1 provides a highly structurally consistent reference baseline, supporting the rapid similarity calculation of local features within each scanned small-scale sliding window in step S2. Simultaneously, the defect response sequence generated in step S2 continuously constructs and optimizes a defect structure response library through a multi-scale confidence mechanism and anomaly clustering process, enabling the system to autonomously identify and memorize unknown defect types. Furthermore, the small-scale sliding window scheduling logic is bound to the adjacent window material context, implementing an adaptive sliding strategy of "neighborhood-driven, prioritized comparison, and timely recycling," ensuring consistent low-load operation when processing large-scale point cloud data. Through the synergistic integration of these steps, the invention achieves high-precision, low-resource, and highly interpretable dynamic recognition of complex building surface defects, with excellent adaptability to field deployments.

[0205] It is understood that the present invention describes the technical solution through multiple embodiments. Under the guidance of the present invention, those skilled in the art can reasonably adjust and equivalently replace the process structure, data organization form, or discrimination algorithm without departing from the core concept. Therefore, the scope of protection of the present invention should not be limited to the specific embodiments described in this specification, but should be based on the scope of the appended claims, covering all variant solutions that essentially achieve the same technical effect.

Claims

1. A lightweight detection method for identifying building surface defects, characterized in that: The method comprises the following steps: Step S11: performing standardized preprocessing on the three-dimensional point cloud data collected by the terahertz radar to obtain a standardized point cloud; Step S12: Use a large-size sliding window to perform local scanning on the standardized point cloud, extract three local structural features: roughness, normal vector change rate, and point density, and construct a local feature vector; Step S13: constructing a normal material template set through cluster analysis, and performing spatial connectivity analysis during the process of constructing the normal material template set; Step S21: Use a small-size sliding window to perform fine-grained scanning on the standardized point cloud, and plan the sliding path of the small-size window according to the spatial adjacency rule, so that any small-size sliding window has spatial overlap or shared boundary points with the previous small-size sliding window; Step S22: performing matching of the local feature vector of the first small-size sliding window with the normal material template set to determine defects; Step S23: During the defect identification process of subsequent small-size sliding windows, different processes are performed according to the attribution label of the previous small-size sliding window. If the attribution label of the previous small-size sliding window is a defect mode, the current small-size sliding window is preferentially compared with the defect mode of the previous small-size sliding window to achieve rapid identification of the defect mode; if the attribution label of the previous small-size sliding window is a normal template, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window to quickly determine whether there is a defect.

2. The lightweight detection method for identifying building surface defects according to claim 1 is characterized in that: Step S11 includes the following steps: The three-dimensional point cloud data collected by the terahertz radar is represented as ;in, Indicates the first 3D sampling points, represents the set of real numbers, Represents the number of three-dimensional sampling points; the standardized preprocessing calculation process is as follows: in, After preprocessing, A standardization point, is the geometric center of the point cloud, is the standard deviation of the point cloud spatial distribution, Represents the square of the Euclidean distance.

3. The lightweight detection method for identifying building surface defects according to claim 2, characterized in that: Step S12 includes the following steps: No. The corresponding sampling point set of the sliding window is: in, represents the center of the sliding window, represents the side length of the sliding window, Indicates the sliding window center is the center and the side length is The cubic area of Sliding window Include Points, the number of swipes required to obtain a complete scan of the building surface; No. Sliding window roughness Calculated as follows: in, represents transpose, Indicates the A sliding window local least squares fit of the plane normal vector, For the A sliding window plane offset term; No. Sliding window normal vector change rate Calculated as follows: in, Indicates the Standardization points The main direction normal vector of the neighborhood; No. Sliding window point density Calculated as follows: in, Indicates the The convex hull volume of the point set in the sliding window, Indicates the The convex hull of the point set in the sliding window is formed. It represents the volume function; The three local structural features of roughness, normal vector change rate and point density are integrated into the local feature vector. Sliding window local feature vector It is expressed as follows: 。 4. The lightweight detection method for identifying building surface defects according to claim 3 is characterized in that: In step S13, the normal material template set is constructed as follows: All sliding window local feature vectors constitute the spatial structure feature set , the spatial structure feature set is input into the clustering module to perform unsupervised learning, extract the main features, automatically cluster the areas that meet high consistency and structural continuity, and construct a normal material template set; the normal material template set is expressed as follows: in, Represents a normal material template set; Indicates the total number of feature clusters; Indicates the A normal template; For the feature clusters, Indicates the The size of the eigenvalues of the feature clusters.

5. The lightweight detection method for identifying building surface defects according to claim 4 is characterized in that: The determination of defects in matching the normal material template set in step S22 includes the following steps: Step S221: Perform fine-grained scanning on the standardized point cloud of the current small-size sliding window to extract the roughness , normal vector change rate and point density Three local structural features to construct the local feature vector of the current small-size sliding window ; Step S222: Calculate the defect response value of the current small-sized sliding window. If the defect response value is greater than or equal to the defect threshold, the current small-sized sliding window is considered a normal area, and the attribution label of the current small-sized sliding window is recorded. If the defect response value of the current small-sized sliding window is less than the defect threshold, mark the current small-sized sliding window as a suspected defect window. Step S223: reconstructing the local boundary and performing multi-scale verification on the suspected defect window; Step S224: Calculate the structural confidence under multi-scale radius and output the defect detection results.

6. The lightweight detection method for identifying building surface defects according to claim 5, characterized in that: In step S222, the current small-size sliding window defect response value is calculated as follows: in, Indicates the current small-size sliding window defect response value, Represents the local feature vector of the current small-size sliding window Minimum Euclidean distance to the normal material template set; is a small positive number that prevents division by zero; represents the Euclidean distance; Indicates the Sliding window local feature vector The minimum Euclidean distance to the normal material template set, Represents the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding window and the normal material template set; Indicates that all Take the maximum value.

7. The lightweight detection method for identifying building surface defects according to claim 6, characterized in that: In step S23, the defect pattern of the current small-size sliding window is compared with the defect pattern of the previous small-size sliding window and calculated as follows: Previous small sliding window The attribute label is defect mode , the local feature vector of the current small-size sliding window Prioritize defect patterns with the previous small sliding window size Compare and calculate the similarity to obtain the defect response value of the current small-size sliding window defect mode , calculated as follows: in, Indicates the defect response value of the current small-size sliding window defect mode; Indicates the Sliding window local feature vector Minimum Euclidean distance to the defect characterization library; Represents the maximum value of the minimum Euclidean distance between all local feature vectors of the sliding window and the defect characterization library.

8. The lightweight detection method for identifying building surface defects according to claim 6, characterized in that: In step S23, the current small-size sliding window is preferentially compared with the normal template of the previous small-size sliding window and calculated as follows: Previous small sliding window The attribute tag is a normal template , the local feature vector of the current small-size sliding window Prioritize the normal template of the previous small sliding window Perform a quick similarity comparison to obtain the local defect response value of the current small-size sliding window , calculated as follows: in, Indicates the local defect response value of the current small-size sliding window.

Images