Electric pole quality detection method

Abstract

The invention relates to the technical field of electric pole detection, in particular to an electric pole quality detection method, which comprises the following steps: acquiring electric pole three-dimensional point cloud data of an electric pole to be detected; comparing the electric pole surface feature point coordinate set with a preset standard electric pole feature point coordinate set point by point; judging the coordinate offset of each feature point and a preset coordinate offset threshold; analyzing the distribution information of the surface deformation area of the electric pole; performing noise filtering and structure layering processing on the original scanning data of the internal structure of the electric pole; defect feature extraction is carried out on the distribution data of the steel bars in the electric pole and the concrete matrix data; and inputting the electric pole surface deformation area distribution information, the steel bar layer defect distribution information and the concrete layer defect distribution information into the electric pole quality comprehensive analysis model for quality state evaluation. According to the invention, the deformation area can be accurately positioned, the internal detection range can be associated, the internal structure damage caused by external deformation can be accurately identified, and the accuracy of electric pole quality detection is improved.

Description

Technical Field

[0001] This invention relates to the field of pole inspection technology, and in particular to a method for inspecting pole quality. Background Technology

[0002] As a core supporting component in power transmission networks, utility poles are widely used in urban and rural power grid construction, communication line laying, and other fields. Their structural integrity directly affects the stability and safety of power transmission. Utility poles are typically made of concrete, steel, or composite materials and are exposed to the natural environment for extended periods, enduring various external forces such as wind loads, temperature variations, mechanical stress, and potential collisions. This makes them prone to surface damage and internal structural deterioration. Surface damage includes cracks, spalling, corrosion, and localized deformation; internal structural damage involves reduced material strength, steel reinforcement corrosion, prestress loss, and internal voids. If these damages are not detected and assessed in a timely manner, they may cause poles to tilt, break, or even collapse, leading to power outages, communication failures, and safety accidents.

[0003] Currently, pole quality inspection mainly relies on manual inspection and general non-destructive testing (NDT) techniques. Manual inspection involves visual observation, tapping and listening, and simple measuring tools. The results are heavily influenced by the inspector's experience and subjective judgment, and it is difficult to detect hidden internal damage. General NDT techniques include ultrasonic testing, infrared thermography, radar detection, and vibration frequency analysis. Ultrasonic testing identifies internal defects based on the propagation characteristics of sound waves in materials, but it lacks sufficient analysis of the correlation between surface deformation areas and internal damage. Infrared thermography identifies anomalies based on surface temperature distribution, but it is easily affected by ambient temperature and sunlight. Radar detection can detect the distribution of internal reinforcing steel and voids, but it is difficult to accurately distinguish the range of internal damage caused by external deformation. Vibration frequency analysis assesses overall performance through changes in structural dynamic characteristics, but it cannot locate localized damage areas. The detection of surface feature points and the identification of internal damage are often independent, lacking a method to accurately correlate surface deformation areas with the internal detection range. Surface feature points refer to geometric or textural anomalies formed on the pole surface due to damage, such as crack endpoints, depression centers, or corrosion patches. While traditional methods can identify some surface feature points, they fail to establish a spatial mapping relationship between these points and internal structural damage, leading to ambiguity in the detection range. For example, when localized dents appear on the pole surface, current technology struggles to accurately infer the depth and extent of internal damage based on the geometric features of the dent, potentially resulting in missed or false detections. Furthermore, pole damage often progresses from the outside in; external deformation induces stress concentration that gradually diffuses into the internal structure, making it difficult to accurately identify internal structural damage caused by external deformation. This frequently leads to missed or false detections, reducing the accuracy and reliability of pole quality inspection and failing to meet the high standards required by power systems for pole quality and safety assurance.

[0004] Chinese Patent Publication No. CN117092115A discloses a method for detecting defects in cement poles, including: S1. acquiring a three-dimensional image of the cement pole to be inspected, and analyzing the appearance impact index of the cement pole; S2. identifying and dividing the structural information of the cement pole to be inspected, obtaining the steel reinforcement layer and the concrete layer; S3. detecting defects in the steel reinforcement layer of the cement pole to be inspected, and calculating the defect impact index of the steel reinforcement layer; S4. detecting defects in the concrete layer of the cement pole to be inspected, and calculating the defect impact index of the concrete layer; S5. comprehensively evaluating the defect degree coefficient of the cement pole to be inspected, and providing control and prompting for the defect degree of the cement pole to be inspected. This solution only divides the cement pole into steel reinforcement layer and concrete layer and calculates the defect impact degree separately. It cannot accurately locate the deformation area through surface feature points and correlate it with the internal detection range, making it difficult to accurately identify internal structural damage caused by external deformation, thus reducing the accuracy of pole quality inspection. Summary of the Invention

[0005] To address this issue, the present invention provides a method for inspecting the quality of utility poles, which overcomes the problem in the prior art that the deformation area cannot be accurately located and associated with the internal detection range through surface feature points, resulting in difficulty in accurately identifying internal structural damage caused by external deformation and reducing the accuracy of utility pole quality inspection.

[0006] To achieve the above objectives, the present invention provides a method for inspecting the quality of utility poles, comprising the following steps: S1. The pole to be inspected is scanned using a 3D scanning device to obtain 3D point cloud data of the pole. The initial image set of the entire pole is converted to grayscale according to the image grayscale processing algorithm to obtain a grayscale image set of the entire pole. The surface feature points of the grayscale image set of the entire pole are extracted to obtain a set of coordinates of surface feature points of the pole. S2. Compare the coordinate set of feature points on the pole surface with the preset standard pole feature point coordinate set point by point to obtain the coordinate offset of each feature point. S3. The coordinate offset of each feature point is compared with the preset coordinate offset threshold to obtain the set of deformation feature points whose coordinate offset exceeds the preset coordinate offset threshold. The set of deformation feature points is then aggregated to obtain the distribution information of the deformation area on the pole surface. S4. The spatial range of the deformation area distribution information on the surface of the pole is analyzed by the detection range positioning algorithm to obtain the three-dimensional coordinate range of the detection target area inside the pole. The internal structure of the pole part corresponding to the three-dimensional coordinate range is scanned by the ultrasonic detection device to obtain the original scanning data of the internal structure of the pole. S5. Noise filtering is performed on the original scanning data of the internal structure of the pole to obtain the noise-reduced data of the internal structure of the pole. The noise-reduced data of the internal structure of the pole is then processed by the layer segmentation algorithm to obtain the data of the distribution of steel bars and the concrete matrix inside the pole. S6. Extract defect features from the internal steel reinforcement distribution data and concrete matrix data of the pole respectively to obtain the defect distribution information of the steel reinforcement layer and the concrete layer. S7. Input the distribution information of the deformation area on the pole surface, the distribution information of the defects in the steel reinforcement layer, and the distribution information of the defects in the concrete layer into the pole quality comprehensive analysis model for multi-dimensional feature fusion and quality status assessment to obtain the target pole quality detection result. The pole quality comprehensive analysis model is obtained by supervised training using a historical dataset containing pole surface deformation features, internal structural defect data, and corresponding quality rating labels, with the training objective being to minimize the model prediction error.

