1. A wheel welding seam surface defect detection method based on computer vision is characterized by comprising the following steps:

obtaining a wheel weld image, performing semantic segmentation processing on the wheel weld image, and identifying a weld connected domain in the wheel weld image;

obtaining rough weld edges according to the weld connected domain, and performing edge refining treatment on the rough weld edges to obtain fine weld edges;

calculating the pit rate corresponding to each pixel between the edge of the fine weld joint and the extending edge according to the gray value of each pixel between the edge of the fine weld joint and the extending edge;

calculating the undercut rate corresponding to each pixel point on the inner line of the weld joint connected domain according to the pit rate;

and marking the undercut defect area on the weld seam according to the undercut rate.

2. The computer vision-based wheel weld surface defect detection method according to claim 1, wherein the method for performing edge refinement on the rough weld edge to obtain the fine weld edge comprises the following steps:

extending each edge pixel point corresponding to the rough welding seam edge along the direction vertical to the central line of the welding seam communication domain to the left for a set number of pixel points to obtain a left extending pixel point corresponding to each edge pixel point corresponding to the rough welding seam edge;

extending each edge pixel point corresponding to the rough weld edge rightwards along the direction perpendicular to the central line of the weld connected domain for a set number of pixel points to obtain a right extending pixel point corresponding to each edge pixel point corresponding to the rough weld edge;

taking a direction vertical to the central line of the welding seam connected domain as a judgment direction, and constructing a pixel gray value sequence taking a pixel point on the central line of the welding seam connected domain as a starting point and an extended pixel point as a terminal point;

and constructing a pixel gray variation curve corresponding to the pixel gray value sequence, and correcting each edge pixel point corresponding to the rough weld edge according to the pixel gray variation curve to obtain a fine weld edge.

3. The computer vision based wheel weld surface defect detection method according to claim 1, wherein the method of calculating the pit rate for each pixel between the fine weld edge to the extended edge comprises:

taking the direction of the central line of the vertical connected domain as a judgment direction, and constructing a gray sequence by taking the first pixel point outside the edge pixel point as a starting point and the extension pixel point as an end point;

and calculating the pit rate corresponding to each pixel according to the gray value of each pixel point in the gray sequence, wherein the pit rate and the gray value form a negative correlation relationship.

4. The computer vision-based wheel weld surface defect detection method according to claim 1, wherein the method for calculating the undercut rate corresponding to each pixel point on the line in the weld connected domain according to the crater rate comprises the following steps:

calculating the discontinuous undercut rate and the continuous undercut rate corresponding to each pixel point on the middle line of the weld joint communication domain according to the pit rate corresponding to each pixel between the edge of the fine weld joint and the extended edge;

and judging the sizes of the discontinuous undercut rate and the continuous undercut rate, and taking the larger value of the discontinuous undercut rate and the continuous undercut rate as the undercut rate corresponding to each pixel point on the line in the weld joint communication domain.

5. The computer vision-based wheel weld surface defect detection method according to claim 4, wherein the method for calculating the discontinuous undercut rate corresponding to each pixel point on the line in the weld joint connection domain comprises the following steps:

counting pixel points with the maximum pit rate in the gray sequence corresponding to each pixel point on the inner line of the welding joint connected domain, and calculating the distance between the pixel points with the maximum pit rate and the corresponding pixel points on the inner line of the welding joint connected domain;

sequentially numbering each pixel point on the middle line of the welding seam communicating domain, and establishing a middle line-distance curve by taking the number as a horizontal axis and the distance as a vertical axis;

identifying wave crests and wave troughs on the central line-distance curve, and dividing the central line-distance curve into unimodal regions, wherein each unimodal region corresponds to one wave trough-wave crest-wave trough region;

and performing Gaussian fitting on each unimodal region to obtain a fitting curve, and calculating the discontinuous undercut rate corresponding to each pixel point on the middle line of the welding joint communication domain according to the fitting curve.

6. The computer vision-based wheel weld surface defect detection method according to claim 4, wherein the method for calculating the continuous undercut rate corresponding to each pixel point on the line in the weld joint connection domain comprises the following steps:

counting the pixel points with the maximum pit rate in the gray sequence corresponding to each pixel point on the welding seam connected domain line to obtain the sequence numbers of the pixel points with the maximum pit rate in the corresponding gray sequence;

sequentially numbering each pixel point on the welding seam communication center line, taking the number corresponding to each pixel point on the welding seam communication center line as a horizontal axis, taking the sequence number as a vertical axis, and establishing an edge-distance curve;

and calculating the average value of the distance difference between each pixel point on the edge-distance curve and other pixel points on the left side and the right side, and calculating the continuous undercut rate corresponding to each pixel point on the middle line of the welding joint connected domain according to the average value of the distance difference.

