A method for detecting surface unevenness defects of a high-voltage cable joint

Abstract

The present application belongs to the technical field of non-destructive testing of cable, and particularly relates to a method for detecting uneven defects on the surface of high-voltage cable joint, which acquires three-dimensional point cloud data by reconstructing the laser line projected on the surface of the cable; in order to realize accurate surface curve fitting, the noise in the curve point cloud is removed by the gradient value change in the point neighborhood; the influence of the defect area on the fitting model is reduced by rejecting the points with large deviation values for multiple times, so that the accurate equation expression of the cable surface contour line is obtained; then the residual value of the high-voltage cable surface point relative to the fitting curve is compared to mark the points of the defect area, and finally the detection and extraction of the concave and convex defects on the intermediate joint of the cable are realized through point cloud segmentation and clustering; the present application can effectively detect the uneven defects of the high-voltage cable surface polishing, and is more economical, simple and efficient compared with the high-voltage cable defect detection method based on partial discharge signal.

Description

Technical Field

[0001] This invention belongs to the field of cable non-destructive testing technology, and particularly relates to a method for detecting surface unevenness defects in high-voltage cable joints. Background Technology

[0002] Currently, the power industry mainly relies on manual methods such as visual inspection and measurement to detect surface defects in cable joints. This method suffers from low efficiency and inconsistent standards. Although the partial discharge signal method can detect defects in high-voltage cables by detecting abnormal fluctuations in electromagnetic fields, light, heat, and electrical signals around the cable, it requires applying high voltage to the cable, which accelerates the aging of the cable edges. Furthermore, the partial discharge signal detection method is an online, live detection method that cannot be performed before cable laying, resulting in complex procedures and high costs. Summary of the Invention

[0003] In order to solve the technical problems existing in the prior art, the present invention provides a method for detecting uneven surface defects of high voltage cable joints, which aims to solve the problems of complex detection procedures and high detection costs.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A method for detecting surface unevenness defects in high-voltage cable joints includes the following steps:

[0006] Step 1: Project a laser onto the surface of the cable, capture images of the laser line, and obtain three-dimensional point cloud data of the cable surface based on the captured images;

[0007] Step 2: Based on the obtained 3D point cloud data of the cable surface, a nearest neighbor search is performed using a Kd-Tree;

[0008] Step 3: Map the coordinates of the point cloud in the 3D point cloud data and calculate point p. i With adjacent point p i-1 p i+1 The gradient value projected onto the straight line is compared with a preset threshold to determine point p. i Is it noise?

[0009] Step 4: Iteratively fit the noise-removed 3D point cloud data to accurately fit and model the cable surface curve, and obtain the curve model of the cable surface.

[0010] Step 5: Based on the fitted modeled curves, extract the defect points on each curve.

[0011] Preferably, step 1 includes the following steps:

[0012] projecting laser to the surface of the cable, the laser forms a laser curve with the surface of the cable, and a camera captures the laser curve;

[0013] extracting the laser center line in the captured laser curve image, and obtaining the laser three-dimensional point cloud profile of the surface of the high-voltage cable according to the following formula:

[0014] F(x, y, z, ) = 0

[0015] A = λQ;

[0016] In the formula, F is the light plane equation obtained by calibration, A is the vector [x c y c ] T representing the point in the camera coordinate system, Q represents the vector [u v T representing the point in the image coordinate system, and the coefficient λ is wherein z c is the depth value of the reconstructed three-dimensional point, and f is the focal length of the camera;

[0017] moving the camera and the laser projector to form a plurality of laser line three-dimensional profiles, and then performing three-dimensional reconstruction of the entire cable based on three-dimensional point cloud splicing.

[0018] Preferably, the step 2 comprises the following steps:

[0019] Step 2.1: Two data points in the three-dimensional point cloud data are located in two-dimensional space, and the variances of the data points in the X and Y axis directions are calculated respectively and

[0020]

[0021]

[0022] In the formula, x n represents the value of the nth data point on the X axis; y n represents the value of the nth data point on the Y axis; represents the average value of all data points on the X axis; represents the average value of all data points on the Y axis;

[0023] Step 2.2: Establishing a Kd-Tree, and finding the nearest node from the search node based on the established Kd-Tree.

