Three-dimensional imaging detection method for surface defects of connector

Abstract

The invention relates to the technical field of three-dimensional imaging, in particular to a three-dimensional imaging detection method for surface defects of a connector, and provides the following scheme: collecting a point cloud image of the connector; extracting a suspected disturbance region through a morphological operator to generate a target point; constructing an axial symmetry mapping relation of the target points, and generating a symmetric residual distribution map; based on the disturbance space gradient field, eliminating a disturbance convergent point and reserving a disturbance divergent point; and further extracting a normal abrupt change relation and a curvature minimum value region through a defect identification network, and outputting a three-dimensional coordinate set of a defect region. According to the method, the artifact elimination accuracy and the defect identification precision are effectively improved, and the method is suitable for a high-robustness three-dimensional defect detection task under a complex inner wall structure.

Description

Technical Field

[0001] The present application relates to the field of three-dimensional imaging technology, and in particular to a three-dimensional imaging detection method for surface defects of connectors. Background Art

[0002] With the widespread application of micro-connectors in consumer electronics, automotive electronics, and medical devices, the manufacturing quality of their surface structures has become an increasingly critical factor affecting the reliability and safety of the entire device. Minor defects such as scratches, dents, deplating, and bubbles on the connector surface can cause poor contact, electrical faults, or even mechanical failure during subsequent plugging, unplugging, connection, or sealing processes. Therefore, high-precision and robust detection of connector surface defects has become a pressing technical challenge.

[0003] For example, the Chinese patent application with publication number CN115953356A discloses a defect identification method for battery swap connectors, which includes the following steps: Step 1: Defect location, using an image acquisition system to collect and analyze the connector and locate the defect position; Step 2: Defect classification, using DCNN to classify defects through the convolution layer; Step 3: Defect analysis, using CNN to extract defect features and analyze sample defects. This application method is used to detect whether the vehicle socket and battery pack plug have abnormalities and defects before and after connection, and perform image-based automatic detection of specific connector sockets and plug objects. The detection object supports user template addition and learning capabilities to continuously improve the adaptability of the system, timely detect abnormalities and defects, greatly improve the timeliness and efficiency of maintenance, and reduce the economic losses of users.

[0004] For another example, a Chinese patent with authorization announcement number CN113808131B discloses a connector defect identification method, system, device and medium, which relates to the field of smart industry, including: obtaining a first image of a connector to be inspected; extracting a first region of interest corresponding to the shell of the connector to be inspected from the first image; performing threshold segmentation processing on the first region of interest to obtain a first binary image, performing contour detection on the first binary image to obtain a first contour detection result, and calculating the contour circumference and contour area based on the first contour detection result. If the contour circumference is greater than the circumference threshold and the contour area is greater than the area threshold, it is determined that the shell of the connector to be inspected has a detachment defect, thereby realizing automatic detection of connector detachment defects.

[0005] The above existing technologies all have the problems raised by this background technology: since the connector cavity structure usually has mirror symmetry, it is easily affected by multipath reflection when collecting point clouds inside the structure, resulting in a large number of mirror artifact points. These artifact points are very similar to real defects in morphology and can easily cause misjudgment. In order to solve the above problems, this application designs a three-dimensional imaging detection method for surface defects of connectors. Summary of the Invention

[0006] The technical problem to be solved by this application is to address the shortcomings of the existing technology and provide a three-dimensional imaging detection method for connector surface defects, which collects connector point cloud images; extracts suspected disturbance areas through morphological operators to generate target points; constructs an axisymmetric mapping relationship of the target points to generate a symmetric residual distribution map; based on the disturbance space gradient field, eliminates disturbance convergence points and retains disturbance divergence points; further extracts normal mutation relationships and curvature minimum areas through a defect recognition network, and outputs a three-dimensional coordinate set of the defect area.

[0007] To achieve the above objectives, this application provides the following technical solutions:

[0008] A three-dimensional imaging detection method for surface defects of a connector, the method comprising:

[0009] According to the collected point cloud data, a surface map in the depth direction is obtained;

[0010] Taking the central axis of the surface graph as the symmetry axis, identifying the target points through morphological operators, calculating the spatial residual between each target point and its axisymmetric position point, and obtaining a symmetric residual distribution map;

[0011] In the symmetric residual distribution map, an artifact point removal strategy is executed to obtain a connector point cloud image, wherein the artifact point removal strategy determines whether there are artifact points based on the disturbance trend of the target point, and removes the target point with the disturbance trend converging;

[0012] The connector point cloud image is input into a preset defect recognition network, the connector point cloud image is processed by the defect recognition network, the defect area is output, and the defect area is three-dimensionally imaged, wherein the defect recognition network is trained by historical defect data of the connector.

[0013] Taking the central axis of the surface graph as the axis of symmetry, the target point is identified by using a morphological operator, including:

[0014] Determine the normal projection direction of the point cloud according to the symmetry axis and the distribution of the point cloud in the surface diagram, and calculate the projection signal curve according to the normal projection direction and the spatial density of the point cloud;

[0015] Performing morphological opening and closing operations on the projection signal curve to obtain a morphological reference curve;

[0016] The projection signal curve and the morphological reference curve are compared for residuals, and point clouds with residual amplitudes greater than a disturbance judgment threshold are screened to generate target points.

[0017] Performing morphological opening and closing operations on the projection signal curve to obtain a morphological reference curve includes:

[0018] generating a nonlinear structural element according to the morphology of the surface image and the density variation of the point cloud in the normal projection direction, wherein the nonlinear structural element is calculated according to the periodic characteristics of the density variation;

[0019] Using the nonlinear structural element as a sliding window, performing an erosion operation and a dilation operation on the projection signal curve, wherein the erosion operation is performed according to a local minimum value within the sliding window, and the dilation operation is performed according to a local maximum value within the sliding window;

[0020] The results of the erosion operation and the dilation operation are fused into a smooth envelope curve, which serves as a morphological reference curve of the projection signal curve.

[0021] Calculate the spatial residual between each target point and its axisymmetric position point, including:

[0022] Obtain the axisymmetric mapping point of each target point;

[0023] The spatial residual, normal residual and curvature residual between the target point and the axisymmetric mapping point are calculated.

[0024] In the symmetric residual distribution map, an artifact point removal strategy is executed to obtain a connector point cloud image, including:

[0025] Constructing a disturbance spatial gradient field according to the symmetric residual distribution map, and calculating the disturbance intensity spatial gradient of each target point in the map;

[0026] Dividing the target point into a disturbance convergence point and a disturbance divergence point according to the gradient direction of the disturbance intensity spatial gradient;

[0027] The disturbance convergence points are eliminated, and a connector point cloud image is generated based on the point clouds corresponding to the remaining disturbance divergence points.