Background

At present, the detection of the welding seam defects based on computer vision is usually aimed at the defects of gaps, air holes, slag inclusion and the like on the surface of the welding seam, for example, black round holes or cracks exist on a silver welding seam, and the image airspace characteristics corresponding to the defects are obvious and definite and are easy to detect.

For undercut defects, the color of the background master batch where the undercut defects are located is variable, and the undercut defects are difficult to extract by giving a proper segmentation threshold limit; and the occlusion defect has no corresponding relatively clear characteristics, and is not easy to detect.

Disclosure of Invention

In order to realize accurate detection of occlusion defects, the invention provides a technical scheme of a wheel weld surface defect detection method based on computer vision, which comprises the following steps:

obtaining a wheel weld image, performing semantic segmentation processing on the wheel weld image, and identifying a weld connected domain in the wheel weld image;

obtaining rough weld edges according to the weld connected domain, and performing edge refining treatment on the rough weld edges to obtain fine weld edges;

calculating the pit rate corresponding to each pixel between the edge of the fine weld joint and the extending edge according to the gray value of each pixel between the edge of the fine weld joint and the extending edge;

calculating the undercut rate corresponding to each pixel point on the inner line of the weld joint connected domain according to the pit rate;

and marking the undercut defect area on the weld seam according to the undercut rate.

The detection method has the beneficial effects that: according to the method, the fine weld joint edge is obtained based on the wheel weld joint image, the undercut rate corresponding to each pixel point on the inner line of the weld joint communication domain is calculated according to the pit rate corresponding to each pixel between the fine weld joint edge and the extension edge, and the accurate identification of the undercut defect area on the weld joint is realized. The invention provides an automatic detection method for the surface defects of wheel welding seams, which solves the problem that the existing method cannot accurately detect the undercut defects of the welding seams.

Further, the method for performing edge refining processing on the rough weld edge to obtain the fine weld edge includes:

extending each edge pixel point corresponding to the rough welding seam edge along the direction vertical to the central line of the welding seam communication domain to the left for a set number of pixel points to obtain a left extending pixel point corresponding to each edge pixel point corresponding to the rough welding seam edge;

extending each edge pixel point corresponding to the rough weld edge rightwards along the direction perpendicular to the central line of the weld connected domain for a set number of pixel points to obtain a right extending pixel point corresponding to each edge pixel point corresponding to the rough weld edge;

taking a direction vertical to the central line of the welding seam connected domain as a judgment direction, and constructing a pixel gray value sequence taking a pixel point on the central line of the welding seam connected domain as a starting point and an extended pixel point as a terminal point;

and constructing a pixel gray variation curve corresponding to the pixel gray value sequence, and correcting each edge pixel point corresponding to the rough weld edge according to the pixel gray variation curve to obtain a fine weld edge.

Further, the method for calculating the pit rate corresponding to each pixel between the fine weld edge and the extended edge comprises the following steps:

taking the direction of the central line of the vertical connected domain as a judgment direction, and constructing a gray sequence by taking the first pixel point outside the edge pixel point as a starting point and the extension pixel point as an end point;

and calculating the pit rate corresponding to each pixel according to the gray value of each pixel point in the gray sequence, wherein the pit rate and the gray value form a negative correlation relationship.

Further, the method for calculating the undercut rate corresponding to each pixel point on the line in the weld joint connected domain according to the pit rate comprises the following steps:

calculating the discontinuous undercut rate and the continuous undercut rate corresponding to each pixel point on the middle line of the weld joint communication domain according to the pit rate corresponding to each pixel between the edge of the fine weld joint and the extended edge;

and judging the sizes of the discontinuous undercut rate and the continuous undercut rate, and taking the larger value of the discontinuous undercut rate and the continuous undercut rate as the undercut rate corresponding to each pixel point on the line in the weld joint communication domain.

