CN110659675A - Welding seam defect detection method based on AdaBoost algorithm - Google Patents
Welding seam defect detection method based on AdaBoost algorithm Download PDFInfo
- Publication number
- CN110659675A CN110659675A CN201910838781.4A CN201910838781A CN110659675A CN 110659675 A CN110659675 A CN 110659675A CN 201910838781 A CN201910838781 A CN 201910838781A CN 110659675 A CN110659675 A CN 110659675A
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- defects
- adaboost
- method based
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The invention belongs to the technical field of weld joint detection, and particularly relates to a weld joint defect detection method based on an Adaboost algorithm. The invention adopts the AdaBoost algorithm with the punishment item, thereby improving the defect detection rate. When the defect classifier and the non-defect classifier are trained, the defect detection rate is improved by adjusting the punishment coefficient and weighting the defect samples. In addition, AdaBoost is used for feature selection, and features which better represent defect characteristics are screened out. Meanwhile, in defect type judgment, the second classification is expanded into a multi-classification AdaBoost. The classifier has different priorities for different defect categories and has good classification performance for non-uniformly distributed samples.
Description
Technical Field
The invention belongs to the technical field of weld joint detection, and particularly relates to a weld joint defect detection method based on an Adaboost algorithm.
Background
Nondestructive inspection is an important technology with a wide application range in the field of industrial detection. According to the statistics of the central intelligence office of the united states, by 2014, there are approximately 3,500,000 kilometers of pipes in 120 countries around the world. Annular welding areas appear at intervals of a distance of the pipeline, and the two pipelines are connected. Welding technology has found widespread use in modern industries as a primary method of work joining operations. Circular welding is an important welding process for joining seamless steel pipes. The quality of welding between pipelines is directly related to the service life and safe use of the pipelines. And thus is important for defect detection in the weld seam. The method can quickly and accurately detect and locate the defects by a nondestructive inspection method, realizes defect location, qualification and quantification, and provides objective and reasonable basis for pipeline quality evaluation and safety evaluation.
At present, the quality evaluation of the pipeline is mainly completed by X-ray nondestructive inspection. X-rays, also known as roentgen rays, can penetrate substances, and thus can detect structures inside objects. The X-ray imaging technology plays a great role in the fields of medicine, security inspection, nondestructive inspection and the like. The X-ray nondestructive inspection method is a method in which an object to be inspected is irradiated with a ray having a uniform intensity, the transmitted ray is exposed to a photographic film, the photographic film is developed to obtain a negative film having different blackness corresponding to the internal structure and defects of the material, and the types, sizes, distributions, etc. of the defects are inspected by observing the negative film.
At present, the processing of the X-ray negative film is mainly divided into two modes, namely manual detection and computer-aided detection. Manual testing relies on expertise, operational experience and a portion of subjective guessing held by the tester. And the quality evaluation result of the pipeline welding seam is easily influenced due to different professional skills and experiences among people. And the defects of inaccurate defect size acquisition, long time consumption and the like exist in manual evaluation, and objective and accurate evaluation on the pipeline welding seam is difficult to perform. Computer-aided detection represents a significant advance over manual detection. The digital image is obtained by scanning the X-ray negative film, the display of the image and the storage of the evaluation result are realized by using a computer, the evaluation process is simplified, and the time is saved. However, the auxiliary evaluation also realizes the qualitative, positioning and quantitative determination of the defects through human judgment, and has the defect of manual detection. In recent years, researchers at home and abroad have gradually focused on automatically evaluating weld defects of X-ray films by means of computer technology. Because the X-ray film detection system is limited by imaging conditions, the obtained radiographic image has the characteristic points of low contrast, unbalanced gray distribution, much noise, dynamic blurring and the like. These features also make the detection of defects in X-ray films difficult.
Disclosure of Invention
Aiming at the detection difficulty caused by the problems of poor image shooting quality, uneven exposure, noise and the like of an X-ray negative film, the invention provides a welding seam defect detection method based on an Adaboost algorithm, and the Adaboost algorithm with a punishment item is adopted, so that the defect detection rate is improved. When the defect classifier and the non-defect classifier are trained, the defect detection rate is improved by adjusting the punishment coefficient and weighting the defect samples. In addition, AdaBoost is used for feature selection, and features which better represent defect characteristics are screened out. Meanwhile, in defect type judgment, the second classification is expanded into a multi-classification AdaBoost. The classifier has different priorities for different defect categories and has good classification performance for non-uniformly distributed samples.