[0024] Preferably, the step 2.2 comprises the following steps:

[0025] Based on the established Kd-Tree computing search path other node space point and the search point example, the distance is less than the value of the nearest node to the search point, if other child control exists the nearest node, then jump to the subspace, repeat the above search process until the nearest point is found, the specific calculation formula is as follows:

[0026]

[0027] Preferably, the step 3 comprises the following steps:

[0028] Step 3.1: the point cloud P in the point cloud data is sorted according to the value of X axis;

[0029] Step 3.2: the first point and the last point on the X axis constitute a straight line l;

[0030] Step 3.3: according to the order, the distance of the point p i to the straight line l is calculated, recorded as d i ;

[0031] Step 3.4: the gradient value Δp i of the projection of the point p i-1 , p i+1 to the straight line l is calculated; i-1 i+1

[0032] Step 3.5: if ||Δp i-1 ·Δp i+1 ||≥α, the point is a noise point, wherein α is a set threshold value.

[0033] Preferably, the iterative residual fitting in the step 4 comprises the following steps:

[0034] According to the light plane equation, the plane normal vector n

[0035] Based on the normal vector n and the Z axis unit vector e , the rotation matrix R is determined, and the point cloud P is rotated by the rotation matrix R to obtain the point cloud P':

[0036] P'=R.P;

[0037] The z axis value of all points in the point cloud P' is equal, and the least square fitting is performed on the z axis value of all points in the point cloud P';

[0038] Therefore, the points in P' are obtained by the following formula:

[0039] F(s,t)=s·t=ax 2 ​​+ bxy + cy 2 + dx + ey + f = 0;

[0040] In the formula: coefficient vector s = [a, b, c, d, e, f] T ; parameter vector t = [x 2 , xy, y 2 , x, y, 1] T ; F(s, t) is the elliptical expression of the current point cloud;

[0041] The coefficient s is identified by least square method:

[0042] E = ||sT|| 2 = 0;

[0043] In the formula: T = [t1, t2,..., t n ] T is determined by the point p i，i＝1，2，...，n in the point cloud P';

[0044] When the fitted points cannot form a complete ellipse, the parameter s is constrained, and when b 2 -4ac = 1 ensures that F(s, t) is always an ellipse;

[0045] The residual value of the point after the first fitting is r1, the points with residual value less than the threshold value a are selected to form a new point cloud P", the point cloud P" is least square fitted and the residual value r2 of the point in P' after the second fitting is calculated;

[0046] The offset between the second fitting and the first fitting is represented by o 12 = ||r1-r2||, and the above operation is repeated until o i-1，i < β, and the curve model of the cable surface is obtained.

[0047] The curve elliptical model formed by the laser projection to the cable surface is identified, for a smooth cable surface, the reconstructed points have good linearity and have small residual value relative to the fitted curve; correspondingly, the curve of the defect such as the concave, convex and scratch area will be deformed, and the residual value of the corresponding points relative to the curve is usually large, and the prerequisite for calculating the residual value of each point of the curve is the parametric equation of the straight curve, so the key to detecting the defect is to accurately fit the curve model of the cable surface.

[0048] Preferably, the step 5 comprises the following steps:

[0049] Based on the curve model, the points with residual absolute value greater than the residual screening threshold value are screened by setting the residual screening threshold value, and the defect points are obtained;

[0050] Using Kd-Tree nearest neighbor search, the average distance of all nearest neighbors is calculated, and points that are outside the threshold of the average distance are removed from the data;

[0051] The extracted defect points are clustered into nearest neighbors. The distance from the defect point to the plane is compared with the constructed threshold. If the distance is less than the constructed threshold, the nearest neighbor cluster of the defect point is obtained. This enables the extraction of uneven defect areas such as depressions and bulges on the cable joint.