Further, the method for calculating the discontinuous undercut rate corresponding to each pixel point on the line in the weld joint connected domain comprises the following steps:

counting pixel points with the maximum pit rate in the gray sequence corresponding to each pixel point on the inner line of the welding joint connected domain, and calculating the distance between the pixel points with the maximum pit rate and the corresponding pixel points on the inner line of the welding joint connected domain;

sequentially numbering each pixel point on the middle line of the welding seam communicating domain, and establishing a middle line-distance curve by taking the number as a horizontal axis and the distance as a vertical axis;

identifying wave crests and wave troughs on the central line-distance curve, and dividing the central line-distance curve into unimodal regions, wherein each unimodal region corresponds to one wave trough-wave crest-wave trough region;

and performing Gaussian fitting on each unimodal region to obtain a fitting curve, and calculating the discontinuous undercut rate corresponding to each pixel point on the middle line of the welding joint communication domain according to the fitting curve.

Further, the method for calculating the continuous undercut rate corresponding to each pixel point on the line in the weld joint connected domain comprises the following steps:

counting the pixel points with the maximum pit rate in the gray sequence corresponding to each pixel point on the welding seam connected domain line to obtain the sequence numbers of the pixel points with the maximum pit rate in the corresponding gray sequence;

sequentially numbering each pixel point on the welding seam communication center line, taking the number corresponding to each pixel point on the welding seam communication center line as a horizontal axis, taking the sequence number as a vertical axis, and establishing an edge-distance curve;

and calculating the average value of the distance difference between each pixel point on the edge-distance curve and other pixel points on the left side and the right side, and calculating the continuous undercut rate corresponding to each pixel point on the middle line of the welding joint connected domain according to the average value of the distance difference.

Drawings

FIG. 1 is a schematic view of a weld undercut defect of the present invention;

FIG. 2 is a flow chart of a computer vision based wheel weld surface defect detection method of the present invention;

FIG. 3 is a schematic view of a rough weld extension of the present invention;

FIG. 4 is a schematic illustration of the rough weld of the present invention without adjustment;

FIG. 5 is a schematic view of the rough weld requiring adjustment of the present invention;

FIG. 6 is a schematic diagram of a gray scale curve of the present invention;

FIG. 7 is a schematic view of an edge-distance curve of the present invention;

FIG. 8 is a schematic view of a weld seam engagement area marking of the present invention;

description of reference numerals: 1 is a seam undercut, 2 is a connected domain center line, 3 is a connected domain left side edge, 4 is a connected domain right side edge, 5 is an edge band A extending edge, and 6 is an edge band B extending edge.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention.

The embodiment of the wheel weld surface defect detection method based on computer vision comprises the following steps:

the weld seam undercut 1 is a groove or a recess generated in the base material portion due to incorrect parameters or improper operation in the welding process, as shown in fig. 1. The undercut defect reduces the effective cross-sectional area of the base material, and stress concentration may be caused at the undercut, and particularly, for welding of low-alloy high-strength steel, the edge structure of the undercut is hardened, and cracks are easily caused.

In order to realize accurate detection of occlusion defects, as shown in fig. 2, the wheel weld surface defect detection method based on computer vision of the embodiment includes the following steps:

(1) obtaining a wheel weld image, performing semantic segmentation processing on the wheel weld image, and identifying a weld connected domain in the wheel weld image;

in the embodiment, the wheel weld image is collected by the camera, and for the obtained wheel weld image, the processing procedure of the embodiment is as follows:

the perspective distortion correction processing is carried out on the wheel welding line image so as to eliminate the perspective distortion influence of the arc-shaped wheel when the welding line image is shot, the perspective distortion correction process is the prior art, and the perspective distortion correction process is not repeated.

For the wheel weld image subjected to the perspective distortion correction processing, the image includes a background portion in addition to the weld due to the problem of the camera shooting angle of view. In order to avoid the influence of other noises on the detection of the weld surface defects, the DNN network is first used to identify the weld area in the image, and the specific process is as follows: and inputting the wheel weld image into a trained DNN network, and reasoning the wheel weld image by using the trained DNN network to obtain a weld connected domain in the wheel weld image.

In this embodiment, the DNN network identifies a weld in an image by adopting a semantic segmentation method, the DNN network has a task of classifying, pixels to be segmented in the DNN network are divided into two types, and a process of labeling a label corresponding to a training set is as follows: and adopting a single-channel semantic label, wherein the label of the pixel at the corresponding position belonging to the background class is 0, and the label of the pixel belonging to the welding seam class is 1. The loss function used in this embodiment is a cross entropy loss function.