In order to achieve the purpose, the invention adopts the following technical scheme:
a welding seam defect detection method based on an AdaBoost algorithm comprises the following steps,
step S1, shooting an X-ray image of the welding seam by using an X-ray machine;
step S2, defect segmentation, which is to perform defect segmentation on the original X-ray image, and includes three processes: image preprocessing, background subtraction and potential defect extraction. Image pre-processing includes image filtering and brightness correction. The histogram equalization improves the brightness distribution of the image by a brightness correction method. The background subtraction method obtains an image containing only defect information. And finally, marking the connected region to obtain an independent defect region.
The potential defect regions are also referred to as a candidate defect set. Candidate defects are contaminated with a large number of non-defects, which can cause serious defect misdetection. The invention can remove non-defects and reduce the false detection rate while ensuring the defect detection rate. Through experimental analysis, interferences such as overexposure, underexposure, character areas, image quality lines and the like are all causes of non-defects. Taking these factors into account, a rich set of negative examples (non-defects) is constructed. And classifying defects and non-defects by a statistical learning method to remove the non-defects.
Step S3, extracting the characteristics of the candidate defect area, extracting the gray characteristic and the geometric characteristic of the defect, calculating the gray difference, and inputting the gray difference into an Adaboost classifier;
and step S4, judging the defects by the Adaboost classifier, realizing the removal of non-defects and classifying the defects at the same time.
The classification process of the defects mainly extracts features through a horizontal region comparison method, and the defects are judged by adopting an improved AdaBoost algorithm. In the AdaBoost training process, two parameters need to be determined, wherein one parameter is a threshold value of a classifier, and the other parameter is the weight of the classifier; and adjusting the distribution of the samples, and improving the attention degree of the next weak classifier on the misclassified samples by adjusting the distribution of the samples. The improved AdaBoost algorithm increases punishment on a certain classification condition on the basis of an original algorithm, so that the classification performance of a classifier on corresponding classes is improved. And the defect types are classified by adopting a machine learning method, and two classes AdaBoost are expanded into a multi-class framework.
Different from the prior art, the invention has the advantages that: extracting features by combining a horizontal region comparison method, and selecting the features by adopting an AdaBoost algorithm; an AdaBoost algorithm with a penalty term is provided, and defect samples are weighted in the defect and non-defect secondary classification process; different classes are given different biases in multiple classifications. The 3 defects of crack, unfused and incomplete penetration are more harmful and the sample data size is small. The AdaBoost algorithm has a higher recall in the classification of these 3 defects.
Drawings
FIG. 1 is a block diagram of weld defect detection and identification;
FIG. 2 is a defect segmentation flow chart;
FIG. 3 is a diagram of candidate defects;
FIG. 4 is a schematic diagram of defect comparison;
FIG. 5 is a schematic view of a defect geometry;
FIG. 6 is a classification flow chart;
FIG. 7 is a chart of recall statistics for five defect types for four classifiers.
Detailed Description
To explain technical contents, structural features, and objects and effects of the technical solutions in detail, the following detailed description is given with reference to the accompanying drawings in conjunction with the embodiments.
Referring to fig. 1, the invention provides a weld defect detection method based on AdaBoost algorithm, including the following steps,
and step S1, acquiring images, namely, shooting the X-ray images of the welding seams by using an X-ray machine, namely, scanning the X-ray films of the welding seams shot by the X-ray camera to form pictures so as to facilitate subsequent processing.
In step S2, the defect segmentation may be divided into three processes, i.e., image preprocessing, background subtraction and candidate region extraction. Referring to FIG. 2, a defect segmentation flow chart is shown. The image preprocessing is mainly used for eliminating noise and enhancing the contrast of the image. Image pre-processing includes image filtering and brightness correction. Noise reduction is first achieved using median filter processing, and the luminance distribution of the image is improved by luminance correction method histogram equalization. The background subtraction method obtains an image only containing defect information, a Gaussian background modeling method is used for modeling a background, the modeled image is smoothed, a mean filter is used for smoothing the background model, a difference image is obtained by background subtraction, and a region filter is used for removing noise from the difference image. Finally, the connected region mark obtains independent candidate defect regions, which is shown in fig. 3.