[0007] Compared with the prior art, the beneficial effects of this application are as follows: Three-dimensional point cloud data of the pole is acquired by a 3D scanning device. Combined with an image grayscale processing algorithm, a global grayscale image set of the pole is obtained. Then, surface feature point extraction yields a set of coordinates for the pole's surface feature points. This set is then compared point-by-point with a preset standard set of pole feature point coordinates to obtain coordinate offsets. A set of deformation feature points is selected based on a preset coordinate offset threshold and their neighborhood is aggregated to accurately obtain the distribution information of the deformation area on the pole's surface. A detection range positioning algorithm is then used to analyze this distribution information to obtain the 3D coordinate range of the target area inside the pole for detection. This guides the ultrasonic detection device to perform targeted scanning, overcoming the problem that existing technologies cannot accurately locate the deformation area and correlate it with the internal detection range using surface feature points. This achieves a precise correspondence between surface deformation and the internal detection area, avoiding the blindness of ultrasonic detection and ensuring accurate capture of internal structural damage caused by external deformation, significantly improving the accuracy of pole quality inspection.

[0008] Noise filtering is applied to the raw scanning data of the internal structure of the utility pole obtained by ultrasonic detection to improve data purity and obtain noise-reduced data of the internal structure. Then, a layered segmentation algorithm is used to accurately separate the distribution data of the reinforcing steel and the concrete matrix inside the pole, and defect features are extracted to obtain the distribution information of defects in the reinforcing steel layer and the concrete layer, respectively. Finally, the distribution information of surface deformation areas, the distribution information of defects in the reinforcing steel layer and the distribution information of defects in the concrete layer are input into the comprehensive quality analysis model of the utility pole. Through multi-dimensional feature fusion, the quality status assessment is realized, which further overcomes the problem that existing technologies are difficult to accurately identify internal structural damage caused by external deformation. Noise reduction and layered segmentation ensure the accuracy of internal defect information. Combined with a well-trained comprehensive analysis model, the correlation assessment of surface and internal defects is realized, avoiding the limitations of single-dimensional detection. The detection results more comprehensively reflect the quality status of the utility pole. The entire process from data processing to assessment improves the detection accuracy and effectively solves the problem that existing technologies lack multi-dimensional correlation assessment, resulting in one-sided detection results and insufficient accuracy.

[0009] Furthermore, in S1, the mathematical expression for the image grayscale processing algorithm is: In the formula, Represents coordinates in the image The grayscale value of a pixel. Represents coordinates in the image The red channel pixel value of the pixel. Represents coordinates in the image The green channel pixel value of the pixel. Represents coordinates in the image The blue channel pixel value of the pixel. Indicates the use of adjusting coordinates in the image The red channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Green channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Blue channel pixel value of a pixel The weighting coefficients, .

[0010] In this solution, by adjusting the red, green, and blue channel pixel values ​​of coordinate pixels in the image, the color information of each channel is precisely integrated, so that the resulting grayscale value can truly reflect the color distribution characteristics of the original image, thereby improving the targeting and accuracy of image grayscale processing.

[0011] Furthermore, S2 includes the following steps: S21. Obtain a preset standard pole 3D model, and extract the preset standard pole feature point coordinate set from the preset standard pole 3D model according to the pole surface feature point coordinate set; S22. Establish the nearest neighbor correspondence between the set of feature point coordinates on the pole surface and the preset standard pole feature point coordinate set, and calculate the coordinate offset of each pair of feature points with established correspondence in the nearest neighbor correspondence. The mathematical expression for the coordinate offset is: In the formula, This represents the coordinate offset of the i-th feature point. This represents the three-dimensional coordinates of the i-th feature point in the set of feature point coordinates on the pole surface. This represents the three-dimensional coordinates of the i-th feature point in the preset set of feature point coordinates for standard utility poles.

[0012] In this scheme, by acquiring a 3D model of a pre-set standard pole and extracting and generating a set of coordinates of feature points of the pre-set standard pole, a nearest neighbor correspondence is established between the set of coordinates of feature points on the pole surface and the set of coordinates of feature points of the pre-set standard pole. The coordinate offset of each pair of feature points with established correspondence is calculated, which can accurately quantify the geometric differences between the actual pole and the standard model.

[0013] Furthermore, S3 includes the following steps: S31. Obtain the coordinate offset of each feature point, and compare the coordinate offset of each feature point with the preset coordinate offset threshold one by one. Mark the feature points whose coordinate offset exceeds the preset coordinate offset threshold to form an initial set of deformation feature points. S32. Based on the cylindrical structural features of the pole, a spherical spatial neighborhood search range is constructed. Then, a neighborhood search is performed on each feature point in the initial set of deformation feature points according to the spherical spatial neighborhood search range to obtain other deformation feature points within the spherical spatial neighborhood search range. Finally, the aggregation degree between any two deformation feature points within the spherical spatial neighborhood search range is calculated to obtain the aggregation degree matrix between feature points. The mathematical expression for the aggregation degree is: In the formula, This represents the degree of aggregation between the i-th deformation feature point and the j-th deformation feature point. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point along the x-axis in the world coordinate system. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point in the y-axis direction of the world coordinate system. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point in the z-axis direction of the world coordinate system; S33. Based on the aggregation degree matrix between feature points, deformable feature points that meet the preset aggregation conditions are grouped into the same deformable region, and the boundary coordinates, area and contour morphology information of each deformable region are extracted and integrated to obtain the distribution information of deformable regions on the pole surface.

[0014] In this scheme, the coordinate offset of each feature point is obtained and compared with a preset coordinate offset threshold one by one to accurately mark the initial set of deformation feature points. Based on the cylindrical structure features of the pole, a spherical spatial neighborhood search range is constructed. The neighborhood search is carried out in a targeted manner and the aggregation degree between deformation feature points is calculated to obtain the aggregation degree matrix between feature points. Then, the deformation feature points that meet the preset aggregation conditions are grouped into the same deformation region. The boundary coordinates, area and contour morphology information of each deformation region are completely extracted to obtain the distribution information of deformation regions on the surface of the pole, thereby improving the accuracy of pole deformation recognition and region division.

[0015] Furthermore, S4 includes the following steps: S41. Extract the three-dimensional coordinate boundaries of each deformation region from the distribution information of deformation regions on the surface of the pole, and input the three-dimensional coordinate boundaries of each deformation region into the detection range positioning algorithm. Based on the corresponding mapping relationship between the surface of the pole and the internal structure, perform spatial coordinate expansion calculation to obtain the three-dimensional coordinate range of the detection target area inside the pole. S42. Based on the three-dimensional coordinate range of the target area inside the pole, adjust the detection parameters of the ultrasonic detection device to obtain the adjusted ultrasonic detection device, and scan the pole part corresponding to the three-dimensional coordinate range area by area based on the adjusted ultrasonic detection device. S43. Receive the reflected and transmitted signals returned by the adjusted ultrasonic detection device during the scanning process, preprocess the reflected and transmitted signals to obtain the original scanning data of the internal structure of the pole.

[0016] In this scheme, the three-dimensional coordinate boundaries of each deformation region in the distribution information of deformation regions on the pole surface are extracted. Combined with the corresponding mapping relationship between the pole surface and the internal structure, the spatial coordinate expansion calculation is performed by the detection range positioning algorithm to accurately determine the three-dimensional coordinate range of the detection target area inside the pole. Based on the three-dimensional coordinate range, the detection parameters of the ultrasonic detection device are adjusted and a region-by-region scan is carried out. The returned reflected and transmitted signals are preprocessed to improve the accuracy of the positioning of the detection target area inside the pole and the targeting of ultrasonic detection.