[0052] The beneficial effects of this invention include:

[0053] This invention acquires three-dimensional point cloud data by reconstructing the laser lines projected onto the cable surface. To achieve accurate surface curve fitting, noise in the curve point cloud is removed by varying the gradient values ​​within the neighborhood of each point. Points with large deviations are repeatedly removed to reduce the impact of defect areas on the fitting model, thus obtaining an accurate equation representing the cable surface contour. Then, the residual values ​​of the high-voltage cable surface points relative to the fitted curve are compared to mark the defect areas. Finally, point cloud segmentation and clustering are used to detect and extract concave and convex defects on the cable joints. This invention can effectively detect unevenness on the surface of high-voltage cables, and is more economical, simple, and efficient than high-voltage cable defect detection methods based on partial discharge signals. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the steps in this invention.

[0055] Figure 2 This is a schematic diagram illustrating the principle of line structured light reconstruction in this invention.

[0056] Figure 3 A physical structural diagram of a real high-voltage cable connector.

[0057] Figure 4 This is a schematic diagram of a single-sided three-dimensional reconstruction process based on a physical cable joint, according to a specific embodiment of the present invention. (a) Acquiring laser line images of the high-voltage cable surface; b) Extracting the center of the laser line; (c) Reconstructing the point cloud outline from the laser line; (d) Reconstructing the point cloud from a single side.

[0058] Figure 5 The following is a line structured light noise diagram of a specific embodiment of the present invention: (a) noise outside the laser centerline; (b) noise of three-dimensional point cloud imaging.

[0059] Figure 6 This is a two-dimensional data point distribution map in this invention.

[0060] Figure 7 This is a diagram of the Kd-Tree model established in this invention.

[0061] Figure 8 A single-line point cloud neighborhood noise distribution diagram for one specific embodiment of the present application.

[0062] Figure 9 An iterative fitting flowchart for the present application.

[0063] Figure 10 An iteration number diagram for the present application. DETAILED DESCRIPTION

[0064] To make the purposes, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0065] The specific embodiments of the present application will be described below with reference to the accompanying drawings. Figure 1 - The accompanying drawings Figure 10 The present application will be further described in detail:

[0066] Referring to the accompanying drawings Figure 1 to the accompanying drawings Figure 3 As shown in the drawings, a high-voltage cable joint surface uneven defect detection method comprises the following steps:

[0067] Step 1: Constructing cable surface three-dimensional point cloud data;

[0068] The laser is projected onto the calibration plate, and the camera photographs the laser line on the calibration plate. The camera is calibrated for internal and external parameters by combining Zhang's calibration method, and the laser plane and camera pose are calibrated.

[0069] The laser is projected onto the cable surface to form a bright and regular laser curve. The camera photographs the laser line, and the collected high-voltage cable surface laser line is as shown in Figure 4 (a);

[0070] The laser center line in the laser line image is as shown in Figure 4 (b), and the laser line three-dimensional point cloud profile of the high-voltage cable surface is reconstructed according to the following formula (as shown in Figure 4 (c)):

[0071] F(x, y, z, ) = 0

[0072] A = λQ;

[0073] Wherein: F is the light plane equation obtained by calibration, A is the vector [x c y c ] T representing the point in the camera coordinate system, Q represents the vector [u v] T representing the point in the image coordinate system, and the coefficient λ is wherein z c is the depth value of the reconstructed three-dimensional point, and f is the focal length of the camera;

[0074] The moving camera and the laser projector form a plurality of laser lines three-dimensional profile, and finally realize the three-dimensional reconstruction of the whole cable through three-dimensional point cloud splicing, as shown in Figure 4 (d).

[0075] Step 2: Based on the obtained cable surface three-dimensional point cloud data, Kd-Tree is used for nearest point search;

[0076] In the acquisition of three-dimensional data, the point cloud data collected will contain noise due to the interference of environmental light and camera sensitivity; As shown in Figure 5 (a), when the points outside the light center are extracted, noise points will be formed around the curve point cloud, as shown in Figure 5 (b) ;

[0077] According to the collected three-dimensional point cloud data, Kd-Tree is established, which is as follows:

[0078] The Kd-Tree has the property of binary search tree; When the right subtree is not empty, the value of the root node on the right subtree is less than that of the child node; When the left subtree is not empty, the value of the root node on the left subtree is greater than that of the child node; The Kd-Tree needs to select the dimension k with the maximum variance in the K-dimensional data set, and divide the data set in the dimension to obtain two subsets and store them in the root node; The two subsets are divided in the same way until they cannot be divided.

