A method for automatic identification and defect screening of cylindrical shell welds

Abstract

The embodiment discloses a kind of cylindrical shell weld automatic identification and defect screening method, can realize pixel level accurate segmentation in complex curved surface, reflection and splash interference working condition in weld area, and the detection of identification and screening of defect is carried out, improves weld identification accuracy and defect screening reliability.Firstly, through the multi-view original surface image data of the weld area of the cylindrical shell workpiece, the enhanced image after the surface of the cylindrical shell is unfolded and the high reflection area mask are obtained, and according to the enhanced image after the surface is unfolded, the weld candidate area is generated, the weld body fine segmentation mask, the welding spatter area mask and the suspected defect area mask are obtained by deep learning, the weld pixel level subdivision is realized, and the weld body, spatter and defect area can be distinguished;Simultaneously, the consistency verification score after introducing welding spatter suppression is used for defect judgment, significantly reduces two-dimensional misjudgment and improves defect screening accuracy.

Description

Technical Field

[0001] This invention relates to the field of defect identification technology, and in particular to a method for automatic identification and defect screening of weld seams in cylindrical shells. Background Technology

[0002] Cylindrical shell components are typically formed by segmented welding, with welds distributed on complex curved surfaces and exhibiting irregular shapes. These welds are prone to defects such as uneven weld reinforcement, undercut, and incomplete penetration. Furthermore, spatter often adheres to the surface, accompanied by metallic reflection. Traditional machine vision methods relying on threshold segmentation or simple template matching are easily affected by reflections, noise, and surface distortion, making it difficult to achieve pixel-level accurate weld identification and reliable defect screening. Summary of the Invention

[0003] This invention discloses an automatic identification and defect screening method for weld seams in cylindrical shells to overcome the above-mentioned technical problems.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows: An automatic identification and defect screening method for weld seams in cylindrical shells includes the following steps: S1: Acquire multi-view raw surface image data and contour scan data of the weld area of ​​the cylindrical shell workpiece; S2: Based on the multi-view original surface image data of the weld area and the contour scan data of the weld area, respectively, obtain the three-dimensional points in the workpiece coordinate system and the three-dimensional contour points in the workpiece coordinate system; S3: Based on the three-dimensional points in the workpiece coordinate system, obtain the enhanced image and high-reflection area mask after unfolding the curved surface of the cylindrical shell; S4: Based on the enhanced image after surface unfolding, generate weld candidate regions to obtain the weld candidate region constraint mask, the candidate region center axis initial value prior map, and the length continuity evaluation prior map; S5: Based on the enhanced image after surface unfolding, the high-reflectivity area mask, the weld candidate area constraint mask, the candidate area central axis initial value prior map, and the length continuity evaluation prior map, the U-Net semantic segmentation network is used to obtain the multi-class probability output results at the unfolded coordinates in the enhanced image after surface unfolding, so as to obtain the weld body fine segmentation mask, the welding spatter area mask, and the suspected defect area mask. S6: Based on the fine segmentation mask of the weld body, obtain the result of the two-dimensional geometric representation of the weld, including the two-dimensional centerline discrete point, the arc length position at the centerline point, the two-dimensional local width at the centerline point, the left boundary point at the centerline point, and the right boundary point at the centerline point; S7: Based on the fine segmentation mask of the weld body, the result of the two-dimensional geometric representation of the weld, and the three-dimensional contour points in the workpiece coordinate system, obtain the three-dimensional contour point set of the weld to establish a three-dimensional surface model of the weld, and then obtain the three-dimensional centerline point, three-dimensional local section, three-dimensional local normal geometric reference, and weld reinforcement height. S8: Based on the suspected defect area mask, the two-dimensional centerline discrete points, the arc length position at the centerline point, and the two-dimensional local width at the centerline point, determine the defect candidate area to obtain the two-dimensional appearance anomaly intensity score of the defect candidate area; and obtain the arc length center position of the candidate area, and obtain the overlap ratio between the defect candidate area and the welding spatter area based on the welding spatter area mask; S9: Determine the arc length position at the centerline point closest to the arc length center position of the defect candidate region, and obtain the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the local concavity depth of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the local convexity height of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the residual height deviation of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, and the average defect confidence of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region; then, based on the overlap ratio between the defect candidate region and the welding spatter region, obtain the consistency verification score of the defect candidate region after the introduction of welding spatter suppression, so as to determine whether the defect candidate region is a valid defect region.

[0005] Beneficial Effects: The present invention provides an automatic identification and defect screening method for weld seams in cylindrical shells. This method achieves pixel-level precise segmentation of the weld seam area under conditions of complex curved surfaces, reflections, and spatter interference, and detects and screens defects, improving the accuracy of weld seam identification and the reliability of defect screening. Firstly, using multi-view original surface image data of the weld seam area of ​​the cylindrical shell workpiece, an enhanced image and a high-reflectivity area mask are obtained after unfolding the curved surface of the cylindrical shell. Based on the enhanced image after unfolding the surface, candidate weld seam areas are generated. Deep learning is used to obtain fine segmentation masks for the weld body, welding spatter areas, and suspected defect areas, achieving pixel-level fine segmentation of the weld seam and distinguishing between the weld body, spatter, and defect areas. Simultaneously, a consistency verification score with welding spatter suppression is used for defect judgment, significantly reducing two-dimensional misjudgments and improving defect screening accuracy. It adapts to curved surface distortion and reflection conditions, possesses strong robustness, and can achieve pixel-level precise identification of weld seams and reliable defect screening. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart of the automatic identification and defect screening method for cylindrical shell welds according to the present invention; Figure 2 This is a flowchart of the automatic identification and defect screening method for cylindrical shell welds according to an embodiment of the present invention. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] This embodiment introduces a method for automatic identification and defect screening of weld seams in cylindrical shells, including the following steps: Figure 1 and Figure 2 As shown: S1: Acquire multi-view raw surface image data and contour scan data of the weld area of ​​the cylindrical shell workpiece; This embodiment uses a ring array area array industrial camera and a line laser profile sensor to synchronously acquire data of the weld area of ​​a cylindrical shell, including: arranging a ring array area array industrial camera on the outer periphery of the area to be inspected (weld area) of the cylindrical shell to acquire multi-view original surface image data of the weld area of ​​the cylindrical shell; and arranging a line laser profile sensor in the detection area corresponding to the ring array area array industrial camera to acquire profile scanning data of the same weld area.

[0010] In this embodiment, a programmable logic controller (PLC) trigger unit synchronously triggers the ring array industrial camera and the line laser contour sensor, ensuring a one-to-one correspondence between the multi-view raw surface image data and the line laser contour scan data in the time dimension. During the acquisition process, the circumferential angle position, axial displacement position, and acquisition sequence number of the cylindrical shell workpiece are recorded simultaneously. The multi-view raw surface image data, line laser contour scan data, circumferential angle position, and axial displacement position acquired at the same time are combined into a synchronous acquisition data set and output. Finally, the multi-view raw surface image data, line laser contour scan data, and acquisition sequence number are output.

[0011] Specifically, the first The data sets obtained from the second synchronous acquisition are: (1) In the formula: Indicates the first Secondary synchronous data acquisition group; Indicates the first The multi-view raw surface image data obtained from the second acquisition; Indicates the first Line laser contour scanning data obtained from the first acquisition; Indicates the first The position of the workpiece's circumferential corner during the next data acquisition; Indicates the first The axial position of the workpiece during the first acquisition; Indicates the first A unified trigger time for each data collection; This serves as both a data collection count index and a data collection viewpoint index; specifically, each data collection corresponds to a viewpoint, therefore in this embodiment, the first... The data collected in the first synchronous acquisition is the first... Data from multiple perspectives.