[0017] Furthermore, S5 includes the following steps: S51. An adaptive filtering algorithm based on wavelet transform is used to filter noise from the original scanning data of the internal structure of the pole, resulting in noise-reduced data of the internal structure of the pole. S52. Construct the segmentation threshold conditions for the hierarchical segmentation algorithm, and divide the noise reduction data of the internal structure of the pole according to the segmentation threshold conditions to obtain several structural layer data. S53. Perform acoustic feature extraction and material attribute identification on the data of each structural layer to obtain the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material. Perform coordinate calibration and format regularization on the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material to obtain the steel reinforcement distribution data and concrete matrix data inside the pole.

[0018] In this scheme, an adaptive filtering algorithm based on wavelet transform is used to filter noise from the original scanning data of the internal structure of the pole, thereby improving the purity of the noise-reduced data. The segmentation threshold conditions of the hierarchical segmentation algorithm are used to divide the noise-reduced data into several structural layers. The structural layer data corresponding to the steel reinforcement material and concrete matrix material are accurately distinguished by acoustic feature extraction and material attribute identification. Through coordinate calibration and format regularization, the accuracy and standardization of the steel reinforcement distribution data and concrete matrix data inside the pole are ensured.

[0019] Furthermore, S6 includes the following steps: S61. Perform connected component analysis on the distribution data of the steel bars inside the pole, identify the set of reflection points belonging to the same steel bar, calculate the geometric center line, diameter fluctuation and reflection signal continuity index of each connected component, and when the diameter fluctuation exceeds the preset diameter fluctuation range or the reflection signal continuity index is lower than the preset reflection signal continuity index threshold, determine that there is a steel bar layer defect in the connected component and generate steel bar layer defect distribution information. S62. The concrete matrix data is analyzed by a signal amplitude feature statistical algorithm to obtain the signal amplitude uniformity index and average amplitude value of each analysis region. When the signal amplitude uniformity index of the analysis region is lower than the preset signal amplitude uniformity index threshold and the average amplitude value is lower than the preset average amplitude value threshold, it is determined that there is a concrete layer defect in the analysis region, and concrete layer defect distribution information is generated.

[0020] In this scheme, connected component analysis is performed on the distribution data of the reinforcing bars inside the pole to extract the geometric centerline, diameter fluctuation, and reflection signal continuity index. Defects in the reinforcing bar layer are determined by combining the preset diameter fluctuation range and preset reflection signal continuity index threshold. The concrete matrix data is analyzed using a signal amplitude feature statistical algorithm to obtain the signal amplitude uniformity index and average amplitude value. Based on the preset signal amplitude uniformity index threshold and preset average amplitude value threshold, concrete layer defects are identified, and corresponding defect distribution information is generated. This refines the defect judgment dimensions and achieves accurate identification of layered defects through targeted index analysis, reducing missed and false judgments.

[0021] Furthermore, in step S7, the loss function of the pole quality comprehensive analysis model during the training process... The mathematical expression is: In the formula, Indicates the number of training samples. Indicates the first The true quality score labels for each training sample This indicates that the comprehensive analysis model of pole mass is applicable to the first... The prediction quality score for each training sample. Represents the regularization coefficient. This represents the total number of layers in the neural network. Indicates the first The weight matrix of a layered neural network, This represents the L2 norm.

[0022] In this scheme, a loss function is constructed that includes the difference between the predicted quality score and the actual quality score label of the training samples, the regularization coefficient, and the relevant terms of the weight matrix of each layer of the neural network. This effectively constrains the model parameters during the training process of the comprehensive analysis model for pole quality, reduces overfitting, and improves the accuracy and stability of the model's prediction of pole quality scores. This provides reliable model support for the comprehensive evaluation of pole quality and ensures the credibility of the quality analysis results. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a method for detecting the quality of utility poles according to an embodiment of the present invention. Detailed Implementation

[0024] The following detailed description illustrates the specific implementation method: like Figure 1 As shown, it is a flowchart illustrating a pole quality inspection method according to an embodiment of the present invention, including the following steps: S1. The pole to be inspected is scanned using a 3D scanning device to obtain 3D point cloud data of the pole. The initial image set of the entire pole is converted to grayscale according to the image grayscale processing algorithm to obtain a grayscale image set of the entire pole. The surface feature points of the grayscale image set of the entire pole are extracted to obtain a set of coordinates of surface feature points of the pole. S2. Compare the coordinate set of feature points on the pole surface with the preset standard pole feature point coordinate set point by point to obtain the coordinate offset of each feature point. S3. The coordinate offset of each feature point is compared with the preset coordinate offset threshold to obtain the set of deformation feature points whose coordinate offset exceeds the preset coordinate offset threshold. The set of deformation feature points is then aggregated to obtain the distribution information of the deformation area on the pole surface. S4. The spatial range of the deformation area distribution information on the surface of the pole is analyzed by the detection range positioning algorithm to obtain the three-dimensional coordinate range of the detection target area inside the pole. The internal structure of the pole part corresponding to the three-dimensional coordinate range is scanned by the ultrasonic detection device to obtain the original scanning data of the internal structure of the pole. S5. Noise filtering is performed on the original scanning data of the internal structure of the pole to obtain the noise-reduced data of the internal structure of the pole. The noise-reduced data of the internal structure of the pole is then processed by the layer segmentation algorithm to obtain the data of the distribution of steel bars and the concrete matrix inside the pole. S6. Extract defect features from the internal steel reinforcement distribution data and concrete matrix data of the pole respectively to obtain the defect distribution information of the steel reinforcement layer and the concrete layer. S7. Input the distribution information of the deformation area on the pole surface, the distribution information of the defects in the steel reinforcement layer, and the distribution information of the defects in the concrete layer into the pole quality comprehensive analysis model for multi-dimensional feature fusion and quality status assessment to obtain the target pole quality detection result. The pole quality comprehensive analysis model is obtained by supervised training using a historical dataset containing pole surface deformation features, internal structural defect data, and corresponding quality rating labels, with the training objective being to minimize the model prediction error.

[0025] In this embodiment, the specific implementation process of scanning the pole under inspection with a 3D scanning device to obtain the 3D point cloud data of the pole is as follows: First, a RIEGL VZ-400i terrestrial 3D laser scanner is used as the 3D scanning device. This terrestrial 3D laser scanner works by emitting laser pulses and receiving signals reflected back from the surface of the pole under inspection. The 3D scanning device is first set up at multiple preset station locations around the pole to ensure complete coverage of the entire pole surface. At each station, the 3D scanning device emits a laser beam to perform a high-speed rotating scan of the pole surface, calculates distance information by measuring the laser reflection time, and records the horizontal and vertical angles to form the original scan data for that station. Next, using the multi-site cloud registration function in the 3D scanning device's built-in software, the point cloud data collected from each station is unified into a global coordinate system by setting a common target sphere as a registration reference between adjacent stations. Then, the registered point cloud data is fused to remove noise points caused by environmental interference, and the environmental point cloud outside the pole body is cropped. Finally, continuous and complete 3D point cloud data of the utility poles is generated through point cloud density optimization and surface reconstruction algorithms. Specifically, the number of preset site locations is determined based on the height, perimeter, environmental occlusion of the pole to be inspected, and the effective scanning distance and field of view of the 3D scanning equipment.