[0028] The construction of the surface graph includes:

[0029] Projecting the point cloud data into a polar coordinate space with the center of the connector as a reference axis;

[0030] Based on the projected point cloud data, multiple equidistant sub-regions are divided in the polar direction, and the point set that forms the local extreme point density in the axial direction in each sub-region is extracted;

[0031] The density peak of the point set is used as the reference layer boundary, and a fitting surface with normal continuity constraints is constructed according to the normal change rate of the point set neighborhood to generate a surface graph.

[0032] The defect recognition network includes a point set geometry encoding layer, a spatial disturbance extraction layer, and a defect determination layer, wherein:

[0033] The point set geometry encoding layer is used to receive the connector point cloud image and construct a geometric feature vector based on the spatial coordinates, normal vector and local curvature of each point cloud;

[0034] The spatial perturbation extraction layer is provided with a graph convolutional network for calculating normal mutation relations and local curvature minimum regions in a spatial adjacency graph constructed by geometric feature vectors;

[0035] The defect determination layer is used to output a position label of the defect area and a corresponding three-dimensional coordinate set thereof according to the normal mutation relationship and the local curvature minimum area.

[0036] The spatial perturbation extraction layer includes an adjacency graph construction module, a graph convolution calculation module, and a differentiation module, wherein:

[0037] The adjacency graph construction module is used to generate a spatial adjacency graph with Euclidean distance as edge weight based on the spatial coordinates and geometric feature vector of each point;

[0038] The graph convolution calculation module uses a weighted graph convolution network to propagate and aggregate information on the nodes in the spatial adjacency graph, and extract the normal change rate and curvature gradient of each point in its neighborhood;

[0039] The distinguishing module determines whether the normal change rate of each point in its neighborhood is greater than the weighted standard deviation of the mean normal change rate of the neighborhood based on the normal change rate and curvature gradient of the point in its neighborhood, so as to identify the normal mutation relationship, and identify the local curvature minimum area when the curvature gradient is less than a preset first threshold and the curvature value is the minimum value in its neighborhood.

[0040] The defect determination layer includes:

[0041] Performing local connectivity analysis on the normal mutation relationship and the curvature minimum value region to generate a candidate defect point set;

[0042] Clustering the candidate defect point set according to a clustering algorithm to obtain a clustering result;

[0043] Calculating the defect confidence of the point set area corresponding to the clustering result according to the spatial distribution density, curvature gradient variation range and normal consistency of the point cloud in the clustering result;

[0044] The point set region where the defect confidence is greater than the defect threshold is output as a defect region.

[0045] Performing three-dimensional imaging on the defect area includes:

[0046] Extracting a three-dimensional point cloud coordinate set corresponding to the defect area, and constructing a spatial voxel grid based on the three-dimensional point cloud coordinate set;

[0047] Interpolation fitting is performed on the spatial voxel grid to generate a three-dimensional image of the defect.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] This application achieves efficient pre-screening of suspected defect areas by constructing a projection signal curve based on normal perturbations and introducing morphological operators to extract abnormal disturbance points; further, a symmetric residual distribution map is constructed through axisymmetric mapping, and combined with the directional analysis of the perturbation gradient field, the accurate removal of mirror artifact points is achieved; on this basis, the spatial adjacency graph and graph structure convolution unit are combined to perform multi-dimensional geometric analysis of the candidate areas, ultimately achieving accurate identification of the real defect areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0051] Figure 1 This is a schematic diagram of an exemplary application scenario of an embodiment of the present application;

[0052] Figure 2 This is a schematic diagram of artifact imaging in an embodiment of the present application;

[0053] Figure 3 This is a flow chart of a method for three-dimensional imaging detection of surface defects of a connector according to an embodiment of the present application;

[0054] Figure 4 This is a schematic diagram of the target point recognition process in an embodiment of the present application;

[0055] Figure 5 This is a schematic diagram of the principle of generating the morphological reference curve according to an embodiment of the present application;

[0056] Figure 6 A schematic diagram of axisymmetric mapping point generation for an embodiment of the present application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0058] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It will be understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] This application is applicable to connector products with cavity-type mirror-symmetrical structures and strong surface reflective properties. The 3D defect detection process is limited by the external viewing angle acquisition method of the imaging device. There are artifact points near the defect area caused by structural reflection. Application scenarios include but are not limited to:

[0060] High-speed signal connectors with multiple rows of metal pins and micro deep cavity holes;

[0061] The pins inside the circular connector are arranged symmetrically around the circular axis. During the imaging process, high reflection angle areas are prone to secondary or multiple reflection interference;

[0062] The selection of application scenarios is determined based on the common characteristics of connectors. The selected application scenarios must have at least one of the following characteristics:

[0063] The defect is located in a deep cavity, slit, inverted cone or other imaging blind area structure;

[0064] The surface has specular reflection characteristics, and multipath reflection easily forms symmetrical artifacts;

[0065] Due to the limited external acquisition angle, the point cloud distribution is unilaterally tilted or sparse;

[0066] Defects are non-structural geometric mutations that only appear as minimal perturbations.

[0067] It should be pointed out that the three-dimensional imaging detection method for connector surface defects proposed in this application does not rely on the clear structural partitioning or fixed cavity division of the target connector as a prerequisite, but is aimed at imaging scenarios with typical mirror reflection path constraints and viewing angle occlusion limitations.

[0068] It is worth noting that the artifact removal strategy described in this application is not designed specifically for a certain structural module or a specific connector series, but can also be applied to:

[0069] In engineering imaging wiring, there is a possibility of multiple optical path reversals;

[0070] There are scenarios where path dependencies are projected or received by default within the structural space;

[0071] In the model space, the same defect behavior can be mirrored or perturbations can be propagated in the point cloud.

[0072] See also Figure 1 , which is a schematic diagram of an exemplary application scenario provided in an embodiment of the present application.

[0073] like Figure 1 As shown, the method of the present application is applicable to defect detection of the pinholes of the connector. The connector generally includes a shell main body structure and a plurality of pinhole array units. The pinhole array units are evenly arranged around the central symmetry axis and present a deep cavity columnar shape, which is a typical axisymmetric mirror structure.

[0074] Figure 1 It is shown that the imaging system includes three parts: a signal source, a connector and an imaging device. The signal source in this application can be a structured light projection device, a laser scanner or a controllable light source with modulation and coding capabilities, which is projected onto the connector structure to be detected.

[0075] Figure 1 It shows that during the detection process, the signal emitted by the signal source is irradiated onto the inner wall of the connector. After being reflected by the surface, the reflected signal image is collected by the imaging device, and a three-dimensional point cloud image of the connector is generated based on it.