[0012] Specifically, to ensure that the image corresponds to the contour data, the following conditions must be met: (2) in: Indicates the camera exposure time; Indicates the sampling time of the line laser profile sensor; This indicates the permissible synchronization error threshold.

[0013] Let the radius of the cylindrical shell be... Then the position of the arc length in the circumferential direction of the workpiece is: (3) in: Indicates the first The corresponding circumferential arc length position is collected each time; This indicates the radius of the cylindrical shell.

[0014] S2: Based on the multi-view original surface image data and the contour scan data of the weld area, respectively, obtain the corrected multi-view original surface image data (3D points in the workpiece coordinate system) and the corrected contour scan data of the weld area (3D contour points in the workpiece coordinate system). ).

[0015] Specifically, in this embodiment, the multi-view raw surface image data is used to calibrate the internal parameters of each ring array industrial camera to obtain the focal length, principal point coordinates, and distortion coefficients of the ring array industrial camera. Simultaneously, the multi-view raw surface image data is used to calibrate the external parameters of each ring array industrial camera to obtain the rotation matrix and translation vector of each camera relative to a unified workpiece coordinate system.

[0016] In this embodiment, the projection model of the ring array area array industrial camera is represented as follows: Let the three-dimensional points in the workpiece coordinate system be: (4) Then the pixel coordinates of a 3D point in the workpiece coordinate system in the ring array industrial camera satisfy: (5) In the formula: Represents a three-dimensional point in the workpiece coordinate system used for projection by a ring-array industrial camera. Represents the coordinate components of a three-dimensional point in the workpiece coordinate system; Indicates the scale factor; All represent the pixel coordinates of multi-view surface images acquired by a ring-array industrial camera; Represents the camera intrinsic parameter matrix; This represents the rotation matrix from the camera to the workpiece coordinate system; This represents the translation vector from the camera to the workpiece coordinate system.

[0017] The distortion model of the ring array area scan industrial camera in this embodiment is represented as follows: Let the ideal normalized image coordinates corresponding to the multi-view surface images acquired by the ring array area array industrial camera be... The coordinates after radial distortion correction are : (6) (7) (8) In the formula: All are ideal normalized image coordinates corresponding to multi-view surface images acquired by a ring array industrial camera. All images are distortion-normalized image coordinates corresponding to multi-view surface images acquired by a ring array industrial camera. All are radial distortion coefficients; All are tangential distortion coefficients; To normalize the image radius; In this embodiment, the contour scanning data of the weld area is used to perform spatial pose calibration on the line laser contour sensor to obtain the installation pose parameters of the line laser contour sensor relative to a unified workpiece coordinate system.

[0018] Specifically, the triangulation model of the line laser profile sensor is represented as follows: Let the normalized direction vector corresponding to the internal imaging of the line laser profilometry sensor be: (9) The laser plane equation is: (10) The three-dimensional contour points in the coordinate system of the line laser contour sensor are: (11) in: (12) therefore: (13) Then transform to the workpiece coordinate system: (14) In the formula: This indicates the direction of the ray received by the area array camera inside the line laser profile sensor. All represent the normalized coordinates of the image received from the area array camera inside the line laser profile sensor; Represents the laser plane normal vector; For transpose; Represents three-dimensional contour points in the coordinate system of a line laser contour sensor; Represents the laser plane constant term; This represents the scaling factor along the ray direction from the imaging center to the intersection point of the laser plane; Represents the three-dimensional contour points in the workpiece coordinate system; This represents the rotation matrix from the line laser profile sensor coordinate system to the workpiece coordinate system; This represents the translation vector from the line laser profile sensor coordinate system to the workpiece coordinate system.

[0019] Specifically, after completing the joint calibration between the ring array industrial camera and the line laser contour sensor, a unified mapping relationship is established between the image pixel coordinate system, the line laser contour sensor coordinate system, and the cylindrical shell workpiece coordinate system. Based on the calibration parameters, distortion correction is performed on the multi-view original surface image data, coordinate normalization is performed on the contour scanning data, and the corrected multi-view surface image data and three-dimensional contour point data in the workpiece coordinate system are output.

[0020] S3: Based on the three-dimensional points in the workpiece coordinate system, obtain the enhanced image and high-reflection area mask after unfolding the curved surface of the cylindrical shell; Specifically, firstly, the surface distortion of the cylindrical shell is corrected based on the corrected multi-view surface image data, resulting in a distorted image. Then, the distorted image is unfolded to create a continuous band-like region for the weld area. The unfolded image is then subjected to bilateral filtering, morphological opening, and contrast-limited adaptive histogram equalization to reduce noise and enhance the local contrast between the weld area and the base material area. Finally, high-grayscale reflection areas are detected, and reflection suppression is achieved using neighborhood mean filling or local interpolation to obtain the enhanced image after surface unfolding. and high reflectivity area mask This enables the preprocessing of surface unfolding, image enhancement, and reflection suppression using corrected multi-view surface image data.

[0021] S31: Obtain the corresponding cylinder surface unfolded coordinates based on the three-dimensional points in the workpiece coordinate system: Let the three-dimensional point in the workpiece coordinate system be... Then the corresponding cylindrical surface unfolded coordinates It is expressed as follows: (15) (16) In the formula: This represents the coordinates of the arc length of the cylindrical shell surface after it has been unfolded along the circumference. Represents the coordinates of the cylindrical shell surface along the axial direction; This indicates the radius of the cylindrical shell.

[0022] S32: Obtain the filtering result based on the expanded coordinates of the cylindrical surface corresponding to the three-dimensional points in the workpiece coordinate system; Let the unfolded image be The bilateral filtering result is ,but: (17) in: (18) S33: Based on the filtering results, perform morphological opening operations to obtain an enhanced image after unfolding the curved surface of the cylindrical shell, as shown below: (19) S34: Based on the enhanced image after unfolding the curved surface of the cylindrical shell, obtain the high-reflectivity area mask using the following formula: (20) In the formula: This indicates the result of bilateral filtering; The unfolded coordinates in the enhanced image after surface unfolding The bilateral filter normalization weight coefficients at the location are used to normalize the spatial distance weights and gray-level similarity weights of each sampling point in the filter neighborhood; Both represent the coordinates of the sampling point within the neighborhood; Indicates the filtering neighborhood; Sampling points in the filtered neighborhood of the enhanced image after surface unfolding The grayscale value at that location; Indicates the standard deviation of the spatial domain; Indicates an expanded image; This represents the standard deviation of the grayscale range; This represents the result of the opening operation, which is the enhanced image after the curved surface of the cylindrical shell is unfolded; This represents the erosion operation; This represents the expansion operation; Represents a structural element; Indicates a mask for highly reflective areas; This indicates the high reflectivity threshold.

[0023] Masking in high-reflection areas middle: Indicates the expanded coordinates This area belongs to a high reflectivity region; Indicates the expanded coordinates This area is not a high-reflectivity region.