(2) Obtaining rough weld edges according to the weld connected domain, and performing edge refining treatment on the rough weld edges to obtain fine weld edges;

although the weld joint connected region is obtained in the embodiment, the weld joint edge can be directly obtained, but the undercut defect exists beside the weld joint edge, and the subsequent calculation of the embodiment depends on the pixel points outside the weld joint, so that the accuracy of the weld joint edge is necessarily ensured. However, the semantic segmentation of the weld is realized by DNN, but DNN has uncertainty, that is, its precision and accuracy are limited, and it is difficult to complete edge extraction with high precision, so the weld region edge obtained by using the DNN network is only a rough weld edge, and in order to obtain a finer weld edge, the embodiment further performs refinement processing on the rough edge obtained by the semantic segmentation, and the detailed refinement process is as follows:

obtaining an edge zone of a welding line according to the welding line communicating region, marking the edge zone as a rough edge zone, extending K/2 pixel points of each edge pixel point in the rough edge zone leftwards along a direction perpendicular to a central line 2 of the welding line communicating region to obtain a left extending pixel point corresponding to each edge pixel point in the rough edge zone, wherein K is the width of the welding line communicating region at the corresponding position of each edge pixel point in the rough edge zone, and marking the range from the central line of the welding line communicating region to the left extending pixel point as a range of the edge zone A, so that an extending edge 5 of the edge zone A corresponding to the extended left edge 3 of the communicating region can be obtained; and (3) extending K/2 pixel points to the right of each edge pixel point in the rough edge zone along the direction vertical to the central line of the welding seam communicating zone to obtain a right extending pixel point corresponding to each edge pixel point in the rough edge zone, and marking the range from the central line of the welding seam communicating zone to the right extending pixel point as the range of the edge zone B, so that the corresponding edge zone B extending edge 6 after the right edge 4 of the communicating zone is extended can be obtained.

The rough edge band is corrected according to the gray level change of the edge band A and the edge band B along the direction vertical to the central line of the welding seam communication domain, and the specific process is as follows:

taking the direction of the central line of the vertical connected domain as the judgment direction, i.e. the direction of the arrow of the dotted line in fig. 3, the gray value sequence G of the pixel points is obtained, which takes the pixel points on the central line as the starting point and the extended pixel points as the end point.

Wherein the content of the first and second substances,is the gray value of the pixel point on the middle line,the gray value of the pixel point on the rough edge,the gray values of the extended pixels.

And drawing a gray level change curve of the pixel points, taking the corner mark serial numbers of the pixel points as a horizontal axis, taking the difference value of the gray level values of the pixel points and the next adjacent pixel point as a vertical axis, fitting the curve after scattered points are drawn to obtain each maximum value point on the curve, and taking the pixel point corresponding to the maximum value point as a fine edge pixel point of the welding line.

If the K/2 position is the maximum point, as shown in fig. 4, it is described that there is no deviation in the edge obtained by using the DNN network, and there is no need to adjust the pixel points on the rough edge, that is, the new edge point。

If the K/2 non-maximum point is not the maximum point, as shown in fig. 5, it is described that the edge obtained by using the DNN network has a deviation, the pixel point on the rough edge needs to be adjusted, and the maximum point closest to K/2 needs to be updated as the new edge point。

Through the above process, all new edge points can be obtainedAll new edge pointsAnd fitting connection is carried out, so that a fine weld seam edge can be obtained. In the embodiment, the DNN network is used for obtaining the rough weld edge, and the rough weld edge is refined, so that the identification speed is increased on the premise of ensuring the identification precision compared with a method for identifying the weld edge by directly using the gray difference.

This embodiment extends K/2 pixels outward, and as another embodiment, it may extend other numbers of pixels.

(3) Calculating the pit rate corresponding to each pixel between the edge of the fine weld joint and the extending edge according to the gray value of each pixel between the edge of the fine weld joint and the extending edge;

the undercut defect is spatially expressed as a crater of the weld bead, but the undercut defect on the two-dimensional image is a crater on a plane, and may have a change in light and shadow, and thus the undercut defect appears as an area having a change in gray scale on the two-dimensional image.