[0076] It is understandable that the imaging device in this application refers to a depth imaging system with the ability to obtain three-dimensional structural information, which is used to extract the spatial position point set of the connector surface from the structured light or laser reflection image.

[0077] Furthermore, the imaging device in this application also integrates or is connected to a processor for performing point cloud data preprocessing, symmetric residual distribution map construction, disturbance behavior analysis, and defect determination logic. The processor can be a computing unit with point cloud processing capabilities, such as a central processing unit, graphics processing unit, edge AI chip, FPGA, or DSP.

[0078] See also Figure 2 , which is a schematic diagram of artifact imaging provided in an embodiment of the present application.

[0079] Figure 2 The image shows a signal source projecting a signal into the pinhole area inside a connector. After specular reflection from the inner wall of the pinhole, the reflected signal is received by an imaging device and used to generate 3D point cloud data. Ideally, the imaging device should capture real points representing the actual surface structure to accurately reconstruct the geometry of the connector's inner wall.

[0080] It is easy to understand that since the interior of the connector is usually an axisymmetric structure and the inner wall is mostly made of metal or electroplated materials with strong mirror reflection ability, the signal light may be reflected one or more times between the two sides of the pinhole and finally received by the imaging device, resulting in part of the reflected signal not coming directly from the actual surface being measured, but from non-target positions on the reflection path of the optical path inside the structure.

[0081] Figure 2 It shows that these indirectly reflected imaging signals are misjudged by the system as point cloud information of the actual surface, which appears as a group of artifact points in the point cloud image. Especially in symmetrical structures, these artifact points often appear in a mirrored manner around the central axis of the connector and present a symmetrical mapping relationship with the real structural points in space, forming artifact interference.

[0082] Those skilled in the art will recognize that while artifact points are highly similar to real points in spatial distribution, their locations lack physical significance, significantly interfering with the stability and accuracy of defect detection. This application addresses this technical issue by proposing a method combining perturbation residual analysis with spatial gradient discrimination to remove artifact points caused by multipath reflection from point clouds, thereby improving the quality of 3D imaging for defect detection in connector inner wall structures.

[0083] Next, in conjunction with the accompanying drawings, a three-dimensional imaging detection method for surface defects of a connector provided in an embodiment of the present application is introduced. Figure 3 The method shown includes the following steps S1-S4, and the specific steps are as follows:

[0084] S1: Obtain a surface map in the depth direction based on the collected point cloud data;

[0085] In this application, point cloud data can be derived from three-dimensional imaging devices based on imaging mechanisms such as structured light projection, laser triangulation, time-of-flight depth perception, and multi-view stereo matching. Specific equipment may include industrial-grade structured light three-dimensional scanners, laser profilers, ToF depth cameras, or imaging devices with active illumination and spatial reconstruction capabilities. Point cloud data is usually composed based on the three-dimensional coordinates of each spatial point, and some detection methods can further directly provide additional attributes such as normal, reflectivity, or grayscale intensity. The point cloud acquisition process can be combined with different viewing angles, different exposure strategies, or active focusing to optimize data density and accuracy. Those skilled in the art will understand that the method of collecting point cloud data can be flexibly selected according to the specific detection environment, connector material, and hole structure characteristics, and is not limited here.

[0086] Furthermore, in order to improve the directionality and distribution regularity of the structural analysis, the point cloud data is projected into the polar coordinate space with the connector center as the reference, and the normal direction of the point cloud is used as the benchmark to generate a surface graph representing the depth disturbance trend.

[0087] S2: Using the central axis of the surface graph as the axis of symmetry, identify target points using morphological operators, calculate the spatial residual between each target point and its axisymmetric position point, and obtain a symmetric residual distribution map;

[0088] In this embodiment, based on the surface image generated by S1, the connector centerline is set as the symmetry reference axis. Projection density analysis is performed along the point cloud normal, extracting the projection signal curve and performing opening and closing operations to obtain a morphological reference curve. The residual segment between the original projection signal curve and the morphological reference curve is further calculated to identify local target points with significant curvature perturbation amplitudes. Subsequently, for each target point, its mirror image position about the symmetry axis is constructed, and the spatial coordinate residual is calculated, resulting in a symmetric residual distribution map encompassing all target point pairs.

[0089] S3: Execute an artifact point removal strategy in the symmetric residual distribution map to obtain a connector point cloud image;

[0090] In this embodiment, the perturbation intensity gradient direction of each target point in the local space is further calculated for the symmetric residual distribution map generated by S2. Based on the local distribution trend of the perturbation value, a determination is made as to whether the point is a perturbation convergence point or a perturbation divergence point. A joint analysis is then performed combining the spatial distribution relationship between symmetrical point pairs, the gradient direction angle, and the consistency of the perturbation direction. When a symmetrical point pair is simultaneously a perturbation convergence point and exhibits gradient convergence behavior with consistent symmetric directions, it is marked as a high-confidence artifact point pair and is removed as a whole.

[0091] It can be understood that the processing logic corresponding to this step is based on the multi-path reflection physical mechanism disclosed in this application. By modeling the structure of the disturbance behavior characteristics, it avoids the problem of false deletion of real defects caused by rough threshold judgment in traditional methods, and is suitable for typical connector configurations with a central axis structure and complex reflection interference paths.

[0092] S4: inputting the connector point cloud image into a preset defect recognition network, processing the connector point cloud image through the defect recognition network, outputting a defect area, and performing three-dimensional imaging of the defect area;

[0093] In this example, a point cloud image with artifacts removed is used as network input, invoking a trained defect recognition neural network model. The model comprises a point set geometry encoding layer, a spatial perturbation extraction layer, and a defect determination layer. A graph-structured convolution mechanism extracts the normal mutation relationship and curvature perturbation trend between point clouds, identifies the actual defect area, and further outputs the precise location of the defect area in three-dimensional space.

[0094] Before developing the specific technical content corresponding to the steps, the embodiments of the present application need to emphasize again that in the pinhole structure of a typical connector, due to the small space and regular structure, the inner wall is often made of metal or electroplated material, which has strong mirror reflection characteristics. When using structured light and other methods for three-dimensional imaging acquisition, the light signal may be reflected once or multiple times in the inner cavity, causing the position of the non-real surface to be recognized by the receiver as a valid signal point on the reflection path, and finally forming regularly distributed false structural information in the point cloud map. Unlike actual defects, this type of point is often distributed in space as a symmetrical mapping relationship around the central axis with a real point or a secondary refraction point, and the geometric perturbation trend is similar to that of the real defect. If only relying on conventional symmetry recognition or error filtering, the artifact point is often mistakenly retained as a defect point due to the overlap of the disturbance amplitude or the similarity of the texture, or the real defect is eliminated.