[0024] S4: Based on the enhanced image after surface unfolding, generate weld candidate regions to obtain the weld candidate region constraint mask, the candidate region center axis initial value prior map, and the length continuity evaluation prior map; so as to realize the generation of weld candidate regions using the enhanced image after surface unfolding. In this embodiment, based on the enhanced image after surface unfolding The Otsu algorithm is used for adaptive threshold segmentation to extract gray-level abnormal regions, resulting in an initial binary candidate mask. For the initial binary candidate mask Morphological closing operations and small region removal are performed to connect discontinuous weld seam regions and remove isolated noise points, resulting in a corrected candidate mask. For the corrected candidate mask Connectivity labeling is performed to obtain multiple candidate connected regions. For each candidate connected region Calculate the area Candidate connected region center point Main direction of the region Main direction deviation and the projected length along the principal direction Based on area threshold Directional deviation threshold and length threshold The candidate connected regions are filtered to obtain the retained weld candidate regions. Based on the retained weld candidate area Generate a constraint mask for the candidate weld region. For the reserved weld candidate areas Extracting the central axis Based on this, an initial prior map of the central axis of the candidate region is generated. For the reserved weld candidate areas Perform a length continuity evaluation and calculate a length continuity score. Based on this, a priori graph for evaluating length continuity is generated. .

[0025] Among them, the length continuity evaluation result is the evaluation result of whether a certain weld candidate area is "continuous, stable and like a real weld" along the weld extension direction.

[0026] Preferably, S4 includes: S41: Based on the enhanced image after surface unfolding The Otsu algorithm is used for adaptive threshold segmentation to extract gray-level abnormal regions and obtain initial binary candidate masks. ; First, the formula used to obtain the optimal threshold for adaptive thresholding in the Otsu algorithm is as follows: (twenty one) (twenty two) in: Indicates the optimal segmentation threshold; The segmentation threshold; Indicates the segmentation threshold The corresponding inter-class variance; Indicates the segmentation threshold Probability of lower foreground pixels; Segmentation represents threshold Background pixel probability; Indicates the segmentation threshold Mean gray level of the lower foreground; Indicates the segmentation threshold Mean grayscale value of the background; Then, the initial binary candidate mask is obtained as follows: (twenty three) In the formula: Indicates the initial binary candidate mask; This represents an enhanced image after the curved surface of the cylindrical shell has been unfolded. in, Indicates the expanded coordinates This location belongs to the initial candidate foreground region obtained through threshold segmentation; Indicates the expanded coordinates The location does not belong to the initial candidate foreground region. S42: Obtain the morphologically corrected candidate mask based on the initial binary candidate mask; (twenty four) In the formula: Represents the candidate mask after morphological correction; This indicates a small area removal operation; This represents the expansion operation; This represents the erosion operation; Represents morphological structural elements used to connect discontinuous regions; This represents the threshold number of pixels to be removed from a small region.

[0027] S43: Based on the morphologically corrected candidate mask, obtain the candidate connected regions to realize the candidate connected region calculation; First of all Connectivity components are labeled as follows: (25) in: Indicates the first One candidate connected region; Indexes for candidate connected regions; Indicates the total number of candidate connected regions; This indicates the operation of connecting component labeling; Furthermore, the first The candidate connected regions are represented as follows: (26) in: This indicates the results of connected component labeling; Represents the two-dimensional coordinates of the unfolded surface of the cylindrical shell. Belongs to the first One candidate connected region; S44: Obtain the area of ​​the candidate connected region using the following formula: (27) in: Indicates the first The area of ​​each candidate connected region; S45: Obtain the center point of the candidate connected region based on its area, using the following formula: (28) in: Indicates the first The center point of each candidate connected region; S46: Based on the area and center point of the candidate connected region, obtain the coordinate covariance matrix of the candidate connected region to obtain the main direction of the region; The formula used to obtain the coordinate covariance matrix of candidate connected regions is as follows: (29) Then to Perform eigenvalue decomposition and take the unit eigenvector corresponding to the largest eigenvalue as the principal direction of the region: (30) in: Indicates the first The coordinate covariance matrix of each candidate connected region; Indicates the first The principal direction unit vectors of the candidate connected regions; Represents the covariance matrix The unit eigenvector corresponding to the largest eigenvalue.

[0028] S47: Based on the main direction of the region, obtain the angle deviation between the main direction of the candidate connected region and the desired direction of the weld, using the following formula: Let the unit vector of the desired weld direction be... Then the first The principal directional deviation of each candidate connected region is: (31) in: Indicates the first The angular deviation between the main direction of each candidate connected region and the desired direction of the weld; is the unit vector representing the desired direction of the weld in the unfolded image; It is the absolute value; It represents the absolute value of the cosine of the angle between two direction vectors.

[0029] S48: Based on the main direction of the region, obtain the projection length of the candidate connected region along the main direction, using the following formula: (32) in: Indicates the first The projection length of each candidate connected region along the main direction; S49: Based on the area of ​​the candidate connected region and the projection length of the candidate connected region along the main direction, the retained weld candidate region is obtained, using the following formula: (33) in: This indicates the number of items retained after filtering. One candidate area for weld seam; This indicates an empty set, meaning the candidate connected region has been eliminated. Indicates the area threshold; Indicates the directional deviation threshold; Indicates the length threshold; S410: Based on the retained weld candidate regions, obtain the weld candidate region constraint mask using the following formula: (34) in: This represents the constraint mask for the candidate weld region. Specifically, Indicates the expanded coordinates This area belongs to the candidate weld area that has been selected and retained. Indicates the expanded coordinates This area is not within the candidate area for weld seam.

[0030] S411: Based on the retained candidate weld areas, obtain the center axis of the candidate areas using the following formula: (35) in: Indicates the center axis of the area to be reserved for weld candidates; This indicates a skeleton extraction operation; S412: Obtain the initial value prior map of the candidate region's center axis based on the candidate region's center axis. The formula used is as follows: (36) in: This represents the prior diagram of the initial value of the central axis of the candidate region. Specifically, Indicates the expanded coordinates The location is situated on the central axis of the candidate region; Indicates the expanded coordinates The location is not situated on the central axis of the candidate region. S413: Based on the projected length of the candidate connected region along the main direction, obtain the length continuity score of the candidate weld region to be retained. The formula used is as follows: (37) in: This indicates the score for the continuity of the length of the candidate weld area. This indicates the projected length of the candidate weld area along the main direction; Indicates the penalty coefficient for the fracture segment; This indicates the number of fracture segments along the main direction in the candidate weld area. S414: Based on the length continuity score of the candidate weld area, obtain the prior map for length continuity evaluation. The formula used is as follows: (38) in: Represents the prior graph for evaluating length continuity.

[0031] Specifically, Indicates the expanded coordinates Belongs to the When selecting a candidate area for a reserved weld, its value is the length continuity score of that area; Indicates the expanded coordinates It does not belong to any candidate area for retained welds, and the length continuity score of this area is 0.