[0079] Suppose there are n two-dimensional data points {(x1, y1), (x2, y2),..., (x n , y n )} in two-dimensional space, as shown in Figure 6 ; Calculate the variance of the data in X and Y axis directions and

[0080]

[0081]

[0082] Specifically, when Select the median value x in the X-axis direction i , with point (x i y i Let be the root node, and draw a straight line x = x perpendicular to the X-axis. i The plane is divided into two spaces, left and right; as follows: Figure 7 As shown, the above process is repeated for the left and right subspaces, with the left median x respectively. k And right median x l The plane is divided and recursively divided. When the final path contains only one leaf node, the Kd-Tree is completed.

[0083] After the Kd-Tree is built, given a search node, find the node closest to the search node. By building the Kd-Tree, the search node is compared with the root node of the Kd-Tree. If the value of the search node is greater than the root node, then the root node of the right subtree is compared. If the value of the search node is less than the value of the root node, then the root node of the left subtree is compared.

[0084] If finding the neighboring node does not yet determine if it is the nearest neighbor, the distance between the search point and other points in the search path needs to be calculated. If the calculated distance is less than the value from the neighboring node to the search point, a nearest neighbor exists in another subspace. The search is then performed in that subspace, and the search process is repeated until the nearest neighbor is found. The formula for calculating the proximity distance is shown below:

[0085]

[0086] Step 3: Map the coordinates of the point cloud in the 3D point cloud data and calculate point p. i With adjacent point p i-1 p i+1 The gradient value projected onto the straight line is compared with a preset threshold to determine point p. i Is it noise?

[0087] See Figure 8 As shown, the sparsity of the single-line point cloud means that radius filtering and statistical filtering will cause feature loss. The noise generated by the reconstruction will be on the same plane as the curve, with a small number and random distribution. The gradient value of the noise points in the Y direction in the neighborhood of the single-line point cloud is much larger than that of the point cloud in the normal region and the point cloud in the defect region.

[0088] Non-noise points and adjacent points show relatively small fluctuations in the Y-axis direction, while outliers appear inside or outside the main point, resulting in significant fluctuations in the Y-axis direction. The directional position of the wire point cloud is uncertain, and judging it based on changes in a single X and Y direction is inaccurate. The coordinates of the point cloud are mapped, and the changes in the mapped values ​​enable the judgment of the noise points.

[0089] Specifically, the rotating point cloud P-plane is parallel to the XOY plane;

[0090] Sort the point cloud P in order of the X value;

[0091] Construct a straight line l from the first point and the last point;

[0092] Calculate point p in sequence. i The distance to line l is denoted as d. i ;

[0093] Calculate point p i With adjacent point p i-1 p i+1 The gradient value Δp after projection onto the line i-1 , Δp i+1 ;

[0094] Satisfy ||Δp i-1 ·Δp i+1 Points with ||≥α are considered noise points, where α is a set threshold.

[0095] The Y-direction gradient value of the noise point in the neighborhood is much larger than that of the normal point and the defective area point.

[0096] Step 4: Iteratively fit the noise-removed 3D point cloud data to accurately fit and model the cable surface curve, and obtain the curve model of the cable surface.

[0097] The curve formed by the laser projecting onto the cable surface can be identified using an ellipse model; the points reconstructed from the smooth cable surface have good linearity; the residual of the fitted curve is relatively small; conversely, the curve in the defect area will be deformed, and the residual value of the curve is large.

[0098] See appendix Figure 9 As shown, the specific calculation steps are as follows:

[0099] The plane normal vector is obtained from the aforementioned light plane equation.