[0095] The processing logic adopted in this embodiment does not focus on improving defect recognition itself, but on calculating the changing trend of spatial disturbance behavior, that is, using the gradient directionality of the disturbance value as an additional identification basis, thereby adding a discrimination clue in addition to the defect morphology. Unlike traditional judgment criteria, this application judges whether the disturbance is concentrated or divergent in space. In engineering practice, it can be regarded as whether the interference behavior has a folded path or overlapping projections due to the mirror structure. Furthermore, when it is found that the gradient directions between multiple target points are symmetrical and point to each other, it can be logically deduced that it is the result of mirror interference imaging, and is eliminated with high confidence. In comparison, the method of the present application has higher stability and adaptability for imaging artifact problems in pinhole and blind cavity connector structures.

[0096] Next, the part of the method of the present application regarding curved surface images is further expanded.

[0097] In deep cavities like connector pinholes, the point cloud data acquired during acquisition can be spatially non-uniform due to strong inner wall reflections, limited angles of incidence, and the fact that imaging equipment is typically located externally. In particular, the number of point clouds tends to decay exponentially in depth, with only a small number of points concentrated in a narrow effective layer, and the overall structure presenting a discontinuous and irregular sheet-like structure.

[0098] Under the influence of this flaky structure, it is understandable that using conventional Cartesian coordinates for patch fitting can lead to geometric distortion due to insufficient support points in the fitting area or sharp edge transitions. In this embodiment, the original point cloud data is projected onto a two-dimensional surface corresponding to the normal direction, and a polar coordinate fitting method is used to interpolate and smooth the local high-density point areas. This results in a more stable restoration of the true curved surface structure of the pinhole inner wall and the specific location of the point cloud.

[0099] In an example, the specific steps for constructing a surface graph in S1 are as follows:

[0100] S1.1: Projecting the point cloud data into a polar coordinate space with the center of the connector as a reference axis;

[0101] In this embodiment, the central axis of the connector can be obtained by axially extracting the overall point cloud through the principal component analysis method. Specifically, the covariance matrix is ​​calculated in the original three-dimensional point set, the first principal direction is extracted as the initial axis direction, and then the outlier interference points are eliminated through the RANSAC algorithm to further accurately fit the structural axis of the connector pinhole.

[0102] Furthermore, the polar diameter, axial direction and polar angle of the three-dimensional coordinates of all points are calculated with the structural axis as the polar axis, thereby completing the conversion of the polar coordinate space. This application will not go into details here.

[0103] S1.2: Based on the projected point cloud data, multiple equidistant sub-regions are divided in the polar direction, and the point set that forms the local extreme point density in the axial direction in each sub-region is extracted;

[0104] Specifically, after projecting to the polar coordinate system, the distribution of the point cloud along each polar radius reflects the structural morphology of the connector's inner wall and deep cavity, with effective imaging points typically concentrated within a certain axial height range. In this step, the polar radius is divided into several equally spaced subregions. Each subregion is a fan-shaped flat ribbon, whose radial width can be adaptively set based on the total polar radius range to ensure resolution across connectors of varying sizes.

[0105] In this embodiment, to eliminate isolated high-density errors caused by single strong reflections, density statistics are performed on the point cloud within each subregion in the axial direction, and a sliding window extreme value extraction method is used to identify the axial point set with the main density peak. The density statistics process considers the minimum distance between local point clouds to avoid high-frequency fluctuations caused by angular occlusion or projection errors. Kernel-based smoothing filtering is also applied to ensure that the extracted extreme point sets not only represent the actual physical structure surface but also have a certain degree of spatial continuity. These extracted point sets constitute the candidate data for surface fitting.

[0106] S1.3: Using the density peak of the point set as a reference layer boundary, constructing a fitting surface with normal continuity constraints based on the normal change rate of the point set neighborhood to generate a surface graph;

[0107] Specifically, valid surfaces in pinhole-like structures typically only have dense and reliable data distribution within a certain spatial region. Therefore, this application selects only the identified extreme point set as the core region for geometric reconstruction. Based on this core region, we further use the normal estimation information of each point to analyze the normal change rate within its local neighborhood to determine whether the region has continuous and smooth morphological characteristics.

[0108] In this embodiment, the normal vector is obtained using a method based on PCA local neighborhood fitting, and the neighborhood radius or the number of K neighboring points can be dynamically adjusted to adapt to changes in point density in different polar radius areas.

[0109] Furthermore, the present application introduces normal continuity constraints in the process of generating the fitted surface. Specifically, the local interpolation model of the surface can be optimized by jointly evaluating the fitting residual and the change amplitude of the normal angle at each point.

[0110] See also Figure 4 , Figure 4 This is a schematic diagram of the target point recognition process in an embodiment of the present application. It can be understood that the present application is aimed at three-dimensional imaging of defective parts rather than three-dimensional imaging of the entire connector. There are two solutions to achieve this goal. The first solution is to image the collected point cloud data as a whole and then locate the defective area from it. The second solution is to pre-process the point cloud and separately filter out point clouds with abnormal morphology and regard them as point clouds with possible defective areas.

[0111] Understandably, the compact structure and complex reflection paths of typical micro-connectors, pinhole interfaces, and housing interiors make overall point cloud imaging highly susceptible to noise, redundant artifacts, and optical distortion. Performing symmetry analysis and 3D reconstruction on the entire point cloud indiscriminately would significantly increase computational costs and lead to higher misjudgment rates.

[0112] Specifically, Figure 4 The method flow shown is based on the normal perturbation and spatial density change of the point cloud in the surface image, pre-screening the areas with abnormal perturbations in the local structure, focusing on the selective modeling of the real defect areas from the overall point cloud, reducing the invalid calculation overhead of the complete data, and improving the recognition reliability and imaging resolution of local defects in complex structures.

[0113] Next, the target point recognition part of the present application method will be further expanded.

[0114] Figure 4 The method shown can be applied to the aforementioned S2, and the specific steps are as follows:

[0115] S2.1: Determine a normal projection direction of the point cloud based on the symmetry axis and the distribution of the point cloud in the surface image, and calculate a projection signal curve based on the normal projection direction and the spatial density of the point cloud;

[0116] Specifically, in the surface map constructed by the point cloud, the point cloud is distributed on the three-dimensional surface of the inner wall of the connector cavity, and the surface as a whole is constructed with the central axis of the jack as the axis of symmetry. Therefore, the projection direction can be established through the radial or normal direction defined by the axis to capture the distribution of the morphological disturbance of the curved surface in this direction.

[0117] In this example, the connector's structural center axis is first determined by extracting dense point cloud regions from the surface image using a least-squares fitting method. Subsequently, a normal direction referenced to the structural center axis is defined as the primary projection direction, and all point clouds are one-dimensionally projected along this direction. During this one-dimensional projection, the local spatial density of the point cloud is used as a weighting parameter to enhance resolution in sparse regions, resulting in a representative projection signal curve.