[0032] S5: Based on the enhanced image after surface unfolding, the high-reflectivity area mask, the weld candidate area constraint mask, the candidate area central axis initial value prior map, and the length continuity evaluation prior map, the U-Net semantic segmentation network is used to obtain the multi-class probability output results at the unfolded coordinates in the enhanced image after surface unfolding, so as to obtain the weld body fine segmentation mask, the welding spatter area mask, and the suspected defect area mask. This embodiment obtains a fine segmentation mask for the weld body, a mask for the welding spatter area, a mask for the suspected defect area, and a pixel-level confidence distribution map based on the enhanced image after surface unfolding, the mask for the candidate weld area, the initial value of the central axis of the candidate area, and the length continuity evaluation results, thereby achieving constrained and precise segmentation of the weld area. Specifically, the surface is unfolded to enhance the image. As the main input; high-reflectivity area mask As an auxiliary input for reflective interference; weld candidate region constraint mask As spatial constraint input; initial value prior map of the candidate region's central axis. Prior graph for evaluating length continuity As geometric prior input, an improved U-Net semantic segmentation network is employed. The main input, glare interference auxiliary input, spatial constraint input, and geometric prior input are all fed into the improved U-Net semantic segmentation network to perform pixel-level classification of the weld area, welding spatter area, suspected defect area, and background area, achieving pixel-level accurate segmentation of the weld area. The improved U-Net semantic segmentation network outputs a weld body probability map, a welding spatter probability map, a suspected defect probability map, and a background probability map, and based on these, a fine-grained segmentation mask for the weld body is generated. Weld spatter area mask Mask of suspected defective areas and pixel-level defect confidence distribution map ， It is expressed as follows: (38) In the formula: Input tensors into the network; This indicates a channel splicing operation; This represents an enhanced image after the curved surface of the cylindrical shell has been unfolded. This represents the constraint mask for the candidate weld region. Indicates a mask for highly reflective areas; This represents the prior diagram of the initial value of the central axis of the candidate region. Represents the prior graph for evaluating length continuity.

[0033] Specifically, the output representation of the improved U-Net semantic segmentation network is as follows: (39) in: Represents the unfolded coordinates in the enhanced image after surface unfolding. The multi-class probability output results at the location; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability that it belongs to the weld body area; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability of belonging to the welding spatter area; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability that the area belongs to a suspected defective region; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability of belonging to the background region; The finely segmented mask representation of the weld body is as follows: (40) The welding spatter area mask is represented as follows: (41) The mask for the suspected defective region is represented as follows: (42) in: Indicates the first Fine segmentation mask of weld body from the perspective of secondary synchronous acquisition; Indicates the first Welding spatter area mask from the perspective of secondary synchronous acquisition; Indicates the first Mask of suspected defective areas from the perspective of secondary synchronous acquisition; Represents the unfolded coordinates in the enhanced image after surface unfolding. It is the maximum value among the four types of probabilities.

[0034] Specifically, for fine segmentation of the weld body mask : Indicates the expanded coordinates The area was identified as the weld body area; Indicates the expanded coordinates This area is not part of the weld body area. (For the weld spatter area mask) : Indicates the expanded coordinates The area was identified as a welding spatter zone; Indicates the expanded coordinates This area is not within the weld spatter zone. Masking for suspected defect areas. : Indicates the expanded coordinates The area was identified as a suspected defective region. Indicates the expanded coordinates This area does not belong to the suspected defect area.

[0035] S6: Finely segment the mask based on the weld body. The results of obtaining the two-dimensional geometric representation of the weld include the two-dimensional discrete points of the centerline, the arc length position at the centerline point, the two-dimensional local width at the centerline point, the left boundary point at the centerline point, and the right boundary point at the centerline point. In this embodiment, a fine-segmentation mask of the weld body is used to extract the boundary of the weld region, obtaining a two-dimensional weld boundary contour. Furthermore, the skeleton of the two-dimensional weld region is extracted and the centerline is fitted to obtain the two-dimensional weld centerline. The local width is then measured along the normal direction of the two-dimensional weld centerline to obtain the two-dimensional width distribution of the weld. The two-dimensional weld centerline is parameterized into arc-length coordinates, and the mileage position and lateral offset of the suspected defect region relative to the weld centerline are established, achieving a two-dimensional geometric representation of the weld. The two-dimensional weld centerline, the two-dimensional weld boundary contour, and the parameterized arc-length index of the weld are output.

[0036] The discrete points of the two-dimensional centerline are represented as follows: (43) The arc length parameter is defined as: (44) (45) The local width is defined as: (46) in: Indicates the first Two-dimensional centerline points; Indicates the first The circumferential arc length coordinates of a two-dimensional centerline point; Indicates the first The axial coordinates of a two-dimensional centerline point; This indicates the number of discrete points on the two-dimensional centerline; Index of discrete points on the two-dimensional centerline; Indicates the first The position of the arc length corresponding to each centerline point; Indicates the first The left boundary point at each centerline point; Indicates the first The right boundary point at the center line point; Indicates the first The two-dimensional local width at each centerline point; This represents the Euclidean distance norm.

[0037] S7: Finely divide the mask according to the weld body. The results of the two-dimensional geometric characterization of the weld and the three-dimensional contour points in the workpiece coordinate system The process involves 3D reconstruction and local datum modeling of the weld seam to obtain a set of 3D contour points, thereby establishing a 3D surface model of the weld seam and ultimately obtaining the 3D centerline points. Three-dimensional local cross-section, three-dimensional local normal geometric datum, weld reinforcement; Preferably, S7 includes: S71: Finely divide the mask according to the weld body. Obtain the two-dimensional weld area, defined as: (47) In the formula: Indicates the first Two-dimensional weld area under a single acquisition sequence or perspective; S72: Based on the two-dimensional weld area, obtain the set of three-dimensional contour points of the weld to obtain the mapping points of the left boundary point and the right boundary point at the center line point on the three-dimensional contour of the weld. Then, based on the left boundary point and the right boundary point at the center line point, obtain the mapping points of the three-dimensional left boundary point at the center line point and the three-dimensional right boundary point at the center line point on the three-dimensional contour of the weld. (48) in, (49) In the formula: Indicates the first The set of three-dimensional contour points of the weld obtained by screening under the perspective of the next synchronous acquisition; Indicates the first The first synchronous acquisition perspective Three-dimensional contour points of candidate connected regions; This represents the mapping function from the three-dimensional contour points of the workpiece coordinate system to the unfolded coordinates of the cylinder surface; Representing three-dimensional contour points The corresponding unfolded coordinates; S73: Based on the mapping points of the left boundary point at the centerline point on the three-dimensional contour of the weld and the mapping points of the right boundary point at the centerline point on the three-dimensional contour of the weld, the three-dimensional centerline point is obtained as follows: (50) In the formula: Indicates the first The mapping point of the three-dimensional left boundary point at the center line point on the three-dimensional contour of the weld; Indicates the first The mapping point of the three-dimensional right boundary point at the center line point on the three-dimensional contour of the weld; Indicates the first Three-dimensional centerline points at each centerline point; Based on the three-dimensional centerline point, obtain a three-dimensional local section to obtain the coordinates within the three-dimensional local section; The formula used to obtain the three-dimensional local cross-section is as follows: (51) In the formula: Represents any point within a three-dimensional local section; Indicates the first Tangential unit vector of the three-dimensional centerline at each centerline point; Indicates the first Three-dimensional local cross-section at a centerline point; S74: Obtain the three-dimensional local normal geometric datum based on the coordinates within the three-dimensional local section, as shown below: (52) S75: The weld reinforcement height is obtained based on the three-dimensional local normal geometric reference, as shown below: (53) In the formula: Represents the lateral coordinates within a three-dimensional local section; Indicates the normal geometric reference height in a three-dimensional local area; All represent the fitting coefficients of a three-dimensional local normal geometric reference; This indicates the actual observed weld section height; Indicates the first The weld excess height at each centerline point.