Where the median filter and the mean filter, both using a 3 x 3 kernel, can remove white noise and point noise very effectively.
The defect regions are significantly darker and smaller than the surrounding pixels, so a large median filter is used for background modeling. Kernel size h of filterwThe smoothed image is represented by f (i, j) and the background image b (i, j) depending on the resolution of the image. (i, j) represents the pixel coordinates of the image.) the difference image fsComprises the following steps:
b(i,j)＝median(f(i+x,j+y),x，y∈[-hw,hw]) (1)
the difference image is a binary image that may contain many non-defects, such as noise and edges. Since defects are generally smaller than these non-defects, an area filter may be used to remove noise. Also, the etching operation has a good effect on such non-defects.
And step S3, extracting the characteristics of the candidate defect area, extracting the gray characteristic and the geometric characteristic of the defect, calculating the gray difference, and inputting the gray difference into an Adaboost classifier.
The features have a great influence on the classification result, so that the selection and extraction of the features are very important. Regarding the characteristics of defects, two aspects are mainly focused on: grayscale features and geometric features. By observing the X-ray image, it can be seen that the defect and the surrounding area are significantly different, especially in the horizontal direction, the density distribution is significantly unbalanced, see fig. 4, which is a schematic diagram of the defect. Therefore, the difference of the gray scale in the horizontal direction can be calculated, a feature descriptor, namely a feature vector, is formed, and the feature vector is sent to the Adaboost classifier.
And step S4, judging the defects by the Adaboost classifier, realizing the removal of non-defects and classifying the defects at the same time.
In Adaboost, a weak classifier selects a binary decision tree. The decision tree is of the form
Where x is the feature, f (x) is the corresponding feature value, φ is the polarity (1 or-1), θ is the threshold of the decision tree. Thus, 61 features are selected in the invention, a weak classifier is established for each feature, and finally a strong classifier is composed of 61 weak classifiers.
The specific training process of the AdaBoost classifier is as follows:
(1) initializing sample weights: when y isiWhen 1, D1(i)＝1/mp(ii) a When y isiWhen is-1, D1(i)＝1/mn； mpIs the number of positive samples, mnIs the number of negative samples. The weights are then normalized.
(2) Training T weak classifiers, and for the T weak classifier:
(ii) based on the current sample distribution DtThe t-th weak classifier, i.e. the above-mentioned two-classification decision tree, is trained to find a suitable threshold θ so as to achieve the minimum classification error rate.
② obtaining the prediction label h under the current weak classifiertAnd calculating the classification error rate as follows:
calculating the weight coefficient of the current classifier according to the error rate of the classifier, and recording as:
fourthly, updating the sample distribution, wherein the updating rule is as follows:
wherein ZtIs a normalization factor. It can be seen that, compared with the original training process, for the condition that the defect is classified as non-defect, the method adds a multiplication term, namely the penalty coefficient p, and by adjusting the value of the penalty coefficient p, the sample weight of the defect classified as non-defect is increased, so as to increase the degree of importance of the subsequent classifier on the sample weight, and finally, the ratio of the defect classified as defect, namely the TPR parameter, is increased.
(3) Each combination of T weak classifiers is a strong classifier, which is represented as:
meanwhile, a cascade method of cascade is adopted. Cascading is a method of connecting multiple strong classifiers, as shown in fig. 4, where each stage of classifier separates out negative samples, the remaining samples enter the next stage of classifier, and the prediction by the last classifier is positive samples.
The invention extracts two types of characteristics: the gray scale compares the features with the geometric features. And calculating comparison characteristics by using a horizontal region comparison method, and selecting the characteristics by adopting an AdaBoost algorithm. The result shows that the distance from the vertical direction to the center of the welding seam, the average gray scale, the gray scale standard difference, the average gray scale difference with the left comparison area, the average gray scale difference with the right comparison area and the relative gray scale are the characteristics with good characteristic of defect characteristics; meanwhile, the invention increases the classification process of defects and non-defects, removes a large amount of non-defects and improves the detection rate of the defects; the probability of the defects of different types is different, so that the problem of unbalanced category of the acquired data is difficult to avoid. In the five-defect classification process, the AdaBoost algorithm has good classification performance on non-uniformly distributed samples. The algorithm is applied to the online detection of the pipeline welding seam picture, and the effectiveness and the practicability of the method are verified.