[0118] S2.2: Performing morphological opening and closing operations on the projection signal curve to obtain a morphological reference curve;

[0119] Specifically, morphological operations, as a spatial structure recognition tool, are used in this step to construct a reference signal that reflects the ideal state of the surface.

[0120] Understandably, the signal in the defective area of ​​the projection signal curve in the surface plot is often weak and has a small amplitude, making it difficult to directly identify the disturbance location corresponding to the defect from the projection signal curve. Therefore, morphological operator processing is used to eliminate non-structural disturbances, thereby obtaining a morphological reference curve that can be used as a comparison benchmark.

[0121] Taking the curve as an example, you can refer to Figure 5 To understand, Figure 5 This is a schematic diagram of the principle of generating the morphological reference curve according to the embodiment of the present application. Figure 5 A certain projection signal curve of a certain point cloud is shown.

[0122] Figure 5 This figure shows a projection signal curve in the normal direction based on point cloud data, reflecting the geometric perturbations of the connector surface in a specific area. Several small local fluctuations can be observed in this curve, which may correspond to minor defects in the actual structure, material fluctuations, or interference points caused by specular reflections.

[0123] Figure 5A set of nonlinear structural elements is shown for adaptively matching the periodic and radial geometric variation trends in the connector cavity structure. Based on the nonlinear structural elements, erosion and dilation operations are performed on the projected signal curve.

[0124] Figure 5 The position where the corrosion operation works is shown, that is, the minimum value is selected within the sliding window of the structural element so that the local high point at that position is flattened.

[0125] Figure 5 The position of the expansion operation is also shown, that is, the maximum value is taken in the window to fill the concave segment.

[0126] Figure 5 It is further shown that after the above-mentioned erosion and dilation operations, a morphological reference curve as shown by the dotted line can be formed to reflect the ideal structural trend of the point cloud.

[0127] It is understandable that Figure 5 The shown fragments are merely exemplary, used to assist in explaining the application process of the morphological processing method in the projection signal curve, and do not limit the specific shape and dimension of the surface diagram or point cloud data in this application.

[0128] In an example, the steps for generating a morphological reference curve are as follows:

[0129] S2.2.1: Generate a nonlinear structuring element based on the morphology of the surface image and the density variation of the point cloud in the normal projection direction, wherein the nonlinear structuring element is calculated based on the periodic characteristics of the density variation;

[0130] Specifically, point clouds often exhibit repetitive peaks and valleys within the projection domain, but different regions may exhibit periodic variations or morphological anomalies. Using linear structuring elements would be unable to adapt to these variations. Therefore, in this embodiment, a periodic template signal is first extracted based on the connector's axial projection density function. This template signal is based on the peak position of the point cloud distribution in the projection direction and, combined with the calculated local average period width, a set of structuring elements with a non-uniform distribution is constructed.

[0131] Preferably, the shape of the structural element during the sliding process can be automatically adjusted as the local surface shape changes. Its function is to perform adaptive feature extraction based on the changes in curvature radius and arrangement period at different positions on the connector surface, so as to improve the adaptability of corrosion and expansion operations to morphological changes.

[0132] S2.2.2: Using the nonlinear structuring element as a sliding window, perform an erosion operation and a dilation operation on the projection signal curve, wherein the erosion operation is performed based on a local minimum value within the sliding window, and the dilation operation is performed based on a local maximum value within the sliding window;

[0133] Specifically, the nonlinear structuring element slides along the projected signal curve with a fixed sliding step size, performing morphological operations on the signal data within each window. The erosion operation searches for the local minimum within each sliding window segment and replaces the window center with this minimum, thereby globally weakening the micro-protrusion structure. Correspondingly, the dilation operation selects the local maximum within each window and uses it to replace the center position, effectively filling the micro-depression area.

[0134] S2.2.3: Merging the results of the erosion operation and the dilation operation into a smooth envelope curve as a morphological reference curve of the projection signal curve;

[0135] Specifically, after the morphological operation is completed, the embodiment performs weighted fusion of the erosion curve and the dilation curve, and the weight can be adjusted based on the directionality and rate of change of the curve fluctuation, ultimately constructing an envelope curve that is highly consistent with the original projection signal in trend but with the local disturbance components smoothed.

[0136] S2.3: performing residual comparison between the projection signal curve and the morphological reference curve, and screening point clouds whose residual amplitude is greater than a disturbance judgment threshold to generate target points;

[0137] Next, the part of the target point screening method of this application is further expanded.

[0138] S2.4: Obtain an axisymmetric mapping point for each target point, calculate the normal residual and the curvature residual between the target point and the axisymmetric mapping point, and generate a symmetric residual distribution map based on the normal residual and the curvature residual;

[0139] It can be understood that the target point set is a candidate set that widely includes disturbance positions, which includes both real defect points and artifact points.

[0140] In this application, the inner wall of the connector is a cavity-type mirror-symmetrical structure. It can be understood that the artifact points must come from the reflection effect of the real points or other artifact points, that is,

[0141] In one case, if a target point does not have a corresponding axisymmetric mapping point, then the target point is theoretically a real defect point that does not have a corresponding reflection during the imaging process, and there is no need to perform artifact point screening.

[0142] In another case, if a target point has one or more corresponding axisymmetric mapping points, these points form a reflection point set, which contains only one real defect point. This application needs to further process the reflection point set, eliminate artifact points, and obtain the real defect point.

[0143] In this embodiment, symmetric mapping is first performed on each identified target point, using the central axis of the surface image as the reference axis. This symmetric mapping process is performed in a preset polar coordinate space. By performing an axisymmetric transformation on the polar angle components of the target point, the theoretical symmetric mapping point position is obtained. The actual point correspondence is then searched for in the original point cloud. If no precise mapping point exists, the target point is not further processed. The target point is selected and aggregated with the target points obtained from further processing in the candidate set to generate the connector point cloud image.

[0144] Furthermore, target points with axisymmetric mappings are aggregated. For each symmetrical point pair, the difference in normal vector angle and curvature between the two points is calculated, using the radial position and polar angle information in polar coordinate space as a reference. This yields a corresponding set of structural perturbation consistency indices. Subsequently, all symmetrical point pairs are used as sample points to construct a symmetric residual distribution map.

[0145] It can be understood that the symmetric residual distribution map uses spatial position as the basic coordinate, and normal residual and curvature residual as multi-dimensional perturbation amplitude indicators to describe the structural consistency pattern in the entire surface map that can be derived from the symmetric mapping.