[0026] The specific implementation process for extracting surface feature points from a global grayscale image set of utility poles to obtain a set of coordinates for surface feature points is as follows: The SIFT feature detection algorithm is used for surface feature point extraction. The global grayscale image set of the utility poles is processed using the SIFT feature detection algorithm. For each image set, a Gaussian scale space is constructed using the SIFT feature detection algorithm. Local extrema are detected in this Gaussian scale space using the difference of Gaussians operator as a preliminary candidate set of surface feature points. This preliminary candidate set is then refined. Low-contrast surface feature points are eliminated by fitting a three-dimensional quadratic function, and edge-response surface feature points are removed using the eigenvalue ratio of the Hessian matrix. For the remaining stable surface feature points, the gradient direction histogram of the stable surface feature points is calculated to determine the principal direction, generating a 128-dimensional descriptor vector with scale and rotation invariance. After extracting the two-dimensional surface feature points, the two-dimensional surface feature points of the same physical point in multiple grayscale images of the pole are matched and corresponded using camera calibration parameters and multi-view geometric constraints. The three-dimensional spatial coordinates of the two-dimensional surface feature point in the coordinate system corresponding to the three-dimensional point cloud data of the pole are calculated using the triangulation principle. All the three-dimensional coordinate points reconstructed through this process constitute the final set of coordinates of the surface feature points of the pole.

[0027] Specifically, in S1, the mathematical expression for the image grayscale processing algorithm is: In the formula, Represents coordinates in the image The grayscale value of a pixel. Represents coordinates in the image The red channel pixel value of the pixel. Represents coordinates in the image The green channel pixel value of the pixel. Represents coordinates in the image The blue channel pixel value of the pixel. Indicates the use of adjusting coordinates in the image The red channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Green channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Blue channel pixel value of a pixel The weighting coefficients, .

[0028] Specifically, S2 includes the following steps: S21. Obtain a preset standard pole 3D model, and extract the preset standard pole feature point coordinate set from the preset standard pole 3D model according to the pole surface feature point coordinate set; S22. Establish the nearest neighbor correspondence between the set of feature point coordinates on the pole surface and the preset standard pole feature point coordinate set, and calculate the coordinate offset of each pair of feature points with established correspondence in the nearest neighbor correspondence. The mathematical expression for the coordinate offset is: In the formula, This represents the coordinate offset of the i-th feature point. This represents the three-dimensional coordinates of the i-th feature point in the set of feature point coordinates on the pole surface. This represents the three-dimensional coordinates of the i-th feature point in the preset set of feature point coordinates for standard utility poles.

[0029] In this embodiment, the process of obtaining a preset standard pole 3D model first requires constructing an idealized digital pole model that conforms to standard dimensions and shape using 3D modeling software, based on the pole's design drawings or technical specifications. This preset standard pole 3D model is a 3D digital representation that defines a standard geometric structure and contains complete surface point cloud data. When determining the preset standard pole 3D model, a corresponding, deformation-free, and defect-free reference 3D model needs to be selected or created based on the pole's model, specifications, and other parameters. It is also necessary to ensure that the preset standard pole 3D model has pre-calculated and stored the 3D coordinates of feature points at key locations on its surface, forming a preset standard pole feature point coordinate set. The specific implementation method for generating the preset standard pole feature point coordinate set by extracting the feature point coordinate set from the preset standard pole 3D model is as follows: each feature point in the actual collected feature point coordinate set of the pole surface is spatially matched with the surface point cloud data corresponding to the preset standard pole 3D model. By finding the point in the surface point cloud of the preset standard pole 3D model that is closest in spatial position to each feature point in the feature point coordinate set of the pole surface, the coordinates of these corresponding points are extracted from the preset standard pole 3D model, and finally assembled to generate the preset standard pole feature point coordinate set.

[0030] The specific process for establishing the nearest neighbor correspondence between the set of feature point coordinates on the pole surface and the preset standard pole feature point coordinate set is as follows: First, each feature point in the set of feature point coordinates on the pole surface is used as a query point, and a traversal search is performed in the preset standard pole feature point coordinate set. For each feature point in the set of feature point coordinates on the pole surface, the Euclidean distance between it and each feature point in the preset standard pole feature point coordinate set is calculated. Then, for this feature point in the set of feature point coordinates on the pole surface, the feature point in the preset standard pole feature point coordinate set with the smallest Euclidean distance is selected, and these two feature points are established as a pair. This process is repeated until each feature point in the set of feature point coordinates on the pole surface finds its unique nearest neighbor feature point in the preset standard pole feature point coordinate set, thus systematically establishing a one-to-one nearest neighbor correspondence between the two sets. During this process, the coordinate offset of each pair of feature points with established correspondence in the nearest neighbor correspondence is calculated.

[0031] Specifically, S3 includes the following steps: S31. Obtain the coordinate offset of each feature point, and compare the coordinate offset of each feature point with the preset coordinate offset threshold one by one. Mark the feature points whose coordinate offset exceeds the preset coordinate offset threshold to form an initial set of deformation feature points. S32. Based on the cylindrical structural features of the pole, a spherical spatial neighborhood search range is constructed. Then, a neighborhood search is performed on each feature point in the initial set of deformation feature points according to the spherical spatial neighborhood search range to obtain other deformation feature points within the spherical spatial neighborhood search range. Finally, the aggregation degree between any two deformation feature points within the spherical spatial neighborhood search range is calculated to obtain the aggregation degree matrix between feature points. The mathematical expression for the aggregation degree is: In the formula, This represents the degree of aggregation between the i-th deformation feature point and the j-th deformation feature point. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point along the x-axis in the world coordinate system. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point in the y-axis direction of the world coordinate system. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point in the z-axis direction of the world coordinate system; S33. Based on the aggregation degree matrix between feature points, deformable feature points that meet the preset aggregation conditions are grouped into the same deformable region, and the boundary coordinates, area and contour morphology information of each deformable region are extracted and integrated to obtain the distribution information of deformable regions on the pole surface.

[0032] In this embodiment, in step S31, the coordinate offset of each feature point is obtained. This coordinate offset is typically calculated by comparing the three-dimensional coordinates of the same feature point at two different time points or under different conditions, and is represented as the displacement vector of each feature point in three-dimensional space. Then, the magnitude of the coordinate offset of each feature point (i.e., the Euclidean distance) is compared one by one with a preset coordinate offset threshold. In practice, all feature points need to be traversed, and the coordinate offset of each feature point needs to be calculated one by one. If the coordinate offset of a feature point exceeds the preset coordinate offset threshold, the feature point is marked as a potential deformation point. Finally, all marked potential deformation points are summarized to form an initial set of deformation feature points. Specifically, the value of the preset coordinate offset threshold needs to be determined comprehensively based on the actual structure of the pole, material properties, monitoring accuracy requirements, and environmental factors. Typically, the preset coordinate offset threshold is set based on the allowable range of normal deformation of the pole or historical monitoring data. For example, 0.5% to 1% of the pole diameter can be used as a reference value. It can be set by analyzing the statistical distribution of the coordinate offset of the pole in a state without significant damage (such as the mean plus three times the standard deviation) or by combining engineering experience (such as the limit value of pole deformation in the power industry standard).

[0033] Constructing a spherical spatial neighborhood search range based on the cylindrical structural features of the utility pole involves: The surface of the utility pole can be approximated as a regular cylindrical surface. To accommodate the three-dimensional curved surface features of the pole and ensure the spatial uniformity of the neighborhood search, a spherical spatial neighborhood search range needs to be constructed. Specifically, the radius of the spherical spatial neighborhood search range is first determined. This radius should be set based on the diameter and surface curvature of the utility pole, typically a specific proportion of the pole's radius (e.g., 0.1 to 0.3 times) to ensure that the spherical spatial neighborhood search range can cover local areas of the pole's surface without being too large or too small. Then, using the feature points in each initial set of deformation feature points as the center, a spherical spatial region is constructed with a set radius. This spherical spatial region is the spherical spatial neighborhood search range. During the construction process, the cylindrical structural features of the utility pole must be considered to ensure that the spherical spatial neighborhood search range approximates a sphere locally on the pole's surface, thereby effectively capturing neighboring deformation feature points and providing a spatial basis for aggregation degree calculation.