Example one
The welding seam defect of the pipeline can cause oil gas leakage and other consequences, and the safe use of the pipeline is seriously damaged. Therefore, the invention provides a weld defect detection method based on the AdaBoost algorithm.
1. Firstly, scanning an X-ray negative film to obtain a digital image, using a median filter to realize image noise reduction, adopting histogram equalization to enhance image contrast and improve image brightness, respectively, adopting a Gaussian background modeling method to obtain a background model, using a mean filter to smooth the background model, then adopting background subtraction to obtain a difference image, using a region filter to remove noise from the difference image, and using a kernel function of 3X 3 for both the median filter and the mean filter.
And modeling the background by using a Gaussian background modeling method, and smoothing the modeled image. For the input image, a difference image is calculated by equations (1) and (2). Wherein the resolution of the image is 4000 multiplied by 1024, and the kernel size hwSet to 30.
And then, performing morphological operation on the difference image, and simultaneously performing denoising by using a region filter to obtain a candidate defect region set.
2. Next, feature extraction is performed on the candidate defect regions. The features with 61 total feature numbers 1-7 are the gray scale features of the defect area, the features with 8-14 are the gray scale features of the left comparison area, and the features with 15-21 are the gray scale features of the right comparison area. Also, the difference in gray level between the defect and the surrounding, i.e., the gray level deviation, is an important attribute for describing whether or not the defect is a defect. The features of numbers 22-28 are the gray scale deviation features of the defect and the left comparison area, the features of numbers 28-35 are the gray scale deviation features of the defect and the right comparison area, and the numbers 35-42 are the sums of the left deviation features and the right deviation features. In addition, some of the features, numbered 43-61, mainly describe the blackness, size and shape of the defect, such as contrast, compactness, filling, moment features, etc. The features expressed by numbers 43-61 are as follows:
(43) area of
The area is the area of the defect, and is the number of pixels in the shaded portion.
(44) Relative gray scale
The relative gray is defined as the ratio of the average gray of the defect area to the average gray of the rest of the expanded rectangle except the defect. The extended rectangle is a rectangular area in which the circumscribed rectangle extends outward by 5 pixels. Relative gray scale describes the brightness of the defect compared to surrounding pixels.
(45) Circumference length
The perimeter, i.e., the perimeter of the defect, is the sum of the number of pixels at the edge of the defect.
(46) Compactness of
The crack defect has no fixed shape. Other four types of defects are similar to circular or rectangular in shape compared to cracks. The compactness E is a feature of the shape description and is expressed as:
E＝C2/S (8)
where C is the perimeter of the defect and S is the area of the defect. The smaller E, the more the defect shape approaches a circle; the larger E, the more complicated the defect shape. The shape of the crack is more complex than other defects and therefore the tightness is generally higher.
(47) Distance ratio of vertical direction to weld center
The distance ratio of the vertical direction to the center of the weld is the ratio of the distance from the center of the defect to the center of the weld to the width of the weld in the vertical direction.
(48) Long axis direction angle
The major axis direction angle is an angle between the major axis and the horizontal axis X. The long axis is calculated by Principal Component Analysis (PCA). Firstly, constructing a covariance matrix of coordinates (x, y) of each pixel in the defect; then calculating the eigenvalue and eigenvector of the covariance matrix; the eigenvector corresponding to the larger eigenvalue is the main direction of data distribution, as the long axis direction. Similarly, the eigenvector corresponding to the smaller eigenvalue serves as the direction of the minor axis.
(49) Length of defect
The major and minor axes were calculated as described above. The projection of the image in a certain direction is then calculated using the radon transform, which is expressed as:
where (x, y) is the position of the pixel and phi is the projection angle. And projecting the defect in the long axis direction, wherein the projection length in the direction is the defect length.
(50) Width of defect
Similarly, the projection length in the short axis direction is the defect width.
(51) Aspect ratio
Lx、LyRespectively representing the defect length and the defect width of the defect, wherein the length-width ratio is Lx/Ly. When L isx/LyWhen not less than 3, it is impossible to be circularAnd (5) a defect.
(52) Degree of filling
The degree of filling, as the name implies, is the degree of filling of the defect in its circumscribed rectangle, expressed as:
ρ＝S/(w×h) (10)
where S is the defect area, w is the width of the rectangle circumscribing the defect, and h is the length of the rectangle circumscribing the defect.