[0146] Taking axisymmetric mapping points as an example, please refer to Figure 6 , Figure 6 The schematic diagram of the axisymmetric mapping point generation for the embodiment of the present application is shown below. Figure 6 Three cases are shown: no symmetric relationship, one-to-one mapping symmetric relationship, and multiple-pair symmetric mapping relationships.

[0147] Figure 6 It shows that target point 1, target point 2 and target point 3 are multiple pairs of symmetrical mapping relationships, so these three target points are a group. In subsequent processing, only the corresponding three target points need to be analyzed to obtain the final true defect point.

[0148] Figure 6 It is further shown that the target point 4 and the target point 6 are in a one-to-one mapping symmetric relationship, so these two target points are a group. In the subsequent processing, only the corresponding two target points need to be analyzed to obtain the final true defect point.

[0149] Figure 6 It is further shown that the axially symmetrical mapping point of target point five does not have a corresponding target point, that is, there is no symmetrical relationship between target point five. In this application, there is no need to further process target point five, and the connector point cloud image can be generated after the processing of other target points is completed.

[0150] Next, the part of the method of the present application regarding the removal of artifact points is further expanded.

[0151] It is understandable that during structured light or laser imaging, defects or surface disturbances can generate multipath reflections through the inner wall, forming symmetrically distributed virtual image points. These artifact points closely resemble real defects in surface morphology, with highly overlapping local normal features and curvature disturbance characteristics, but their spatial positions are typically offset by a certain axisymmetric distance from the real points. The main characteristics of these artifact points are: a single source, multiple results, and a regular, symmetrical clustered distribution. In contrast, real defects, because they actually exist on the surface of the structure, their disturbances are not reflected by other points and are therefore more likely to exist independently in a divergent form.

[0152] Those skilled in the art can now understand that the present application believes that: if a disturbance point is located at the starting point of multiple low-residual symmetric points in the symmetric residual distribution map, and its disturbance direction shows an obvious gradient divergence trend compared with the corresponding axisymmetric mapping point, then the point is more likely to be the source of the actual defect; and those points that show convergence characteristics in the disturbance direction are more likely to be artifact points generated by reflection paths or structural coupling.

[0153] In an example, the specific steps for S3 are as follows:

[0154] S3.1: Construct a disturbance spatial gradient field based on the symmetric residual distribution map, and calculate the disturbance intensity spatial gradient of each target point in the map;

[0155] Specifically, for each target point, the corresponding normal residual and curvature residual values ​​are obtained, and the spatial gradient of the disturbance intensity is calculated based on the spatial distribution of the neighboring point set in the three-dimensional polar coordinate system. The calculation process traverses the residual variation trends of each point in the polar radial and polar angular directions, forming a disturbance vector centered on the target point in the local disturbance field. The direction and magnitude of the disturbance vector reflect the dominant flow direction at that point in the surrounding disturbance field.

[0156] S3.2: Dividing the target point into a disturbance convergence point and a disturbance divergence point according to the gradient direction of the disturbance intensity spatial gradient;

[0157] Specifically, in the aforementioned steps, the target points with an axially symmetric relationship have been aggregated, that is, the perturbation space gradient field constructed by the point cloud in the polar coordinate space is used to perform gradient direction analysis on each target point. Specifically, for a certain target point, a perturbation gradient vector is formed by statistically analyzing the changing trend of the perturbation intensity of its adjacent points in the radial and axial directions. The vector reflects whether the point is in the initiation direction or the receiving direction of the perturbation. If there are multiple perturbation gradient vectors pointing to a point among its adjacent points, it means that the point has gathered perturbation energy from multiple directions and is judged to be a perturbation convergence point; on the contrary, if most of the perturbation vectors of the point point to the adjacent points, that is, the perturbation diverges outward from the point, then the point is considered to be a perturbation divergence point.

[0158] S3.3: Eliminate the disturbance convergence points and generate a connector point cloud image based on the point clouds corresponding to the remaining disturbance divergence points;

[0159] It can be understood that, in addition to the point clouds corresponding to the remaining disturbance divergence points, the present application also generates a connector point cloud image based on the point clouds corresponding to the aforementioned target points where there are no axisymmetric mapping points.

[0160] Those skilled in the art will understand that how to generate a point cloud image based on a point cloud is common knowledge and will not be elaborated in this application. However, it should be noted that since the aforementioned steps have performed multiple steps of screening on the point cloud, the final point cloud data may be sparse and not in piecemeal form. Therefore, it is necessary to aggregate the point cloud that was screened out in advance in the aforementioned steps with the final point cloud to generate a complete point cloud image.

[0161] As a preference, the present application further analyzes the elimination of disturbance convergence points.

[0162] Specifically, a perturbation convergence point refers to a point in the symmetric residual distribution map where the spatial gradient direction of the perturbation intensity tends to point to the point itself. In terms of spatial structure, this gradient convergence phenomenon may be reflected as a typical feature of the cause of artifact interference. However, in actual point cloud data, some real defect areas may also present the geometric form of local perturbation convergence. For example, the closed end points of crack-like defects often appear as high-density edge structures in imaging due to their microscopic morphology, forming a similar perturbation centripetal aggregation effect; for example, the inner concave surface area formed by pits, ablation or erosion, its normal mutation and curvature concentration may also generate a convergence gradient pattern similar to the artifact point in the perturbation field.

[0163] It is easy to understand that if all gradient convergence target points are directly excluded when removing artifact points, the point cloud corresponding to the above-mentioned real defect features will most likely be mistakenly deleted, thereby affecting the integrity and confidence of the final detection. At the same time, this embodiment also takes into account that the cavity structure is limited by factors such as the incident angle, non-ideal reflection path, and inconsistent surface coating during the actual imaging process. Some artifact points do not strictly exhibit typical gradient convergence behavior, but exist in the form of weaker perturbation gradients or local asymmetry. If only relying on the directionality of the gradient as the basis for exclusion, this part of the artifact area may also be missed, resulting in misidentification.

[0164] In this preferred embodiment, it is further determined whether these convergence points form a mirror point pair relationship under an axisymmetric structure. Based on the central axis of the connector as the geometric reference, a symmetric mapping operation is performed on each perturbed convergence point, and a corresponding point is searched in the actual point cloud for its axisymmetric position. If a corresponding point exists, the following conditions are further determined:

[0165] First, the symmetric point itself is also a perturbation convergence point, and has similar local perturbation intensity gradient convergence characteristics as the original point;

[0166] Secondly, the gradient directions between the original point and its symmetrical point are mirror images in space, that is, the perturbation gradient directions of the two are basically symmetrical about the central axis;

[0167] Thirdly, the disturbance intensity values ​​of the two points are close within a certain tolerance range, indicating that they may originate from the same imaging interference or reflection path response;

[0168] When the above three conditions are met at the same time, it can be reasonably inferred that the point pair is a typical mirror artifact pair with symmetrical spatial distribution, similar perturbation behavior and consistent interference source, and therefore it is eliminated as a whole.