[0034] Based on the spherical spatial neighborhood search range, a neighborhood search is performed on each feature point in the initial set of deformable feature points to obtain other deformable feature points within the spherical spatial neighborhood search range. Specifically, after constructing the spherical spatial neighborhood search range, a neighborhood search is performed on each feature point in the initial set of deformable feature points. In practice, with the current feature point as the center and the radius of the corresponding spherical spatial neighborhood search range as the boundary, all other feature points located within this spherical spatial neighborhood search range are searched in three-dimensional space. First, the Euclidean distance between the current feature point and all other feature points in the initial set of deformable feature points is calculated. Then, feature points whose distance is less than or equal to the radius of the spherical spatial neighborhood search range are selected; these feature points are the other deformable feature points within the spherical spatial neighborhood search range. This process requires traversing each feature point in the initial set of deformable feature points, sequentially performing the above distance calculation and selection operations, thereby establishing a list of deformable feature points within the spherical spatial neighborhood search range for each feature point, providing neighborhood data for subsequent calculation of the aggregation degree matrix between feature points.

[0035] Based on the aggregation degree matrix between feature points, deformation feature points that meet the preset aggregation conditions are grouped into the same deformation region. The boundary coordinates, area, and contour morphology information of each deformation region are extracted, and the distribution information of the deformation region on the pole surface is integrated to obtain the following: First, based on the aggregation degree matrix between feature points, where the matrix elements... This represents the degree of aggregation between the i-th deformation feature point and the j-th deformation feature point. The preset aggregation condition is typically set as an aggregation degree threshold, for example... A value greater than 0.5 (the specific value can be adjusted according to the actual application) indicates that the two deformation feature points have a high spatial correlation. Then, a clustering algorithm (such as density-based clustering or connected component analysis) is used to traverse the aggregation degree matrix between feature points, and deformation feature points whose aggregation degree meets the preset aggregation conditions are grouped into the same group to form a deformation region. For each deformation region, the boundary coordinates of each deformation region are extracted: by calculating the minimum and maximum three-dimensional coordinates of all deformation feature points in the deformation region, the bounding box or convex hull boundary of the deformation region is determined. The area of ​​the region is obtained by projecting the deformation feature points onto the local plane of the pole surface or by directly estimating based on the three-dimensional point cloud density. The contour morphology information is described by analyzing the distribution characteristics of the edge points of the deformation region (such as smoothness and curvature). Finally, the boundary coordinates, area, and contour morphology information of all deformation regions are integrated to form the distribution information of deformation regions on the pole surface, including comprehensive data such as the number, location, size, and shape of the deformation regions.

[0036] Specifically, S4 includes the following steps: S41. Extract the three-dimensional coordinate boundaries of each deformation region from the distribution information of deformation regions on the surface of the pole, and input the three-dimensional coordinate boundaries of each deformation region into the detection range positioning algorithm. Based on the corresponding mapping relationship between the surface of the pole and the internal structure, perform spatial coordinate expansion calculation to obtain the three-dimensional coordinate range of the detection target area inside the pole. S42. Based on the three-dimensional coordinate range of the target area inside the pole, adjust the detection parameters of the ultrasonic detection device to obtain the adjusted ultrasonic detection device, and scan the pole part corresponding to the three-dimensional coordinate range area by area based on the adjusted ultrasonic detection device. S43. Receive the reflected and transmitted signals returned by the adjusted ultrasonic detection device during the scanning process, preprocess the reflected and transmitted signals to obtain the original scanning data of the internal structure of the pole.

[0037] In this embodiment, the detection range localization algorithm employs a spatial coordinate expansion algorithm. This algorithm is based on the three-dimensional coordinate boundaries of each deformation region in the distribution information of deformation regions on the pole surface. It calculates the spatial coordinate expansion through the corresponding mapping relationship between the pole surface and its internal structure. First, the vertex data of the three-dimensional coordinate boundaries of each deformation region are extracted, and the spatial range of the deformation region on the pole surface is determined by fitting a three-dimensional geometric model. Then, according to the corresponding mapping relationship between the pole surface and its internal structure (usually based on mapping rules established using a standard pole structural model or a historical damage database), the deformation region on the pole surface is extended inward along the normal direction. Considering the thickness of the pole material and the layering characteristics of the internal structure (such as concrete protective layer, steel reinforcement layer, core, etc.), the three-dimensional spatial coordinates are expanded radially and axially. During the expansion process, the surface boundary points are mapped to the internal structural layers through a coordinate transformation matrix, and the expansion distance is corrected by combining a safety factor (such as damage diffusion margin). Finally, the three-dimensional coordinate range of the detection target area inside the pole is output.

[0038] The ultrasonic detection device employs a multi-channel array ultrasonic flaw detector, which consists of an ultrasonic transducer array, a signal generator, a receiver, a data acquisition module, and a motion control module. The ultrasonic transducer array comprises multiple piezoelectric ceramic probes arranged at specific intervals, capable of emitting high-frequency ultrasonic pulses and receiving reflected and transmitted signals. The signal generator generates electrical pulses to drive the transducers, the receiver amplifies and filters the returned ultrasonic signals, and the data acquisition module converts analog signals into digital signals. The motion control module adjusts the spatial position and orientation of the transducer array based on the three-dimensional coordinates of the target area within the pole, using a robotic arm or guide rail system to achieve area-by-area scanning. The multi-channel array ultrasonic flaw detector supports dynamic parameter adjustment, including transmission frequency, pulse width, gain, and depth of focus, to adapt to the detection needs of different materials (such as concrete and reinforcing steel) and defect types.

[0039] In step S41, firstly, the three-dimensional coordinate boundaries of each deformation region are extracted from the distribution information of deformation regions on the pole surface. Three-dimensional point cloud processing techniques (such as boundary extraction algorithms) are used to obtain the contour vertex coordinates of the deformation regions. The contour vertex coordinates are then input into a detection range localization algorithm. This algorithm, based on a predefined mapping relationship between the pole surface and its internal structure (such as a mapping function from surface coordinates to internal reinforcing steel layers or crack propagation zones), determines the potential range of internal damage through spatial coordinate expansion calculations. The expansion calculations include offsetting the surface boundary points radially inward along the pole (the offset amount is based on material thickness and the damage model) and extending them axially to cover potentially continuous defect areas. Finally, a three-dimensional coordinate range of the detection target area inside the pole is generated, represented by a three-dimensional bounding box or mesh.

[0040] In step S42, based on the three-dimensional coordinate range of the target area inside the pole, the scanning path and resolution requirements are calculated, and the detection parameters of the ultrasonic detection device are adjusted by the control unit. The parameter adjustment of the ultrasonic detection device includes setting the ultrasonic emission frequency (high frequency for shallow, fine detection, low frequency for deep penetration), the pulse repetition frequency to adapt to the scanning speed, and the focusing rule and gain value of the transducer array. After adjustment, the motion control module drives the transducer array of the ultrasonic detection device to position itself at the starting point on the pole surface corresponding to the three-dimensional coordinate range, and performs a region-by-region scan according to the preset grid path to ensure coverage of the entire target area inside the pole.