The process of calculating the pit rate corresponding to each pixel between the edge of the weld and the extended edge in the embodiment is as follows:

taking the direction of a central line in a vertical connected domain as a judgment direction, taking the first pixel point outside the edge pixel point as a starting point, and taking the extension pixel point as a terminal point to construct a gray sequence FG:

wherein the content of the first and second substances,as a new edge pointGray value of the first outer pixel, FTo extend edge pixelsIs determined by the gray-scale value of (a),to extend edge pixelsThe gray value of (a). In the above formula willIs further converted intoThe purpose of (2) is to renumber the subscripts starting from 1.

And (3) performing scatter plot on the gray sequence FG, taking the converted serial number as a horizontal axis and the gray value FG as a vertical axis, and fitting a curve to obtain a gray curve as shown in FIG. 6, wherein a solid line and a dotted line in the graph show gray change conditions corresponding to two different gray sequences.

The larger the mutation degree of the gray values of the adjacent pixel points in the gray sequence is, the closer to black the gray sequence is, and the more likely the gray sequence is to be a pit point; the more towards white in the gray-scale image, the more towards 255 the gray-scale value; the more black the gray value tends to be 0. Therefore, the lower the point value in the gray curve and the larger the difference with other kinds of pixel points, the larger the pit rate. In this embodiment, the pit rate corresponding to each pixel between the edge of the weld and the extended edge is calculated according to the following formula, and the pit rate is also the probability of existence of a pit:

wherein the content of the first and second substances,indicating the pit rate of the ith pixel,represents the gray value of the ith pixel point,、and respectively representing the maximum gray value and the minimum gray value in the current gray sequence, wherein QT is the average gray value of other pixel points in other categories.The lower the gray value of the ith pixel point is, the larger the pit rate is;the larger the difference between the ith pixel point and other pixel points is, the larger the pit rate is.

The determination process of the QT in this embodiment is: firstly, clustering gray values in a sequence by using a Kmeans mean algorithm, and setting the type number k of clustering as 2, namely, the pixels in the sequence can be forcedly divided into two categories no matter what value is taken; obtaining the average gray value of the corresponding pixels of each category，. When the pixels are forcedly divided into two types, if the difference of the corresponding gray values of the two types is large, the two types are judged to be respectively corresponding to pits and non-pits, and the pits exist in the current sequence at the moment; if the difference between the corresponding gray values of the two types is not large, it indicates that the two types are both black or white as a whole, but not both of them, and the probability of the existence of the pits in the current sequence is very small. When in useIn the case of class A, QT is takenWhen is coming into contact withIn the case of class B, QT is taken。

Thus, the pit rate corresponding to each pixel between the edge of the welding seam and the extending edge can be obtained。

(4) Calculating the undercut rate corresponding to each pixel point on the inner line of the weld joint connected domain according to the pit rate;

undercut defects can be divided into discontinuous undercuts and continuous undercuts, and the appearance is not the same: the discontinuous undercut is a transient melting master batch, so the external form is represented by a pit with an arc edge; and the undercut edge of the continuous undercut defect is approximately parallel to the edge of the weld bead. In order to accurately judge the undercut defect, the present embodiment calculates the undercut defect for each of the two cases, and then obtains the probability of the undercut defect, that is, the undercut defect rate, by synthesis.

In the embodiment, the discontinuous undercut rate corresponding to each pixel point on the welding seam connected domain line is calculatedThe process of (2) is as follows:

and each gray sequence perpendicular to the central line direction of the welding seam has a pixel point with the largest pit rate, and the pixel distance D from the pixel point with the largest pit rate in each judgment direction to the corresponding central line pixel point is obtained.

And numbering each pixel of the welding seam central line from top to bottom and from left to right in sequence, and taking the number as a horizontal axis and the corresponding distance D as a vertical axis to obtain a central line-distance curve. Firstly, identifying wave crest and wave trough points on a curve, and dividing the curve into single-peak areas to obtain S single-peak areas; each unimodal region corresponds to a trough-peak-trough.

And performing Gaussian fitting on each unimodal region to obtain a fitting curve, and further calculating the discontinuous undercut rate yu corresponding to each pixel point on the middle line of the weld joint connected domain:

wherein s is the sequence number of the unimodal region corresponding to the jth pixel on the midline,showing the single peak amplitude corresponding to the curve after the s single peak region is subjected to Gaussian fitting,representing the average difference of the curves before and after the Gaussian fitting corresponding to the s-th unimodal region, whereinFor the pre-gaussian value corresponding to the jth pixel point on the centerline,is the gaussian fitted value corresponding to the jth pixel point on the centerline,is the difference value before and after the Gaussian fitting corresponding to the jth pixel on the central line,and the discontinuous undercut rate corresponding to the jth pixel point on the central line.