(53) Edge smoothness
Edge smoothness is a parameter that describes the extent of defect edge and surface smoothness. Let I (x, y) be the defect image edge, (x)g,yg) Is the center of gravity of the defect region. The smoothness FS can be expressed as:
(54) sharpness degree
Sharpness describes the sharpness of both ends of a defect. Taking the area from one end point to one quarter of the length as S1Taking the area from the end point of the other end to one quarter of the length as S2Fig. 5 is a schematic diagram of the geometrical features of the defect. Sharpness is expressed as:
SP＝1-(S1+S2)/S (12)
the larger the sharpness parameter, the stronger the sharpness of both ends.
(55) Symmetry property
Symmetry describes the symmetry of the defect shape with the minor axis as the axis of symmetry. The greater the degree of symmetry, the closer the parameter SYM is to 0, which can be expressed as:
SYM＝|1-S1/S2| (13)
(56) degree of deviation
The degree of deviation is defined as the degree to which the center of gravity of the defect deviates from the geometric center of the circumscribed rectangle of the defect. The degree of deviation σ is calculated as:
wherein (x)g，yg) Is a defective areaCenter of gravity of the domain, (x)c，yc) Is the geometrical center of the circumscribed rectangle of the defect.
(57) Flatness in the major axis direction
When calculating the defect length, the projection in the main axis direction can be obtained through radon transformation. The standard deviation of the projection values is defined as the long axis direction flatness. The projection standard deviation may represent the shape of the projection, with larger values indicating more irregular shapes reflecting the smoothness of the pixel distribution in the long axis direction.
(58) Uniformity of gray scale
The gray uniformity is defined as the standard deviation of the gray within the defect area. Unfused grays are not uniformly distributed, one side is higher and the other side is lower, and thus the grayscale uniformity may be relatively high.
(59) Ratio of upper and lower brightness
The up-down luminance ratio is defined as the luminance ratio of the up-down two partial areas in the vertical direction. Dividing the defect into two parts with equal height, the average gray scale of the upper half part is g1The average gray level of the lower half is g2. The upper and lower brightness ratio is g1/g2Similar to the uniformity of gray scale, the property of different brightness between the upper side and the lower side without fusion can be described.
(60) Distance ratio of horizontal direction to weld center
The distance ratio of the horizontal direction to the center of the weld is the ratio of the distance from the center of the defect to the center of the weld to the length of the weld in the horizontal direction.
(61) Center moment
Moment features are often used to describe profile characteristics, and the central moment is used herein, and the calculation formula of the central moment μ is:
3. and sending the obtained feature vectors to an Adaboost classifier for classification. This embodiment addresses five typical defects: cracks, unfused, lack of penetration, strip defects, and circular defects. The defect identification process is divided into defect segmentation, two classification of defects and non-defects and multi-classification of defect types. After defect segmentation, 3951 candidate defects are obtained, which include 1067 defects and 2884 non-defects.
It can be seen that 3951 candidate defects are obtained by image segmentation, and 70% or more of the candidate defects are non-defects, so that the non-defects must be removed. The invention increases the classification process of defects and non-defects, removes a large number of non-defects and simultaneously improves the detection rate of the defects. The invention provides an AdaBoost algorithm with a penalty term, which emphasizes defect samples in the process of classifying defects and non-defects.
In the cascade process, the training of each stage of classifier needs to adjust the penalty coefficient p. In the classification flow of fig. 6, the input samples of each stage of the classifier are different, and thus the p-parameter that satisfies the requirement is also different. Therefore, the penalty factor p needs to be updated, and the update formula is:
p＝p+Δp,Δp＝d×λ×(1.0-TPR)+d×(1.0-λ)×(0.5-FPR) (3.1)
where d is the adjustment step size, λ is the bias weight coefficient, and the larger λ is, the higher the TPR parameter of the trained classifier is. And gradually adjusting the value of the penalty coefficient through an iterative formula so as to achieve the set target.
In the experiment, the training end conditions for each class are:
(1) the current classifier satisfies TPR close to 1.0 and FPR less than 0.5.
(2) The iterative training times reach the set maximum iterative times.
And when any one of the conditions is met in the training process, ending the training, finishing the training of the current classifier, and entering the training of the next-stage classifier.