[0100] Based on normal vector and Z-axis unit vector Determine the rotation matrix R, and rotate the point cloud P using the rotation matrix R to obtain the point cloud P′:

[0101] P' = R.P;

[0102] The z-axis value of all points in the point cloud P' is equal, and the z-axis value can be ignored in the fitting calculation; the least square fitting is performed on the z-axis value of all points in the point cloud P';

[0103] Therefore, the points in P' are obtained by the following formula:

[0104] F(s, t) = s · t = ax 2 +bxy+cy 2 +dx+ey+f = 0;

[0105] In the formula: the coefficient vector s = [a, b, c, d, e, f] T ; the parameter vector t = [x 2 , xy, y 2 , x, y, 1] T ; F(s, t) is an elliptical expression of the current point cloud;

[0106] The coefficient s is identified by the least square method:

[0107] E = ||sT|| 2 = 0;

[0108] In the formula: T = [t1, t2,..., tn] n ] T is determined by the points p i，i＝1，2，...，n in the point cloud P';

[0109] When the fitted points cannot form a complete ellipse, the parameter s is constrained, and when b 2 -4ac = 1 guarantees that F(s, t) is always an ellipse;

[0110] The residual value of all points after the first fitting is r1, the points with a residual value less than a threshold value a are selected to form a new point cloud P", the point cloud P" is least square fitted and the residual value r2 of the points in P' after the second fitting is calculated;

[0111] The offset between the second and first fittings is represented by o 12 = ||r1-r2||, and the above operation is repeated until o i-1，i < β, and a curve model of the cable surface is obtained.

[0112] Specific verification examples are as follows. When the iteration termination offset β is set to 0.005 mm, the curve point cloud containing defects with a point number of 526 is fitted and stopped after the fifth iteration is completed, as shown in Figure 10The maximum residual value increased from 1.395 mm to 1.636 mm, the minimum residual value was less than 0.005 mm, and the mean value of the point residual decreased from 0.177 mm to 0.108 mm. This shows that during the iteration process, the normal points gradually approach the fitted curve, and the defect points gradually deviate from the fitted curve, so that the defect area points are more prominent and accurate in the residual distribution, proving the effectiveness of the fitting method.

[0113] Step 5: Based on the curve after fitting modeling, the defect points on each curve are extracted.

[0114] The fitting result can obtain the deviation degree (residual value) of all points on the curve. The cable defect point is determined by calculating the residual value;

[0115] The spatial filtering is used to remove outliers, the Kd-Tree is used to search for adjacent points, the average distance of all adjacent points is calculated, and the points whose average distance is outside the threshold value can be defined as outliers and removed from the data.

[0116] The defect points are clustered by the nearest neighbor, the distance of the defect points to the plane is calculated and compared with the threshold value, the defect points with a distance less than the threshold value are put into the class, and the nearest neighbor clustering of the defect points is obtained.

[0117] The spatial distance is used to segment the clustering result to obtain the defect area, and the clustering of the defect points is segmented according to the different spatial distances of different clusters to the plane. The defect class is judged according to the distance less than the threshold value, and then segmented to obtain the extraction of the concave and convex uneven defect area on the cable intermediate joint.

[0118] The above embodiments only express the specific implementation of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the protection scope of the present application. It should be noted that for ordinary skilled in the art, without departing from the technical concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A method for detecting surface unevenness defects in high-voltage cable joints, characterized in that, Includes the following steps: Step 1: Project a laser onto the surface of the cable, capture images of the laser line, and obtain three-dimensional point cloud data of the cable surface based on the captured images; Step 2: Based on the obtained 3D point cloud data of the cable surface, a nearest neighbor search is performed using a Kd-Tree; Step 3: Map the coordinates of the point cloud in the 3D point cloud data and calculate point p. i With adjacent point p i-1 ,p i+1 The gradient value projected onto the straight line is compared with a preset threshold to determine point p. i Is it noise? Step 4: Iteratively fit the noise-removed 3D point cloud data to accurately fit and model the cable surface curve, and obtain the curve model of the cable surface. Step 5: Based on the fitted and modeled curves, extract the defect points on each curve; Step 1 includes the following steps: A laser is projected onto the surface of a cable, and the laser and the cable surface form a laser curve, which is then captured by a camera. Extract the laser centerline from the captured laser curve image, and then obtain the 3D point cloud contour of the laser on the surface of the high-voltage cable according to the following formula: F(x,y,z)=0 A = λQ; In the formula: F is the equation of the light plane obtained from calibration, A is a point in the camera coordinate system, Q represents a point in the image coordinate system, and the coefficient λ is... Where z c This represents the depth value of the reconstructed 3D points, where f is the camera focal length. The camera and laser projector are moved to form a three-dimensional outline of multiple laser lines, and then the entire cable is reconstructed in three dimensions based on the stitching of the three-dimensional point cloud. Step 3 includes the following steps: Step 3.1: Sort the point cloud P in the point cloud data according to the size of its value on the X-axis; Step 3.2: Construct a straight line l from the first point and the last point on the X-axis; Step 3.3: Calculate point p sequentially. i The distance to the line l is denoted as d. i ; Step 3.4: Calculate the point p i With adjacent point p i-1 ,p i+1 The gradient value Δp projected onto the line l i-1 ,Δp i+1 ; Step 3.5: If ||Δp i-1 ·Δp i+1 Points with a threshold value ≥ α are considered noise points, where α is a set threshold value.