[0041] In step S43, during the scanning process, the receiver of the ultrasonic detection device acquires reflected and transmitted signals in real time, and the signals are transmitted to the data processing unit for preprocessing. Preprocessing includes signal noise reduction (such as wavelet filtering to remove environmental noise), time-domain gain compensation to correct signal attenuation, and signal alignment and normalization. The preprocessed data is marked and integrated according to the scanning area location to generate original scanning data of the pole's internal structure. This data includes the ultrasonic amplitude, time delay, and spectral information for each scanning point.

[0042] Specifically, S5 includes the following steps: S51. An adaptive filtering algorithm based on wavelet transform is used to filter noise from the original scanning data of the internal structure of the pole, resulting in noise-reduced data of the internal structure of the pole. S52. Construct the segmentation threshold conditions for the hierarchical segmentation algorithm, and divide the noise reduction data of the internal structure of the pole according to the segmentation threshold conditions to obtain several structural layer data. S53. Perform acoustic feature extraction and material attribute identification on the data of each structural layer to obtain the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material. Perform coordinate calibration and format regularization on the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material to obtain the steel reinforcement distribution data and concrete matrix data inside the pole.

[0043] In this embodiment, in step S51, the adaptive filtering algorithm specifically employs an adaptive filtering algorithm based on wavelet transform. The implementation process is as follows: First, the original scan data of the pole's internal structure is used as the input signal, and an appropriate wavelet basis function and decomposition level are selected for wavelet transform, decomposing the original scan data into approximation coefficients and detail coefficients at different scales. Then, by analyzing the statistical characteristics of the detail coefficients at each level, such as variance or thresholds related to the noise model, the filtering threshold for each level is adaptively calculated. Next, a soft-thresholding or hard-thresholding function is used to process the detail coefficients, treating coefficients below the soft-threshold or hard-threshold as noise and attenuating or setting them to zero, while retaining coefficients above the soft-threshold or hard-threshold that represent the true structure. Finally, the processed approximation coefficients and detail coefficients are used to perform an inverse wavelet transform to reconstruct the signal, thereby outputting the denoised data of the pole's internal structure. The entire process, through adaptive threshold adjustment of the wavelet coefficients, effectively separates and suppresses noise while preserving structural edge and detail information.

[0044] In step S52, the hierarchical segmentation algorithm specifically employs a segmentation threshold-based algorithm. First, based on the physical characteristics (such as density and continuous changes in acoustic impedance) and statistical features of the noise reduction data of the pole's internal structure, a clear segmentation threshold is constructed. This threshold typically involves metrics such as signal strength, gradient change, or regional homogeneity, and their critical values. Then, using the spatial coordinate sequence of the noise reduction data of the pole's internal structure as the processing object, a continuous scan is performed starting from one end, calculating its feature values ​​point-by-point or region-by-region and comparing them with the segmentation threshold. When the data features satisfy the boundary criteria defined by the segmentation threshold, it is determined to be a boundary point of a structural layer. Based on all boundary points, the continuous noise reduction data of the pole's internal structure is divided into multiple subsets corresponding to different physical structural layers; these subsets constitute several structural layer data. This process achieves automated identification and data segmentation of the vertical hierarchical structure within the pole.

[0045] In step S53, firstly, acoustic feature extraction and material attribute identification are performed on each segment of structural layer data. Acoustic feature extraction includes calculating feature parameters of the structural layer data, such as sound wave propagation speed, attenuation coefficient, and spectral response. Material attribute identification is accomplished by matching and comparing the extracted acoustic features with preset steel reinforcement material feature libraries and concrete matrix material feature libraries to determine the main material category corresponding to each structural layer data. Next, the structural layer data corresponding to steel reinforcement material and the structural layer data corresponding to concrete matrix material are processed separately. Coordinate calibration is performed on the structural layer data corresponding to steel reinforcement material and the structural layer data corresponding to concrete matrix material, respectively, that is, according to the spatial positioning parameters of the scanning equipment, the relative coordinates in the data are converted into coordinates in a unified absolute coordinate system. Then, the format is standardized to obtain the steel reinforcement distribution data and concrete matrix data inside the pole, respectively.

[0046] Specifically, S6 includes the following steps: S61. Perform connected component analysis on the distribution data of the steel bars inside the pole, identify the set of reflection points belonging to the same steel bar, calculate the geometric center line, diameter fluctuation and reflection signal continuity index of each connected component, and when the diameter fluctuation exceeds the preset diameter fluctuation range or the reflection signal continuity index is lower than the preset reflection signal continuity index threshold, determine that there is a steel bar layer defect in the connected component and generate steel bar layer defect distribution information. S62. The concrete matrix data is analyzed by a signal amplitude feature statistical algorithm to obtain the signal amplitude uniformity index and average amplitude value of each analysis region. When the signal amplitude uniformity index of the analysis region is lower than the preset signal amplitude uniformity index threshold and the average amplitude value is lower than the preset average amplitude value threshold, it is determined that there is a concrete layer defect in the analysis region, and concrete layer defect distribution information is generated.

[0047] In this embodiment, the preset diameter fluctuation range is typically determined based on statistical analysis of a large amount of data on the distribution of reinforcing bars inside standard, defect-free poles. First, a large sequence of diameter data for complete connected rebar domains is extracted from historical inspection data or standard samples. The standard deviation or range of the diameter of each connected domain is calculated as its diameter fluctuation value. Then, the distribution range of diameter fluctuation values ​​for all qualified connected domains is statistically analyzed (e.g., the mean plus several times the standard deviation is taken as the upper limit). This distribution range of diameter fluctuation values ​​for all qualified connected domains is determined as the preset diameter fluctuation range. The preset reflection signal continuity index threshold is determined based on theoretical modeling and experimental calibration of the reflection signal characteristics of ideal continuous rebar. The reflection signal continuity index typically quantifies the coherence of the spatial distribution of reflection points within a connected domain or the stability of the signal strength. By analyzing data on known, well-functioning connected rebar domains, the reflection signal continuity index of the connected rebar domain data is calculated. Combined with engineering experience (such as allowable small signal gaps), a minimum acceptable limit is set, which is the preset reflection signal continuity index threshold. The preset threshold value for signal amplitude uniformity is determined based on the definition of signal characteristics in a uniform and dense concrete matrix region. Signal amplitude uniformity indices (such as the coefficient of variation of amplitude in a local area) reflect the homogeneity of the concrete's internal material. By analyzing a large number of confirmed defect-free concrete matrix data areas, the signal amplitude uniformity index for each region is calculated, and the typical distribution of the signal amplitude uniformity index for each region is statistically analyzed. The lower limit of the typical distribution or a lower percentile (such as the 5th percentile) is set as the preset threshold value for signal amplitude uniformity. The preset threshold value for average amplitude depends on the establishment of a signal strength benchmark for a healthy concrete matrix. The average amplitude value is related to the density, strength, and internal interface condition of the concrete. By collecting and analyzing scan data from standard-cured or confirmed intact concrete pole areas, the average level of signal amplitude in these areas is statistically analyzed. Considering normal material fluctuations and system noise, the preset threshold value for average amplitude is usually set as the lower boundary value of the statistical distribution of the average amplitude value in healthy areas (such as the mean minus several times the standard deviation).