When the multi-classification is expanded, each sub-classifier separates one class, and the un-separated class enters the next sub-classifier. As shown in fig. 7, part a is responsible for sorting out non-defects, which is a binary process of defects and non-defects. And the part B is a classification defect type and is sequentially separated according to the sequence of cracks, unfused, incomplete penetration, strip defects and circular defects. 5 defects are classified by 4 sub-classifiers.
The experimental data are cross-validated by five folds, the classification accuracy is 85.5%, and the defect detection rate (TPR) is 91.66%. And 3 sets of comparative tests were set up: SVM, RandomForest and kNN, which are optimally set. Considering only the TP parameter, the proposed method increases 22.37% over kNN, 20.05% over SVM, 10.13% over RF, and AdaBoost shows its superiority.
AdaBoost may give different classes different biases in multiple classifications. The 3 defects of crack, unfused and incomplete penetration are more harmful and the sample data size is small. AdaBoost has higher recall in the classification of these 3 defects than SVM, RandomForest and kNN.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrases "comprising … …" or "comprising … …" does not exclude the presence of additional elements in a process, method, article, or terminal that comprises the element. Further, herein, "greater than," "less than," "more than," and the like are understood to exclude the present numbers; the terms "above", "below", "within" and the like are to be understood as including the number.
Although the embodiments have been described, once the basic inventive concept is obtained, other variations and modifications of these embodiments can be made by those skilled in the art, so that the above embodiments are only examples of the present invention, and not intended to limit the scope of the present invention, and all equivalent structures or equivalent processes using the contents of the present specification and drawings, or any other related technical fields, which are directly or indirectly applied thereto, are included in the scope of the present invention.
Claims (7)
1. A weld defect detection method based on an AdaBoost algorithm is characterized by comprising the following steps: comprises the following steps of (a) carrying out,
step S1, shooting an X-ray image of the welding seam by using an X-ray machine;
step S2, dividing the defect, preprocessing the X-ray image, eliminating noise, enhancing the contrast of the image, modeling and smoothing the image background, calculating a differential image, and processing the differential image to obtain a candidate defect area set;
step S3, extracting the characteristics of the candidate defect area, extracting the gray characteristic and the geometric characteristic of the defect, calculating the gray difference, and inputting the gray difference into an Adaboost classifier;
and step S4, judging the defects by the Adaboost classifier, realizing the removal of non-defects and classifying the defects at the same time.
2. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 1, characterized in that: the image preprocessing in step S2 includes image filtering and brightness correction, and histogram equalization is used to improve the brightness distribution of the image through the brightness correction method.
3. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 2, characterized in that: and the image filtering adopts a median filter to realize noise reduction.
4. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 1, characterized in that: in step S2, a background modeling method using a large median filter is used to model the background.
5. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 1, characterized in that: in step S2, an averaging filter is used to smooth the image.
6. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 1, characterized in that: in step S3, features are extracted by a horizontal area comparison method, and the grayscale differences are combined into a feature vector and input to an Adaboost classifier.
7. The welding seam defect detection method based on the AdaBoost algorithm as claimed in claim 1, characterized in that: the step S4 classifies the defect as: cracks, unfused, lack of penetration, strip defects, and circular defects.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910838781.4A CN110659675A (en) | 2019-09-05 | 2019-09-05 | Welding seam defect detection method based on AdaBoost algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910838781.4A CN110659675A (en) | 2019-09-05 | 2019-09-05 | Welding seam defect detection method based on AdaBoost algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110659675A true CN110659675A (en) | 2020-01-07 |
Family
ID=69037964
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910838781.4A Pending CN110659675A (en) | 2019-09-05 | 2019-09-05 | Welding seam defect detection method based on AdaBoost algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110659675A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292303A (en) * | 2020-01-21 | 2020-06-16 | 湖北文理学院 | Weld defect type detection method and device, electronic equipment and storage medium |