[0169] Furthermore, this embodiment does not directly remove convergence points that fail to form a mirror-image pair relationship, that is, if their symmetrical positions do not have corresponding perturbation convergence points, or if the perturbation direction / intensity does not meet the consistency requirements. Such asymmetric convergence points are temporarily retained as part of the disturbance region to be determined and enter the subsequent defect recognition network. The network determines whether they constitute true defects based on deeper geometric features and contextual information, thereby achieving a high-precision and high-fidelity defect recognition process.

[0170] Preferably, the present application further analyzes the retention of disturbance divergence points.

[0171] Specifically, real defect points often appear in the disturbance gradient field as outward-diverging spatial gradients of disturbance intensity. Therefore, retaining all divergent points helps ensure the complete capture of the defect area. However, under specific structural conditions, some artifact points may also form visual features of disturbance gradient divergence under the interference of three or more optical reflection paths, thereby causing interference and misjudgment of the recognition results.

[0172] It is easy to understand that due to the typical mirror-symmetric cavity structure inside the connector, when the laser or imaging beam is reflected multiple times on the inner wall, secondary artifact points may be formed under the non-main axisymmetric structure. Especially when the reflection path involves multiple surface intersections or multi-cavity turning sections, the spatial position of such artifact points deviates from the initial reflection symmetry axis, and the gradient direction of their perturbation no longer shows strong symmetry, but tends to be discrete or locally perturbed and divergent. Such high-order reflection artifact points are difficult to capture under traditional residual analysis and axisymmetric screening mechanisms. In the perturbation gradient field, they may appear as false divergence points and be mistakenly classified as real defects.

[0173] In this preferred embodiment, among the target points identified as disturbance divergence points, the angle between the disturbance direction and the normal of the inner wall of the connector is further calculated to determine whether the source of the disturbance conforms to the law of single reflection propagation; if the disturbance direction deviates from the normal of the inner wall and the disturbance value is less than the preset lower limit, it is marked as an artifact inheritance point formed by multiple reflections and is eliminated.

[0174] The defect recognition network includes a point set geometry encoding layer, a spatial disturbance extraction layer, and a defect determination layer, wherein:

[0175] The point set geometry encoding layer is used to receive the connector point cloud image and construct a geometric feature vector based on the spatial coordinates, normal vector and local curvature of each point cloud;

[0176] The spatial perturbation extraction layer is provided with a graph convolution network for calculating normal mutation relationships and local curvature minimum regions in a spatial adjacency graph constructed by geometric feature vectors. The spatial perturbation extraction layer includes an adjacency graph construction module, a graph convolution calculation module, and a differentiation module, wherein:

[0177] The adjacency graph construction module is used to generate a spatial adjacency graph with Euclidean distance as edge weight based on the spatial coordinates and geometric feature vector of each point;

[0178] The graph convolution calculation module uses a weighted graph convolution network to propagate and aggregate information on the nodes in the spatial adjacency graph, and extract the normal change rate and curvature gradient of each point in its neighborhood;

[0179] The distinguishing module determines whether the normal change rate of each point in its neighborhood is greater than the weighted standard deviation of the mean normal change rate of the neighborhood based on the normal change rate and curvature gradient of the point in its neighborhood, so as to identify the normal mutation relationship, and identifies the local curvature minimum area when the curvature gradient is less than a preset first threshold and the curvature value is the minimum value in its neighborhood;

[0180] The defect determination layer is used to output the position label of the defect area and its corresponding three-dimensional coordinate set based on the normal mutation relationship and the local curvature minimum area. The defect determination layer includes:

[0181] Performing local connectivity analysis on the normal mutation relationship and the curvature minimum value region to generate a candidate defect point set;

[0182] Clustering the candidate defect point set according to a clustering algorithm to obtain a clustering result;

[0183] Calculating the defect confidence of the point set area corresponding to the clustering result according to the spatial distribution density, curvature gradient variation range and normal consistency of the point cloud in the clustering result;

[0184] The point set region where the defect confidence is greater than the defect threshold is output as a defect region.

[0185] In the embodiments of this application, a defect recognition network is used to identify actual defect areas from artifact-removed connector point cloud images. Its core goal is to accurately identify and locate defects with subtle geometric disturbances. This network structure is specifically designed to fully integrate the spatial geometric characteristics of the point cloud and the structural regularity of the connector's inner wall to support a three-stage recognition process for fine-grained disturbance detection: geometric feature encoding, spatial disturbance extraction, and defect area determination.

[0186] In this embodiment, the point set geometry encoding layer first encodes the input point cloud image. By combining geometric descriptors such as the 3D coordinates, normal vector direction, and local curvature of each point, it constructs a geometric feature vector that describes the local structural state of the point. This feature representation not only preserves the positional distribution of the points but also enhances their distinguishability under slight geometric changes.

[0187] The spatial perturbation extraction layer further builds on these geometric features by constructing a spatial adjacency graph and performing feature propagation and aggregation through a graph convolutional network, thereby encoding the geometric perturbation relationship between each point and its neighborhood. This process effectively captures spatial perturbation features such as normal abrupt changes and curvature gradient changes through the graph convolution computation module. The discrimination module then extracts regional points within the local neighborhood of a point that meet certain abrupt changes and minimum curvature conditions, laying the foundation for the subsequent identification of typical morphological changes such as defect edges, pores, and protrusions.

[0188] The defect judgment layer performs fusion judgment on the extracted spatial disturbance information. Through local connectivity analysis, the network first selects a set of candidate points that may constitute defect boundaries or internal disturbances, and structurally aggregates these points through a clustering algorithm. Finally, based on the multi-dimensional characteristics of the points in the cluster area, such as the spatial density, curvature change amplitude, and normal vector consistency, the confidence score of the area as a defect is comprehensively calculated. Only when the score exceeds a specific threshold, it is identified as a valid defect area and its three-dimensional coordinate set is output. This processing method enables the present application to not only have the ability to identify visible surface structural disturbances, but also effectively eliminate artifact points and retain real defects in point cloud data with complex structures and strong reflection interference.

[0189] Those skilled in the art will understand that although the present embodiment uses a graph convolutional network structure and specific geometric rules as the implementation form of the defect recognition network, it is not limited to this. Other neural network architectures such as PointNet, DGCNN or Transformer can also be used to achieve the defect recognition goals of this application under the premise of meeting the requirements for local disturbance perception capabilities. The construction principle of the defect recognition network is based on the structural consistency of the point cloud in space, the local disturbance pattern and the geometric mutation characteristics. Through high-dimensional feature aggregation and confidence modeling strategies, it can achieve stable recognition and accurate calibration of defects on the inner wall of the connector. This application does not make specific limitations.