[0048] In step S61, the distribution data of the internal steel reinforcement of the pole is input. This data is a set of points containing spatial coordinates and reflection intensity information. Subsequently, connected component analysis is performed on the internal steel reinforcement distribution data. Typically, based on a three-dimensional spatial proximity criterion (e.g., Euclidean distance less than a set threshold), spatially close reflection points are clustered into multiple independent point sets. Each point set is identified as a set of reflection points belonging to the same steel reinforcement, i.e., a connected component. Next, for each connected component, the geometric centerline (obtained through principal component analysis or curve fitting of the spatial point set), diameter fluctuation (calculated and evaluated by analyzing the distribution range of point sets perpendicular to the centerline at each cross-section), and reflection signal continuity index (e.g., a measure of the coherence of reflection point density or signal intensity along the centerline direction) are calculated. Then, the calculated diameter fluctuation for each connected component is compared with a preset diameter fluctuation range, and the calculated reflection signal continuity index for each connected component is compared with a preset reflection signal continuity index threshold. When the diameter fluctuation of a connected component exceeds a preset diameter fluctuation range, or the continuity index of the reflected signal of the connected component is lower than a preset reflection signal continuity index threshold, it is determined that the rebar corresponding to the connected component has a rebar layer defect. Finally, the location, type, and severity information of all connected components determined to have defects are integrated to generate structured rebar layer defect distribution information.

[0049] In step S62, firstly, concrete matrix data is input. Next, the concrete matrix data is analyzed regionally using a signal amplitude feature statistical algorithm. This typically involves dividing the concrete matrix data into regular voxel grids or irregular but characteristically consistent analysis units in three-dimensional space. For each analysis region, the signal amplitude uniformity index (e.g., the ratio of the standard deviation to the mean of the amplitude values ​​of all data points within the region, i.e., the coefficient of variation) and the average amplitude value (the arithmetic mean of the amplitude values ​​of all data points within the region) are calculated. Then, the signal amplitude uniformity index of each analysis region is compared with a preset signal amplitude uniformity index threshold, and the average amplitude value of each analysis region is also compared with a preset average amplitude value threshold. When the signal amplitude uniformity index of an analysis region is lower than the preset signal amplitude uniformity index threshold, and the average amplitude value of the analysis region is also lower than the preset average amplitude value threshold, it is determined that a concrete layer defect exists in that analysis region. Finally, the spatial range and feature information of all analysis regions that meet the above joint judgment conditions are summarized to generate concrete layer defect distribution information.

[0050] Specifically, in S7, the loss function of the pole quality comprehensive analysis model during the training process. The mathematical expression is: In the formula, Indicates the number of training samples. Indicates the first The true quality score labels for each training sample This indicates that the comprehensive analysis model of pole mass is applicable to the first... The prediction quality score for each training sample. Represents the regularization coefficient. This represents the total number of layers in the neural network. Indicates the first The weight matrix of a layered neural network, This represents the L2 norm.

[0051] In this embodiment, the comprehensive quality analysis model for utility poles is trained under supervision using a feedforward neural network. The specific numerical structure of this feedforward neural network is as follows: the number of input layer nodes is determined by the total dimension of three types of feature vectors: the distribution information of deformation areas on the pole surface, the distribution information of defects in the reinforcing steel layer, and the distribution information of defects in the concrete layer. For example, the distribution information of deformation areas on the pole surface can extract the number of deformation areas (1D), the average depth of deformation areas (1D), the maximum area of ​​deformation areas (1D), the total area of ​​deformation areas (1D), and the spatial centroid coordinates of deformation areas (3D), totaling 7 dimensions of features; the distribution information of defects in the reinforcing steel layer can extract the total length of steel cracks (1D), the spatial density of defect points (1D), the volume ratio of rusted areas (1D), and the direction of the main crack (2D), totaling 5 dimensions of features; the distribution information of defects in the concrete layer can extract the total volume of concrete voids (1D), the average width of the matrix crack network (1D), the strength attenuation coefficient (1D), and the spatial dispersion of defect areas (1D), totaling 4 dimensions of features. Therefore, the total number of input layer nodes is 16. The feedforward neural network consists of two fully connected hidden layers: the first hidden layer contains 32 neurons and uses the ReLU activation function; the second hidden layer contains 16 neurons and also uses the ReLU activation function. The output layer has one neuron and uses a linear activation function to output the predicted quality score of the target pole by the comprehensive pole quality analysis model. During training, a historical dataset containing pole surface deformation features, internal structural defect data, and corresponding quality rating labels is used. The predicted quality score is calculated using forward propagation, and the loss function L (where L is the regularization coefficient) is applied. (Default value is 0.01) Calculate the sum of prediction error and weight regularization term, and iteratively optimize the weight matrix of all neural networks from the input layer to the output layer using the backpropagation algorithm. This continues until the model converges. A well-trained model is defined as follows: on the reserved validation set, the value of the loss function L decreases by less than a preset threshold of 0.001 over 50 consecutive training epochs; the mean absolute error between the model's predicted quality score and the actual quality score is less than a preset threshold of 0.3 points; and the model's coefficient of determination on the independent test set... Reaching or exceeding the preset value of 0.9.

[0052] The specific implementation process of multi-dimensional feature fusion and quality status assessment includes: inputting the distribution information of deformation areas on the pole surface, the distribution information of defects in the reinforcing steel layer, and the distribution information of defects in the concrete layer into the comprehensive quality analysis model of the pole for multi-dimensional feature fusion and quality status assessment. The specific process is as follows: First, the distribution information of deformation areas on the pole surface, the distribution information of defects in the reinforcing steel layer, and the distribution information of defects in the concrete layer are preprocessed by feature vectorization and standardization. The distribution information of deformation areas on the pole surface needs to be transformed into a 7-dimensional normalized feature vector containing the number of deformation areas, the average depth of deformation areas, the maximum area of ​​deformation areas, the total area of ​​deformation areas, and the spatial centroid coordinates of deformation areas. The distribution information of defects in the reinforcing steel layer needs to be transformed into a 5-dimensional normalized feature vector containing the total length of steel cracks, the spatial density of defect points, the volume ratio of rusted areas, and the direction of the main crack. The distribution information of defects in the concrete layer needs to be transformed into a 4-dimensional normalized feature vector containing the total volume of concrete voids, the average width of the matrix crack network, the strength attenuation coefficient, and the spatial dispersion of defect areas. Subsequently, the three normalized feature vectors are concatenated sequentially to form a 16-dimensional comprehensive feature vector, which serves as the input data for the input layer of the pole quality comprehensive analysis model. This comprehensive feature vector undergoes nonlinear transformation and feature abstraction through the first hidden layer (32 neurons) and the second hidden layer (16 neurons) of the pole quality comprehensive analysis model. During this process, the features of the pole surface deformation area distribution, the features of the steel reinforcement layer defect distribution, and the features of the concrete layer defect distribution are automatically interacted and fused. Finally, the output layer of the pole quality comprehensive analysis model maps this fused high-level feature representation into a continuous pole quality prediction score. Based on a preset quality score threshold (e.g., a score above 8.0 is excellent, 6.0 to 8.0 is good, and below 6.0 requires repair), this prediction score is converted into the final target pole quality inspection result, which comprehensively reflects the structural condition characterized by the pole surface deformation area distribution, steel reinforcement layer defect distribution, and concrete layer defect distribution.