2. The method for detecting surface unevenness defects in high-voltage cable joints according to claim 1, characterized in that, Step 2 includes the following steps: Step 2.1: In the three-dimensional point cloud data, two data points are located in two-dimensional space. Calculate the variance of the data points in the X and Y axis directions respectively. and In the formula: x n This represents the value of the nth data point on the X-axis; y n This represents the value of the nth data point on the Y-axis; This represents the average of all data points on the X-axis; This represents the average value of all data points along the Y-axis. Step 2.2: Build a Kd-Tree and find the node closest to the search node based on the built Kd-Tree.

3. The method for detecting surface unevenness defects in high-voltage cable joints according to claim 2, characterized in that, Step 2.2 includes the following steps: Based on the established Kd-Tree, the search path is used to calculate the distance between the midpoint of other nodes in the search space and the search node. If the distance is less than the distance from the nearest node to the search node, and if a nearest node exists in another subspace, the search jumps to that subspace and repeats the above search process until the nearest point is found. The specific calculation formula is as follows:

4. The method for detecting surface unevenness defects in high-voltage cable joints according to claim 3, characterized in that, The iterative residual fitting in step 4 includes the following steps: The plane normal vector is obtained from the equation of the light plane. Based on normal vector and Z-axis unit vector Determine the rotation matrix R, and rotate the point cloud P using the rotation matrix R to obtain the point cloud P. ′ : P ′ =R·P; The point cloud P ′ All points have the same Z-axis value for the point cloud P. ′ Least-squares fitting is performed on the Z-axis values ​​of all points; Therefore P ′ The points in the middle are obtained using the following formula: F(s,t)=s·t=ax 2 +bxy+cy 2 +dx+ey+f=0; In the formula: the coefficient vector s = [a, b, c, d, e, f] T ; Parameter vector t = [x 2 ,xy,y 2 [x,y,1] T F(s,t) is the elliptic expression of the current point cloud; The coefficients s are identified using the least squares method: E=‖sT‖ 2 =0; In the formula: T=[t1,t2,…,t n ] T It is made of point cloud P ′ point p in i Let i = 1, 2, ..., n; When the fitted points cannot form a complete ellipse, the coefficients s are constrained; when b 2 -4ac = 1 ensures that F(s,t) is always an ellipse; The residual value of a point after the first fitting is r1. Points with residual values ​​less than the threshold α are selected to form a new point cloud P. ″ For point cloud P ″ Perform least squares fitting and calculate P. ′ The point in the second fitting is at the residual value r2; via o 12 =‖r1-r2‖ represents the offset between the two fitting operations. Repeat the above operation until o. i-1,i When <β, a curve model of the cable surface is obtained, where β is the iteration termination offset.

5. The method for detecting surface unevenness defects in high-voltage cable joints according to claim 1, characterized in that, Step 5 includes the following steps: Based on the curve model, points with residual absolute values ​​greater than the residual screening threshold are selected by setting a residual screening threshold, thus obtaining defect points; Using Kd-Tree nearest neighbor search, the average distance of all nearest neighbors is calculated, and points that are outside the threshold of the average distance are removed from the data; The extracted defect points are clustered into nearest neighbors. The distance from the defect point to the plane is compared with a preset threshold. If the distance is less than the preset threshold, the nearest neighbor cluster of the defect point is obtained. This enables the extraction of uneven defect areas such as dents and bulges on the cable joint.

Images