| CN111709916A (en) * | 2020-05-29 | 2020-09-25 | 青岛铁木真软件技术有限公司 | Cord detection method, system, device and storage medium |
| CN112102358A (en) * | 2020-09-29 | 2020-12-18 | 南开大学 | Non-invasive animal behavior characteristic observation method |
| CN112651341A (en) * | 2020-12-28 | 2021-04-13 | 长江大学 | Processing method of welded pipe weld joint real-time detection video |
| CN112767329A (en) * | 2021-01-08 | 2021-05-07 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776964A (en) * | 2018-06-04 | 2018-11-09 | 武汉理工大学 | A kind of ship weld defect image detecting system and method based on Adaboost and Haar features |
| CN109859181A (en) * | 2019-01-29 | 2019-06-07 | 桂林电子科技大学 | A kind of PCB welding point defect detection method |
-
2019
- 2019-09-05 CN CN201910838781.4A patent/CN110659675A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776964A (en) * | 2018-06-04 | 2018-11-09 | 武汉理工大学 | A kind of ship weld defect image detecting system and method based on Adaboost and Haar features |
| CN109859181A (en) * | 2019-01-29 | 2019-06-07 | 桂林电子科技大学 | A kind of PCB welding point defect detection method |
Non-Patent Citations (1)
| Title |
|---|
| FENG DUAN ET AL.: "Automatic Welding Defect Detection of X-Ray Images by Using Cascade AdaBoost With Penalty Term", 《IEEE ACCESS》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292303A (en) * | 2020-01-21 | 2020-06-16 | 湖北文理学院 | Weld defect type detection method and device, electronic equipment and storage medium |
| CN111292303B (en) * | 2020-01-21 | 2023-09-19 | 湖北文理学院 | Weld defect type detection method and device, electronic equipment and storage medium |
| CN111709916A (en) * | 2020-05-29 | 2020-09-25 | 青岛铁木真软件技术有限公司 | Cord detection method, system, device and storage medium |
| CN112102358A (en) * | 2020-09-29 | 2020-12-18 | 南开大学 | Non-invasive animal behavior characteristic observation method |
| CN112102358B (en) * | 2020-09-29 | 2023-04-07 | 南开大学 | Non-invasive animal behavior characteristic observation method |
| CN112651341A (en) * | 2020-12-28 | 2021-04-13 | 长江大学 | Processing method of welded pipe weld joint real-time detection video |
| CN112767329A (en) * | 2021-01-08 | 2021-05-07 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
| CN112767329B (en) * | 2021-01-08 | 2021-09-10 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110659675A (en) | Welding seam defect detection method based on AdaBoost algorithm | |
| CN113362326B (en) | Method and device for detecting defects of welding spots of battery | |
| US10565479B1 (en) | Identifying and excluding blurred areas of images of stained tissue to improve cancer scoring | |
| Wang et al. | Automatic identification of different types of welding defects in radiographic images | |
| CN106846316B (en) | Automatic identification method for typical defect images in GIS | |
| WO2019104767A1 (en) | Fabric defect detection method based on deep convolutional neural network and visual saliency | |
| CN114372983B (en) | Shielding box coating quality detection method and system based on image processing | |
| CN110097547B (en) | Automatic detection method for welding seam negative film counterfeiting based on deep learning | |
| CN114663346A (en) | Strip steel surface defect detection method based on improved YOLOv5 network | |
| CN112102229A (en) | Intelligent industrial CT detection defect identification method based on deep learning | |
| CN107610111A (en) | A kind of weld point image detection method based on deep learning | |
| JP2011214903A (en) | Appearance inspection apparatus, and apparatus, method and program for generating appearance inspection discriminator | |
| CN112465746B (en) | Method for detecting small defects in ray film | |
| CN104574418A (en) | Pressure vessel weld defect identification method and device based on neural network | |
| CN113177456B (en) | Remote sensing target detection method based on single-stage full convolution network and multi-feature fusion | |
| WO2021153633A1 (en) | Metal structure phase classification method, metal structure phase classification device, metal structure phase learning method, metal structure phase learning device, material property prediction method for metal material, and material property prediction device for metal material | |
| Nizam et al. | Vision based identification and classification of weld defects in welding environments: A review | |
| CN114419007B (en) | Defect type identification method and system for multi-strategy fusion deep learning network model | |
| CN113313107B (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| CN113313678A (en) | Automatic sperm morphology analysis method based on multi-scale feature fusion | |
| CN109544513A (en) | A kind of steel pipe end surface defect extraction knowledge method for distinguishing | |
| Nacereddine et al. | Computer-aided shape analysis and classification of weld defects in industrial radiography based invariant attributes and neural networks | |
| CN116721294A (en) | Image classification method based on hierarchical fine granularity classification | |
| CN114943843B (en) | Welding defect detection method based on shape sensing | |
| CN110674852A (en) | Weld defect classification method based on geometric features and AdaBoost algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200107 |