[0053] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for inspecting the quality of utility poles, characterized in that: Includes the following steps: S1. The pole to be inspected is scanned using a 3D scanning device to obtain 3D point cloud data of the pole. The initial image set of the entire pole is converted to grayscale according to the image grayscale processing algorithm to obtain a grayscale image set of the entire pole. The surface feature points of the grayscale image set of the entire pole are extracted to obtain a set of coordinates of surface feature points of the pole. S2. Compare the coordinate set of feature points on the pole surface with the preset standard pole feature point coordinate set point by point to obtain the coordinate offset of each feature point. S3. The coordinate offset of each feature point is compared with the preset coordinate offset threshold to obtain the set of deformation feature points whose coordinate offset exceeds the preset coordinate offset threshold. The set of deformation feature points is then aggregated to obtain the distribution information of the deformation area on the pole surface. S4. The spatial range of the deformation area distribution information on the surface of the pole is analyzed by the detection range positioning algorithm to obtain the three-dimensional coordinate range of the detection target area inside the pole. The internal structure of the pole part corresponding to the three-dimensional coordinate range is scanned by the ultrasonic detection device to obtain the original scanning data of the internal structure of the pole. S5. Noise filtering is performed on the original scanning data of the internal structure of the pole to obtain the noise-reduced data of the internal structure of the pole. The noise-reduced data of the internal structure of the pole is then processed by the layer segmentation algorithm to obtain the data of the distribution of steel bars and the concrete matrix inside the pole. S6. Extract defect features from the internal steel reinforcement distribution data and concrete matrix data of the pole respectively to obtain the defect distribution information of the steel reinforcement layer and the concrete layer. S7. Input the distribution information of the deformation area on the pole surface, the distribution information of the defects in the steel reinforcement layer, and the distribution information of the defects in the concrete layer into the pole quality comprehensive analysis model for multi-dimensional feature fusion and quality status assessment to obtain the target pole quality detection result. The pole quality comprehensive analysis model is obtained by supervised training using a historical dataset containing pole surface deformation features, internal structural defect data, and corresponding quality rating labels, with the training objective being to minimize the model prediction error.

2. The method for detecting the quality of utility poles according to claim 1, characterized in that: In S1, the mathematical expression for the image grayscale processing algorithm is: In the formula, Represents coordinates in the image The grayscale value of a pixel. Represents coordinates in the image The red channel pixel value of the pixel. Represents coordinates in the image The green channel pixel value of the pixel. Represents coordinates in the image The blue channel pixel value of the pixel. Indicates the use of adjusting coordinates in the image The red channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Green channel pixel value of a pixel The weighting coefficients, Indicates the use of adjusting coordinates in the image Blue channel pixel value of a pixel The weighting coefficients, .

3. The method for detecting the quality of utility poles according to claim 1, characterized in that: S2 includes the following steps: S21. Obtain a preset standard pole 3D model, and extract the preset standard pole feature point coordinate set from the preset standard pole 3D model according to the pole surface feature point coordinate set; S22. Establish the nearest neighbor correspondence between the set of feature point coordinates on the pole surface and the preset standard pole feature point coordinate set, and calculate the coordinate offset of each pair of feature points with established correspondence in the nearest neighbor correspondence. The mathematical expression for the coordinate offset is: In the formula, This represents the coordinate offset of the i-th feature point. This represents the three-dimensional coordinates of the i-th feature point in the set of feature point coordinates on the pole surface. This represents the three-dimensional coordinates of the i-th feature point in the preset set of feature point coordinates for standard utility poles.

4. The method for detecting the quality of utility poles according to claim 1, characterized in that: S3 includes the following steps: S31. Obtain the coordinate offset of each feature point, and compare the coordinate offset of each feature point with the preset coordinate offset threshold one by one. Mark the feature points whose coordinate offset exceeds the preset coordinate offset threshold to form an initial set of deformation feature points. S32. Based on the cylindrical structural features of the pole, a spherical spatial neighborhood search range is constructed. Then, a neighborhood search is performed on each feature point in the initial set of deformation feature points according to the spherical spatial neighborhood search range to obtain other deformation feature points within the spherical spatial neighborhood search range. Finally, the aggregation degree between any two deformation feature points within the spherical spatial neighborhood search range is calculated to obtain the aggregation degree matrix between feature points. The mathematical expression for the aggregation degree is: In the formula, This represents the degree of aggregation between the i-th deformation feature point and the j-th deformation feature point. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point along the x-axis in the world coordinate system. This represents the coordinate difference between the i-th and j-th deformation feature points along the y-axis in the world coordinate system. This represents the coordinate difference between the i-th deformation feature point and the j-th deformation feature point in the z-axis direction of the world coordinate system; S33. Based on the aggregation degree matrix between feature points, deformable feature points that meet the preset aggregation conditions are grouped into the same deformable region, and the boundary coordinates, area and contour morphology information of each deformable region are extracted and integrated to obtain the distribution information of deformable regions on the pole surface.

5. The method for detecting the quality of utility poles according to claim 1, characterized in that: S4 includes the following steps: S41. Extract the three-dimensional coordinate boundaries of each deformation region from the distribution information of deformation regions on the surface of the pole, and input the three-dimensional coordinate boundaries of each deformation region into the detection range positioning algorithm. Based on the corresponding mapping relationship between the surface of the pole and the internal structure, perform spatial coordinate expansion calculation to obtain the three-dimensional coordinate range of the detection target area inside the pole. S42. Based on the three-dimensional coordinate range of the target area inside the pole, adjust the detection parameters of the ultrasonic detection device to obtain the adjusted ultrasonic detection device, and scan the pole part corresponding to the three-dimensional coordinate range area by area based on the adjusted ultrasonic detection device. S43. Receive the reflected and transmitted signals returned by the adjusted ultrasonic detection device during the scanning process, preprocess the reflected and transmitted signals to obtain the original scanning data of the internal structure of the pole.

6. The method for detecting the quality of utility poles according to claim 1, characterized in that: S5 includes the following steps: S51. An adaptive filtering algorithm based on wavelet transform is used to filter noise from the original scanning data of the internal structure of the pole, resulting in noise-reduced data of the internal structure of the pole. S52. Construct the segmentation threshold conditions for the hierarchical segmentation algorithm, and divide the noise reduction data of the internal structure of the pole according to the segmentation threshold conditions to obtain several structural layer data. S53. Perform acoustic feature extraction and material attribute identification on the data of each structural layer to obtain the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material. Perform coordinate calibration and format regularization on the structural layer data corresponding to the steel reinforcement material and the structural layer data corresponding to the concrete matrix material to obtain the steel reinforcement distribution data and concrete matrix data inside the pole.

7. The method for detecting the quality of utility poles according to claim 1, characterized in that: S6 includes the following steps: S61. Perform connected component analysis on the distribution data of the steel bars inside the pole, identify the set of reflection points belonging to the same steel bar, calculate the geometric center line, diameter fluctuation and reflection signal continuity index of each connected component, and when the diameter fluctuation exceeds the preset diameter fluctuation range or the reflection signal continuity index is lower than the preset reflection signal continuity index threshold, determine that there is a steel bar layer defect in the connected component and generate steel bar layer defect distribution information. S62. The concrete matrix data is analyzed by a signal amplitude feature statistical algorithm to obtain the signal amplitude uniformity index and average amplitude value of each analysis region. When the signal amplitude uniformity index of the analysis region is lower than the preset signal amplitude uniformity index threshold and the average amplitude value is lower than the preset average amplitude value threshold, it is determined that there is a concrete layer defect in the analysis region, and concrete layer defect distribution information is generated.

8. The method for detecting the quality of utility poles according to claim 1, characterized in that: In S7, the loss function of the pole quality comprehensive analysis model during the training process. The mathematical expression is: In the formula, Indicates the number of training samples. Indicates the first The true quality score labels for each training sample This indicates that the comprehensive analysis model of pole mass is applicable to the first... The prediction quality score for each training sample. Represents the regularization coefficient. This represents the total number of layers in the neural network. Indicates the first The weight matrix of a layered neural network, This represents the L2 norm.

Images