[0190] In one example, performing three-dimensional imaging on the defect area includes:

[0191] Extracting a three-dimensional point cloud coordinate set corresponding to the defect area, and constructing a spatial voxel grid based on the three-dimensional point cloud coordinate set;

[0192] Interpolation fitting is performed on the spatial voxel grid to generate a three-dimensional image of the defect.

[0193] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A three-dimensional imaging detection method for connector surface defects, characterized in that: The method comprises: According to the collected point cloud data, a surface map in the depth direction is obtained; Taking the central axis of the surface graph as the symmetry axis, identifying the target points through morphological operators, calculating the spatial residual between each target point and its axisymmetric position point, and obtaining a symmetric residual distribution map; In the symmetric residual distribution map, an artifact point removal strategy is executed to obtain a connector point cloud image, wherein the artifact point removal strategy determines whether there are artifact points based on the disturbance trend of the target point, and removes the target point with the disturbance trend converging; The connector point cloud image is input into a preset defect recognition network, the connector point cloud image is processed by the defect recognition network, the defect area is output, and the defect area is three-dimensionally imaged, wherein the defect recognition network is trained by historical defect data of the connector.

2. The method for detecting surface defects of a connector using three-dimensional imaging according to claim 1, wherein: Taking the central axis of the surface graph as the axis of symmetry, the target point is identified by using a morphological operator, including: Determine the normal projection direction of the point cloud according to the symmetry axis and the distribution of the point cloud in the surface diagram, and calculate the projection signal curve according to the normal projection direction and the spatial density of the point cloud; Performing morphological opening and closing operations on the projection signal curve to obtain a morphological reference curve; The projection signal curve and the morphological reference curve are compared for residuals, and point clouds with residual amplitudes greater than a disturbance judgment threshold are screened to generate target points.

3. The method for detecting surface defects of a connector using three-dimensional imaging according to claim 2, wherein: Performing morphological opening and closing operations on the projection signal curve to obtain a morphological reference curve includes: generating a nonlinear structural element according to the morphology of the surface image and the density variation of the point cloud in the normal projection direction, wherein the nonlinear structural element is calculated according to the periodic characteristics of the density variation; Using the nonlinear structural element as a sliding window, performing an erosion operation and a dilation operation on the projection signal curve, wherein the erosion operation is performed according to a local minimum value within the sliding window, and the dilation operation is performed according to a local maximum value within the sliding window; The results of the erosion operation and the dilation operation are fused into a smooth envelope curve, which serves as a morphological reference curve of the projection signal curve.

4. The method for three-dimensional imaging detection of surface defects of a connector according to claim 1, characterized in that: Calculate the spatial residual between each target point and its axisymmetric position point, including: Obtain the axisymmetric mapping point of each target point; The normal residual and the curvature residual between the target point and the axisymmetric mapping point are calculated.

5. The method for three-dimensional imaging detection of surface defects of a connector according to claim 1, characterized in that: In the symmetric residual distribution map, an artifact point removal strategy is executed to obtain a connector point cloud image, including: Constructing a disturbance spatial gradient field according to the symmetric residual distribution map, and calculating the disturbance intensity spatial gradient of each target point in the map; Dividing the target point into a disturbance convergence point and a disturbance divergence point according to the gradient direction of the disturbance intensity spatial gradient; The disturbance convergence points are eliminated, and a connector point cloud image is generated based on the point clouds corresponding to the remaining disturbance divergence points.

6. The method for three-dimensional imaging detection of surface defects of a connector according to claim 1, characterized in that: The construction of the surface graph includes: Projecting the point cloud data into a polar coordinate space with the center of the connector as a reference axis; Based on the projected point cloud data, multiple equidistant sub-regions are divided in the polar direction, and the point set that forms the local extreme point density in the axial direction in each sub-region is extracted; The density peak of the point set is used as the reference layer boundary, and a fitting surface with normal continuity constraints is constructed according to the normal change rate of the point set neighborhood to generate a surface graph.

7. The method for detecting surface defects of a connector using three-dimensional imaging according to claim 1, characterized in that: The defect recognition network includes a point set geometry encoding layer, a spatial disturbance extraction layer, and a defect determination layer, wherein: The point set geometry encoding layer is used to receive the connector point cloud image and construct a geometric feature vector based on the spatial coordinates, normal vector and local curvature of each point cloud; The spatial perturbation extraction layer is provided with a graph convolutional network for calculating normal mutation relations and local curvature minimum regions in a spatial adjacency graph constructed by geometric feature vectors; The defect determination layer is used to output a position label of the defect area and a corresponding three-dimensional coordinate set thereof according to the normal mutation relationship and the local curvature minimum area.

8. The method for three-dimensional imaging detection of surface defects of a connector according to claim 7, characterized in that: The spatial perturbation extraction layer includes an adjacency graph construction module, a graph convolution calculation module, and a differentiation module, wherein: The adjacency graph construction module is used to generate a spatial adjacency graph with Euclidean distance as edge weight based on the spatial coordinates and geometric feature vector of each point; The graph convolution calculation module uses a weighted graph convolution network to propagate and aggregate information on the nodes in the spatial adjacency graph, and extract the normal change rate and curvature gradient of each point in its neighborhood; The distinguishing module determines whether the normal change rate of each point in its neighborhood is greater than the weighted standard deviation of the mean normal change rate of the neighborhood based on the normal change rate and curvature gradient of the point in its neighborhood, so as to identify the normal mutation relationship, and identify the local curvature minimum area when the curvature gradient is less than a preset first threshold and the curvature value is the minimum value in its neighborhood.

9. The method for three-dimensional imaging detection of surface defects of a connector according to claim 7, characterized in that: The defect determination layer includes: Performing local connectivity analysis on the normal mutation relationship and the curvature minimum value region to generate a candidate defect point set; Clustering the candidate defect point set according to a clustering algorithm to obtain a clustering result; Calculating the defect confidence of the point set area corresponding to the clustering result according to the spatial distribution density, curvature gradient variation range and normal consistency of the point cloud in the clustering result; The point set region where the defect confidence is greater than the defect threshold is output as a defect region.

10. The method for detecting surface defects of a connector using three-dimensional imaging according to claim 1, characterized in that: Performing three-dimensional imaging on the defect area includes: Extracting a three-dimensional point cloud coordinate set corresponding to the defect area, and constructing a spatial voxel grid based on the three-dimensional point cloud coordinate set; Interpolation fitting is performed on the spatial voxel grid to generate a three-dimensional image of the defect.

Images