[0038] In this embodiment, based on the three-dimensional contour points in the workpiece coordinate system Fine segmentation mask of weld body This process maps the two-dimensional region of the weld seam onto three-dimensional contour point data. For each acquisition sequence or viewpoint, the three-dimensional contour points projected onto the two-dimensional region of the weld seam are retained, resulting in a set of three-dimensional contour points for the weld seam. Combining the two-dimensional centerline point Arc length parameter Boundary points and local width The process involves multi-view matching, point cloud fusion, and local surface reconstruction of the weld's 3D contour points to obtain a 3D surface model of the weld. Specifically, based on the correspondence between the left and right boundary points, a 3D left boundary point is constructed from the set of weld 3D contour points. and the three-dimensional right boundary point Based on this, a three-dimensional centerline point is established. .by Based on the reference, the tangential unit vector of the three-dimensional centerline at this location is... Using the normal vector, construct a local cross section. ; in a three-dimensional local section Internal fitting of the parent material reference curve This yields the local normal geometric reference; further calculations are then performed to determine the arc length position. weld reinforcement at the location .

[0039] S8: Based on the suspected defect area mask, the two-dimensional centerline discrete points, the arc length position at the centerline point, and the two-dimensional local width at the centerline point, determine the defect candidate area, and obtain the two-dimensional appearance anomaly intensity score of the defect candidate area based on the two-dimensional appearance characteristics of the defect candidate area; and obtain the arc length center position of the candidate area, and obtain the overlap ratio between the defect candidate area and the welding spatter area based on the welding spatter area mask; Preferably, the formula used to obtain the two-dimensional appearance anomaly intensity score of the defect candidate region is as follows: (54) in: (55) In the formula: Indicates the first Two-dimensional appearance anomaly intensity score for each defect candidate region; Indicates the first Normalized values ​​of the concavity degree of each defect candidate region contour; Indicates the first Normalized values ​​of gray-level anomaly degree in candidate defect regions; Indicates the first Normalized values ​​for the anomaly degree of area of ​​each defect candidate region; Indicates the first Normalized values ​​of edge gradient changes in candidate defect regions; Indicates the first The normalized value of the lateral offset of each defect candidate region; Indicates the first Normalized values ​​of the length distribution of candidate defect regions; Both represent the weights of two-dimensional appearance features; The formula used to obtain the center position of the arc length of the defect candidate region is as follows: (56) In the formula: Indicates the first The center position of the arc length of each defect candidate region; Indicates the first One defect candidate region; express The weights; Index for defect candidate regions; The formula used to obtain the overlap ratio between the defect candidate area and the welding spatter area is as follows: (57) in: Indicates a mask for the welding spatter area; Indicates the first The number of pixels within each defect candidate region; Indicates the first The overlap ratio between the defect candidate area and the welding spatter area.

[0040] Specifically, use a mask for suspected defective areas. By performing connected component partitioning, multiple independent defect candidate regions are obtained. Then combine the two-dimensional centerline Arc length parameter and local width Two-dimensional appearance features are extracted for each defect candidate region. These features include contour indentation degree, mean grayscale value, area characteristics, edge gradient variation, lateral offset, and length distribution along the weld centerline. A weld spatter area mask is then used. Calculate the overlap ratio between each defect candidate region and the weld spatter region. .

[0041] Among them, the contour concavity feature characterizes the degree of local concavity of the defect candidate region boundary relative to the normal weld boundary. The grayscale anomaly feature characterizes the deviation between the mean grayscale value of the defect candidate region and the mean grayscale value of the neighboring normal weld region. The area feature characterizes the degree of anomaly of the defect candidate region area relative to the area of ​​normal noise spots. The edge gradient variation feature characterizes the drastic degree of grayscale variation at the edge of the defect candidate region. The lateral offset feature characterizes the degree of deviation of the defect candidate region relative to the weld centerline. The length distribution feature characterizes the extent of extension of the defect candidate region along the weld centerline direction.

[0042] S9: Determine the center position of the arc length relative to the defect candidate region. Arc length at the nearest centerline point To obtain the center position of the arc length relative to the defect candidate region. Corresponding three-dimensional local cross section , and the center position of the arc length of the defect candidate region The local depression depth corresponding to the three-dimensional local cross-section and the center position of the arc length of the defect candidate region. The corresponding local bulge height of the three-dimensional local section and the center position of the arc length of the defect candidate region. The corresponding three-dimensional local section's residual height deviation and the arc length center position of the defect candidate region. The average defect confidence of the corresponding three-dimensional local cross section is then used to determine whether the defect candidate region is a valid defect region. Furthermore, based on the overlap ratio between the defect candidate region and the welding spatter region, the consistency verification score of the defect candidate region after the introduction of welding spatter suppression is obtained.

[0043] Preferably, S9 includes: S91: Determine the center position of the arc length relative to the defect candidate region. The arc length position at the nearest centerline point is used to determine the arc length center position relative to the defect candidate region. The corresponding three-dimensional local section; Among them, the center position of the arc length relative to the defect candidate region is determined. The formula used to determine the arc length at the nearest centerline point is as follows: (58) In the formula: Indicates the position of the center of the arc length of the defect candidate region. The closest discrete arc length index; Indicates the first The position of the arc length corresponding to each centerline point; Represents the lateral coordinates within a three-dimensional local section; Then, the center position of the arc length of the defect candidate region The corresponding three-dimensional centerline points are: (59) Center position of the arc length of the defect candidate region The corresponding three-dimensional local section is: (60) In the formula: Center position of the arc length of the defect candidate region The corresponding three-dimensional local section; S92: Based on the center position of the arc length relative to the defect candidate region The corresponding three-dimensional local cross-section is used to obtain the center position of the arc length relative to the defect candidate region. The corresponding local depression depth of the three-dimensional local section is: (61) In the formula: Indicates the positive truncation operator; Indicates the first The local depression depth of each defect candidate region; Indicates the center position of the arc length relative to the defect candidate region. The corresponding lateral coordinates within the three-dimensional local section; Indicates the center position of the arc length relative to the defect candidate region. The corresponding three-dimensional local normal geometric reference height; Indicates the center position of the arc length relative to the defect candidate region. The corresponding actual observed weld section height; S93: Obtain the center position of the arc length of the defect candidate region. The corresponding local bulge height of the three-dimensional local section is: (62) In the formula: Indicates the first The local bulge height of each defect candidate region; The local thickness variation in the defect candidate region is: (63) In the formula: Indicates the first The local thickness variation of each defect candidate region; It is the absolute value; For the first The actual observed local thickness value at the lateral coordinate within the three-dimensional local section of each defect candidate region; For the first The reference thickness value at the lateral coordinate of the three-dimensional local section of each defect candidate region is determined by the local normal geometric datum or the workpiece design thickness. S94: Obtain the center position of the arc length of the defect candidate region. The corresponding residual height deviation of the three-dimensional local section is: (64) In the formula: Indicates the first The residual height deviation of each defect candidate region; Indicates the first The actual observed residual height of the corresponding location of each defect candidate region; Indicates the first weld excess height at each centerline point; S95: Obtain the center position of the arc length of the defect candidate region. The average defect confidence level of the corresponding three-dimensional local cross-section is: (65) In the formula: Indicates the first The average defect confidence level of each defect candidate region; For the first The number of pixels within each defect candidate region; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability that the area belongs to a suspected defective region; S96: Based on the overlap ratio between the defect candidate region and the weld spatter region, obtain the consistency verification score of the defect candidate region after introducing weld spatter suppression: First, obtain the basic conformity verification score before the introduction of weld spatter suppression, using the following formula: (66) in: (67) In the formula: This indicates the basic consistency verification score before the introduction of welding spatter suppression; Indicates the first Two-dimensional appearance anomaly intensity score for each defect candidate region; This represents the normalized threshold for the depth of a local depression. This represents the normalized threshold for the height of a local bulge. This represents the normalized threshold for local thickness variation; This represents the normalized threshold for the residual height deviation. All of these are weights in the basic consistency verification score; Then, based on the overlap ratio between the defect candidate region and the weld spatter region, the consistency verification score after introducing weld spatter suppression is obtained, using the following formula: (68) In the formula: Indicates the first The overlap ratio between the candidate defect region and the welding spatter region; This indicates the consistency verification score after introducing welding spatter suppression; This indicates the weight for suppressing welding spatter. Preferably, the formula used to determine whether a candidate defect region is a valid defect region is as follows: (69) In the formula: Indicates the first The final determination label for each defect candidate region; This indicates the consistency verification score threshold; This indicates the threshold for the overlap ratio of welding spatter; Specifically, in the final determination of the label Chinese: When When, it indicates the first One defect candidate region was determined to be a valid defect region; When, it indicates the first One defect candidate region was identified as a false defect region.