In the embodiment, the continuous undercut rate corresponding to each pixel point on the line in the weld joint connected domain is calculatedThe process of (2) is as follows:

firstly, sequence numbers of points with the largest pit rate on a gray sequence are sequentially obtained, and sequence number values corresponding to the largest pit rate on the gray sequence FG are obtained. Similarly, the pixels in the center line of the weld are numbered sequentially from top to bottom and from left to right, and the edge-distance curve can be obtained by taking the number as the horizontal axis and the sequence number value corresponding to the maximum pit rate on the corresponding gray sequence FG as the vertical axis. The curve represents the distance R from the pixel point with the maximum pit rate to the corresponding edge in each judgment direction.

By width of weldDeriving corresponding decision window size for radixAnd judging the horizontal degree of each position according to the judgment length. Wherein a is an adjustment coefficient, which can be adjusted according to actual needs, and in this embodiment, a = 0.5.

Wherein the content of the first and second substances,indicating the distance difference corresponding to the ith pixel and the jth pixel,represents the average value of the distance differences between the total number of the Kaa pixels on the left and the right sides of the jth pixel and the jth pixel,the larger the description of the jth pixelThe more non-parallel the pixels to its left and right are, as shown in fig. 7:

setting a decision threshold value ks whenWhen the value is less than ks, the corresponding curve at the ith pixel point is approximately horizontal. The corresponding continuous undercut rate can be obtained according to ksThe continuous undercut ratio is in the range of [0,1 ]]：

Obtaining the defect rate of discontinuous undercutAnd continuous undercut defect rateThen, according to the discontinuous undercut defect rateAnd continuous undercut defect rateObtaining the undercut defect rate of the jth pixel point on the middle lineNamely:

the embodiment respectively judges the discontinuous undercut defect rate corresponding to the jth pixel point on the central lineAnd continuous undercut defect rateAnd when the edge corresponding to the jth pixel point on the middle line has the discontinuous undercut defect, calculating the discontinuous undercut defect rateWill be larger; when the continuous undercut defect exists on the edge corresponding to the jth pixel point on the middle line, the calculated continuous undercut defect rateWill be larger, so this embodiment will not break the undercut defect rateAnd continuous undercut defect rateThe larger value of the sum is used as the undercut defect rate of the jth pixel point on the middle line. Thus, the undercut defect rate corresponding to each pixel point on the middle line can be obtained.

(5) And marking the undercut defect area on the weld seam according to the undercut rate.

If the undercut defect rate corresponding to each pixel point on the central lineGreater than a set detection accuracyIf the position is considered as undercut defect, the defect needs to be labeled so as to facilitate subsequent manual detection. In this exampleAs other embodiments, modifications may be made as necessary.

Taking the central line of the welding seam as a horizontal axis, taking the undercut defect judgment result as a vertical axis, marking as 1 if the undercut is adopted, and marking as 0 if the undercut is not adopted, so as to obtain a corresponding sequence, wherein the form is as follows:

[000111110000000000000000000000110000000111111000]

when 2 or more 1 s appear continuously, it is considered that the undercut defect has occurred, and the corresponding region is determined as an undercut region.

Determining the size C of the bounding box according to the number of the central pixel points in the undercut region, determining the size H of the pit point with the largest distance corresponding to the bounding box, obtaining the sizes (C, H) of the bounding box, further obtaining a defect frame, and realizing defect marking, as shown in FIG. 8, so as to facilitate subsequent manual inspection.

In the embodiment, the fine weld edge is obtained based on the wheel weld image, the undercut rate corresponding to each pixel point on the inner line of the weld connected domain is calculated according to the pit rate corresponding to each pixel between the fine weld edge and the extended edge, and the accurate identification of the undercut defect area on the weld is realized. The embodiment provides an automatic detection method for the surface defects of wheel welding seams, and solves the problem that the existing method cannot accurately detect the undercut defects of the welding seams.

It should be noted that while the preferred embodiments of the present invention have been described, additional variations and modifications to these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.