[0044] Specifically, the arc length center position of the defect candidate region is used. Find the arc length position at the centerline point closest to it. This allows us to obtain the three-dimensional centerline point at the corresponding centerline point. Three-dimensional local cross-section 3D Local Normal Geometric Reference and weld reinforcement ; and then in the corresponding three-dimensional local section Calculate the depth of the local depression. Local protrusion height Local thickness variation and residual height deviation Combining pixel-level defect confidence levels Calculate the average defect confidence score of the defect candidate region. Then, the two-dimensional appearance anomaly intensity score is combined. and splash overlap ratio Construct a consistency verification score that incorporates welding spatter suppression. Then, based on the consistency verification score and the splash overlap ratio, the candidate region is determined to be a valid defect or a false defect. Finally, the list of valid defects, defect type, defect size, defect severity, and defect spatial location are output.

[0045] Finally, output a list of valid defect areas, defect types, defect sizes, defect severity, and defect spatial locations.

[0046] A specific embodiment of the present invention is as follows: (I) Multi-view acquisition and calibration: Multiple industrial cameras are arranged in a circular array to acquire multi-view images, and a line laser scanner is used to acquire 3D data of the weld area; Zhang Zhengyou calibration method is used to obtain the camera's intrinsic and extrinsic parameters and perform distortion correction; further calibration of the camera and laser scanner is performed to achieve data fusion. Acquisition can be triggered synchronously by PLC, and the acquisition frequency can be set to 10fps to balance real-time performance and data integrity.

[0047] (ii) Image preprocessing: Surface distortion correction can be performed using perspective / geometric correction based on calibration parameters; grayscale conversion can be achieved using a weighted formula. Noise suppression employs bilateral filtering (core 5×5, =2.0, =30) + morphological opening operation (3×3, 1 iteration); contrast enhancement uses CLAHE (gain 2.5, tile 8×8, clipping threshold 0.01); reflection suppression can detect high grayscale spots and fill with neighborhood mean.

[0048] (III) Coarse Segmentation and Deep Learning Fine Segmentation: Coarse segmentation employs Otsu adaptive thresholding and Canny edge detection to filter continuous edges and determine the approximate range of the weld seam, forming a coarse mask to constrain the deep learning inference region. Deep learning segmentation uses an improved U-Net network: the encoder extracts multi-scale features, and the decoder introduces residual connections to preserve weld seam details; the loss function can use a combination of Dice loss and cross-entropy loss to alleviate the imbalance problem caused by the small foreground proportion; training uses data augmentation to improve generalization ability.

[0049] (iv) Three-dimensional morphological verification and defect screening: The finely segmented weld area is reconstructed in three dimensions and the morphological parameters such as the height and width are extracted; the two-dimensional defect detection can obtain candidate defects based on indicators such as contour depression, grayscale mean and area, and then map them to the three-dimensional model to verify their depth / thickness / height and other geometric conditions to eliminate false judgments, and finally output the defect list.

[0050] This embodiment enables pixel-level precise segmentation of the weld area under complex curved surface, reflective, and spatter interference conditions, and improves the accuracy of weld identification and the reliability of defect screening. First, using multi-view original surface image data of the weld area of ​​a cylindrical shell workpiece, an enhanced image and a high-reflectivity area mask are obtained after unfolding the curved surface of the cylindrical shell. Based on the enhanced image after surface unfolding, candidate weld areas are generated. A coarse segmentation mask is used to constrain the deep learning segmentation input, reducing false detections and improving real-time performance. Through deep learning, a fine segmentation mask for the weld body, a welding spatter area mask, and a suspected defect area mask are obtained, achieving pixel-level fine segmentation of the weld, distinguishing between the weld body, spatter, and defect areas. Simultaneously, a three-dimensional morphological verification mechanism is introduced, using a consistency verification score after introducing welding spatter suppression for defect judgment, significantly reducing two-dimensional false judgments and improving defect screening accuracy. Adaptable to curved surface distortion and reflective conditions, it possesses strong robustness and can achieve pixel-level precise weld identification and reliable defect screening.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatic identification and defect screening of weld seams in cylindrical shells, characterized in that, Includes the following steps: S1: Acquire multi-view raw surface image data and contour scan data of the weld area of ​​the cylindrical shell workpiece; S2: Based on the multi-view original surface image data of the weld area and the contour scan data of the weld area, respectively, obtain the three-dimensional points in the workpiece coordinate system and the three-dimensional contour points in the workpiece coordinate system; S3: Based on the three-dimensional points in the workpiece coordinate system, obtain the enhanced image and high-reflection area mask after unfolding the curved surface of the cylindrical shell; S4: Based on the enhanced image after surface unfolding, generate weld candidate regions to obtain the weld candidate region constraint mask, the candidate region center axis initial value prior map, and the length continuity evaluation prior map; S5: Based on the enhanced image after surface unfolding, the high-reflectivity area mask, the weld candidate area constraint mask, the candidate area central axis initial value prior map, and the length continuity evaluation prior map, the U-Net semantic segmentation network is used to obtain the multi-class probability output results at the unfolded coordinates in the enhanced image after surface unfolding, so as to obtain the weld body fine segmentation mask, the welding spatter area mask, and the suspected defect area mask. S6: Based on the fine segmentation mask of the weld body, obtain the result of the two-dimensional geometric representation of the weld, including the two-dimensional centerline discrete point, the arc length position at the centerline point, the two-dimensional local width at the centerline point, the left boundary point at the centerline point, and the right boundary point at the centerline point; S7: Based on the fine segmentation mask of the weld body, the result of the two-dimensional geometric representation of the weld, and the three-dimensional contour points in the workpiece coordinate system, obtain the three-dimensional contour point set of the weld to establish a three-dimensional surface model of the weld, and then obtain the three-dimensional centerline point, three-dimensional local section, three-dimensional local normal geometric reference, and weld reinforcement height. S8: Based on the suspected defect area mask, the two-dimensional centerline discrete points, the arc length position at the centerline point, and the two-dimensional local width at the centerline point, determine the defect candidate area to obtain the two-dimensional appearance anomaly intensity score of the defect candidate area; and obtain the arc length center position of the candidate area, and obtain the overlap ratio between the defect candidate area and the welding spatter area based on the welding spatter area mask; S9: Determine the arc length position at the centerline point closest to the arc length center position of the defect candidate region, and obtain the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the local concavity depth of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the local convexity height of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, the residual height deviation of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region, and the average defect confidence of the three-dimensional local cross-section corresponding to the arc length center position of the defect candidate region; then, based on the overlap ratio between the defect candidate region and the welding spatter region, obtain the consistency verification score of the defect candidate region after the introduction of welding spatter suppression, so as to determine whether the defect candidate region is a valid defect region.

2. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, S3 includes: S31: Based on the three-dimensional point in the workpiece coordinate system, obtain the unfolded coordinates of the cylinder surface corresponding to the three-dimensional point in the workpiece coordinate system, as shown below: In the formula: This represents the coordinates of the arc length of the cylindrical shell surface after it has been unfolded along the circumference. Represents the coordinates of the cylindrical shell surface along the axial direction; Indicates the radius of the cylindrical shell; All coordinates are three-dimensional point coordinates in the workpiece coordinate system; S32: Obtain the filtering result based on the expanded coordinates of the cylinder surface corresponding to the three-dimensional points in the workpiece coordinate system; the formula used is as follows: in: In the formula: This indicates the result of bilateral filtering; The unfolded coordinates in the enhanced image after surface unfolding The normalized weighting coefficients of the bilateral filter at the location; Both represent the coordinates of the sampling point within the neighborhood; Indicates the filtering neighborhood; Sampling points in the filtered neighborhood of the enhanced image after surface unfolding The grayscale value at that location; Indicates the standard deviation of the spatial domain; Indicates an expanded image; This represents the standard deviation of the grayscale range; S33: Based on the filtering results, obtain the enhanced image after unfolding the curved surface of the cylindrical shell, as shown below: In the formula: This represents the result of the opening operation, which is the enhanced image after the curved surface of the cylindrical shell is unfolded; This represents the erosion operation; This represents the expansion operation; Represents a structural element; S34: Based on the enhanced image after unfolding the curved surface of the cylindrical shell, obtain the high-reflectivity area mask using the following formula: In the formula: Indicates a mask for highly reflective areas; This indicates the high reflectivity threshold.

3. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, S4 includes: S41: Based on the enhanced image after surface unfolding Obtain the initial binary candidate mask. The formula used is as follows: In the formula: Indicates the initial binary candidate mask; This represents an enhanced image after the curved surface of the cylindrical shell has been unfolded. Indicates the optimal segmentation threshold; in: The segmentation threshold; Indicates the segmentation threshold The corresponding inter-class variance; Indicates the segmentation threshold Probability of lower foreground pixels; Indicates the segmentation threshold Background pixel probability; Indicates the segmentation threshold Mean gray level of the lower foreground; Indicates the segmentation threshold Mean grayscale value of the background; Indicates selection The segmentation threshold corresponding to the maximum value ; S42: Obtain the morphologically corrected candidate mask based on the initial binary candidate mask; In the formula: Represents the candidate mask after morphological correction; This indicates a small area removal operation; This represents the expansion operation; This represents the erosion operation; Represents morphological structural elements used to connect discontinuous regions; This represents the threshold for the number of pixels to be removed from a small region. S43: Obtain candidate connected regions based on the morphologically corrected candidate mask; First of all Connectivity components are labeled as follows: in: Indicates the first One candidate connected region; Indexes for candidate connected regions; Indicates the total number of candidate connected regions; This indicates the operation of connecting component labeling; The first The candidate connected regions are represented as follows: in: This indicates the results of connected component labeling; Represents the two-dimensional coordinates of the unfolded surface of the cylindrical shell. Belongs to the first One candidate connected region; S44: Obtain the area of ​​the candidate connected region using the following formula: in: Indicates the first The area of ​​each candidate connected region; S45: Obtain the center point of the candidate connected region based on its area, using the following formula: in: Indicates the first The center point of each candidate connected region; S46: Based on the area and center point of the candidate connected region, obtain the coordinate covariance matrix of the candidate connected region to obtain the main direction of the region; The formula used to obtain the coordinate covariance matrix of candidate connected regions is as follows: Then, based on the coordinate covariance matrix of the candidate connected regions, the main direction of the region is obtained: in: Indicates the first The coordinate covariance matrix of each candidate connected region; Indicates the first The principal direction unit vectors of the candidate connected regions; Represents the covariance matrix The unit eigenvector corresponding to the largest eigenvalue; S47: Based on the main direction of the region, obtain the angle deviation between the main direction of the candidate connected region and the desired direction of the weld, using the following formula: in: Indicates the first The angular deviation between the main direction of each candidate connected region and the desired direction of the weld; is the unit vector representing the desired direction of the weld in the unfolded image; It is the absolute value; It represents the absolute value of the cosine of the angle between two direction vectors; S48: Based on the main direction of the region, obtain the projection length of the candidate connected region along the main direction, using the following formula: in: Indicates the first The projection length of each candidate connected region along the main direction; S49: Based on the area of ​​the candidate connected region and the projection length of the candidate connected region along the main direction, the retained weld candidate region is obtained, using the following formula: in: This indicates the candidate weld areas that have been selected and retained. This indicates an empty set, meaning the candidate connected region has been eliminated. Indicates the area threshold; Indicates the directional deviation threshold; Indicates the length threshold; S410: Based on the retained weld candidate regions, obtain the weld candidate region constraint mask using the following formula: in: This represents the constraint mask for the candidate weld region. S411: Based on the retained candidate weld areas, obtain the center axis of the candidate areas using the following formula: in: Indicates the center axis of the area to be reserved for weld candidates; This indicates a skeleton extraction operation; S412: Obtain the initial value prior map of the candidate region's center axis based on the candidate region's center axis. The formula used is as follows: in: This represents the prior diagram of the initial value of the central axis of the candidate region. S413: Based on the projected length of the candidate connected region along the main direction, obtain the length continuity score of the candidate weld region to be retained. The formula used is as follows: in: This indicates the score for the continuity of the length of the candidate weld area. This indicates the projected length of the candidate weld area along the main direction; Indicates the penalty coefficient for the fracture segment; This indicates the number of fracture segments along the main direction in the candidate weld area. S414: Based on the length continuity score of the candidate weld area, obtain the prior map for length continuity evaluation. The formula used is as follows: in: Represents the prior graph for evaluating length continuity.

4. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, The multi-class probability output results at the unfolded coordinates in the enhanced image after surface unfolding include the probability that the unfolded coordinates in the enhanced image after surface unfolding belong to the weld body region, the probability that the unfolded coordinates in the enhanced image after surface unfolding belong to the welding spatter region, and the probability that the unfolded coordinates in the enhanced image after surface unfolding belong to the suspected defect region. The finely segmented mask representation of the weld body is as follows: The welding spatter area mask is represented as follows: The mask for the suspected defective region is represented as follows: in: Indicates the first Fine segmentation mask of weld body from the perspective of secondary synchronous acquisition; Indicates the first Welding spatter area mask from the perspective of secondary synchronous acquisition; Indicates the first Mask of suspected defective areas from the perspective of secondary synchronous acquisition; Represents the unfolded coordinates in the enhanced image after surface unfolding. The maximum value among the four types of probabilities; Represents the unfolded coordinates in the enhanced image after surface unfolding. The multi-class probability output results at the location; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability that it belongs to the weld body area; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability of belonging to the welding spatter area; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability of belonging to a suspected defect area.

5. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, The S7 includes: S71: Based on the fine segmentation mask of the weld body, obtain the two-dimensional weld region, defined as: In the formula: Indicates the first Two-dimensional weld area under a single acquisition sequence or perspective; Indicates the first Fine segmentation mask of weld body from the perspective of secondary synchronous acquisition; S72: Based on the two-dimensional weld area, obtain the set of three-dimensional contour points of the weld to obtain the mapping points of the left boundary point and the right boundary point at the center line point on the three-dimensional contour of the weld. Then, based on the left boundary point and the right boundary point at the center line point, obtain the mapping points of the three-dimensional left boundary point at the center line point and the three-dimensional right boundary point at the center line point on the three-dimensional contour of the weld. in, In the formula: Indicates the first The set of three-dimensional contour points of the weld obtained by screening under the perspective of the next synchronous acquisition; Indicates the first The first synchronous acquisition perspective Three-dimensional contour points of candidate connected regions; This represents the mapping function from the three-dimensional contour points of the workpiece coordinate system to the unfolded coordinates of the cylinder surface; Representing three-dimensional contour points The corresponding unfolded coordinates; S73: Based on the mapping points of the left boundary point at the centerline point on the three-dimensional contour of the weld and the mapping points of the right boundary point at the centerline point on the three-dimensional contour of the weld, the three-dimensional centerline point is obtained as follows: In the formula: Indicates the first The mapping point of the three-dimensional left boundary point at the center line point on the three-dimensional contour of the weld; Indicates the first The mapping point of the three-dimensional right boundary point at the center line point on the three-dimensional contour of the weld; Indicates the first Three-dimensional centerline points at each centerline point; Based on the three-dimensional centerline point, obtain a three-dimensional local section to obtain the coordinates within the three-dimensional local section; The formula used to obtain the three-dimensional local cross-section is as follows: In the formula: Represents any point within a three-dimensional local section; Indicates the first Tangential unit vector of the three-dimensional centerline at each centerline point; Indicates the first Three-dimensional local cross-section at a centerline point; S74: Obtain the three-dimensional local normal geometric datum based on the coordinates within the three-dimensional local section, as shown below: S75: The weld reinforcement height is obtained based on the three-dimensional local normal geometric reference, as shown below: In the formula: Represents the lateral coordinates within a three-dimensional local section; Indicates the normal geometric reference height in a three-dimensional local area; All represent the fitting coefficients of a three-dimensional local normal geometric reference; This indicates the actual observed weld section height; Indicates the first The weld excess height at each centerline point.

6. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, S8 includes: The formula used to obtain the two-dimensional appearance anomaly intensity score of the defect candidate region is as follows: in: In the formula: Indicates the first Two-dimensional appearance anomaly intensity score for each defect candidate region; Indicates the first Normalized values ​​of the concavity degree of each defect candidate region contour; Indicates the first Normalized values ​​of gray-level anomaly degree in candidate defect regions; Indicates the first Normalized values ​​for the anomaly degree of area of ​​each defect candidate region; Indicates the first Normalized values ​​of edge gradient changes in candidate defect regions; Indicates the first The normalized value of the lateral offset of each defect candidate region; Indicates the first Normalized values ​​of the length distribution of candidate defect regions; Represents the weights of two-dimensional appearance features; The formula used to obtain the center position of the arc length of the defect candidate region is as follows: In the formula: Indicates the first The center position of the arc length of each defect candidate region; Indicates the first One defect candidate region; express The weights; Index for defect candidate regions; The formula used to obtain the overlap ratio between the defect candidate area and the welding spatter area is as follows: in: Indicates a mask for the welding spatter area; Indicates the first The number of pixels within each defect candidate region; Indicates the first The overlap ratio between the defect candidate area and the welding spatter area.

7. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, S9 includes: S91: Determine the position of the arc length at the centerline point closest to the center position of the arc length of the defect candidate region, so as to determine the three-dimensional local section corresponding to the center position of the arc length of the defect candidate region; The formula used to determine the arc length position of the centerline point closest to the center position of the defect candidate region is as follows: In the formula: This represents the discrete arc length index that is closest to the center of the arc length of the defect candidate region; Indicates the first The position of the arc length corresponding to each centerline point; Represents the lateral coordinates within a three-dimensional local section; Then, the center position of the arc length of the defect candidate region The corresponding three-dimensional centerline points are: The three-dimensional local section corresponding to the center position of the arc length of the defect candidate region is: In the formula: A three-dimensional local cross-section corresponding to the center position of the arc length of the defect candidate region; S92: Based on the three-dimensional local cross-section corresponding to the center position of the arc length of the defect candidate region, obtain the local depression depth of the three-dimensional local cross-section corresponding to the center position of the arc length of the defect candidate region as follows: In the formula: Indicates the positive truncation operator; Indicates the first The local depression depth of each defect candidate region; Indicates the center position of the arc length relative to the defect candidate region. The corresponding lateral coordinates within the three-dimensional local section; Indicates the center position of the arc length relative to the defect candidate region. The corresponding three-dimensional local normal geometric reference height; Indicates the center position of the arc length relative to the defect candidate region. The corresponding actual observed weld section height; S93: Obtain the local bulge height of the three-dimensional local section corresponding to the center position of the arc length of the defect candidate region: In the formula: Indicates the first The local bulge height of each defect candidate region; The local thickness variation in the defect candidate region is: In the formula: Indicates the first The local thickness variation of each defect candidate region; For the first The actual observed local thickness value at the lateral coordinate within the three-dimensional local section of each defect candidate region; For the first The reference thickness value at the lateral coordinate of the three-dimensional local section of each defect candidate region is determined by the local normal geometric datum or the workpiece design thickness. It is the absolute value; S94: Obtain the residual height deviation of the three-dimensional local section corresponding to the center position of the arc length of the defect candidate region: In the formula: Indicates the first The residual height deviation of each defect candidate region; Indicates the first The actual observed residual height of the corresponding location of each defect candidate region; Indicates the first weld excess height at each centerline point; S95: The average defect confidence level of the three-dimensional local cross-section corresponding to the center position of the arc length of the defect candidate region is obtained as follows: In the formula: Indicates the first The average defect confidence level of each defect candidate region; For the first The number of pixels within each defect candidate region; Represents the unfolded coordinates in the enhanced image after surface unfolding. The probability that the area belongs to a suspected defective region; S96: Based on the overlap ratio between the defect candidate region and the weld spatter region, obtain the consistency verification score of the defect candidate region after introducing weld spatter suppression: First, obtain the basic conformity verification score before the introduction of weld spatter suppression, using the following formula: in: In the formula: This indicates the basic consistency verification score before the introduction of welding spatter suppression; Indicates the first Two-dimensional appearance anomaly intensity score for each defect candidate region; This represents the normalized threshold for the depth of a local depression. This represents the normalized threshold for the height of a local bulge. This represents the normalized threshold for local thickness variation; This represents the normalized threshold for the residual height deviation. All of these are weights in the basic consistency verification score.

8. The method for automatic identification and defect screening of weld seams in cylindrical shells according to claim 1, characterized in that, The formula used to determine whether a candidate defect region is a valid defect region is as follows: In the formula: Indicates the first The final determination label for each defect candidate region; This indicates the consistency verification score threshold; This indicates the threshold for the proportion of welding spatter overlap.

Images