CN110009618A - A kind of Axle Surface quality determining method and device - Google Patents

A kind of Axle Surface quality determining method and device Download PDF

Info

Publication number
CN110009618A
CN110009618A CN201910261832.1A CN201910261832A CN110009618A CN 110009618 A CN110009618 A CN 110009618A CN 201910261832 A CN201910261832 A CN 201910261832A CN 110009618 A CN110009618 A CN 110009618A
Authority
CN
China
Prior art keywords
image
defect
connected domain
axial workpiece
axle surface
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.)
Granted
Application number
CN201910261832.1A
Other languages
Chinese (zh)
Other versions
CN110009618B (en
Inventor
高一聪
李康杰
冯毅雄
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201910261832.1A priority Critical patent/CN110009618B/en
Publication of CN110009618A publication Critical patent/CN110009618A/en
Application granted granted Critical
Publication of CN110009618B publication Critical patent/CN110009618B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30164—Workpiece; Machine component

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pathology (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Signal Processing (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of Axle Surface quality determining method and devices.It is taken pictures by industrial High-Speed Line Scanning Camera to Axle Surface, the axial workpiece industry high-speed line scan image of acquisition is pre-processed, complete image segmentation with improved threshold values iterative algorithm, defect image is extracted by removal background, noise and interference；In conjunction with the area of each connected domain of segmented image, area accounting, tubbiness degree, and the extraction of depth information defect characteristic of image three-dimensional reconstruction is combined, training classifier identifies defect type.Energy detection axis class surface quality of workpieces of the present invention, and the surface defect classification of energy automatic identification axial workpiece, defect recognition rate with higher have preferable robustness to water stain equal false defects.

Description

A kind of Axle Surface quality determining method and device
Technical field
The invention belongs to test technique automatic field, specifically a kind of Axle Surface quality determining method and dress It sets.
Background technique
Metal axial workpiece is the fundamental parts of various kinds of equipment, mainly serves and transmits torque and load bearing.Its matter The quality of amount directly affects the normal operation and and safety in production of equipment.Traditional fluorescentmagnetic particle(powder) detection is grasped manually by worker It completes, not only labor intensity of workers is big, but also checkability is low, unavoidably there is the case where accidentally picking up, missing inspection.
In recent years, with the development of machine vision and photoelectric technology, the reason of surface defects detection is directly carried out using image Significant progress has been obtained by research.Defect detecting technique based on image causes the extensive concern of academia and industry. Saridis etc. is directed to industrial steel plate, and by fixed mode scan image, carrying out judgement to the image after the calculus of differences of front and back is No existing defects.Sayed etc. proposes a kind of new textile industry Fabric Defect Detection, by using entropy filtering and minimum Error threshold segmentation, so that having superperformance in fabric defects detection.Li et al. is directed to cigarette label, improves existing lack Detection algorithm is fallen into, proposes and minimum circumscribed rectangle is applied to defect shape analysis, obtained preferable practicability and higher Detection accuracy.Zhang etc. is proposed a kind of based on wavelet multi-scale analysis for the detection for taking turns tyre defect in mipmap picture Tire defect inspection method detects to distinguish defect and background texture edge using Defect Edge measurement model.What is again It is emerging etc. to propose a kind of hub defect detection method based on defect characteristic and seed filling, it is positioned by mountain peak and obtains defect picture Plain fritter is quickly obtained defect image using seed fill algorithm.Wang Xuanyin etc. proposes a kind of based on multivariate image analysis Surface defects detection algorithm, is influenced small by illumination unevenness, improves the accuracy and robustness of image detecting system.Li Manli Aiming at the problem that Deng being difficult to after board surface knot defect is by dyeing, proposed using image fusion technology a kind of new Board surface flaw detection method, and it is best to demonstrate the syncretizing effect based on Laplacian pyramid algorith.In axial workpiece Defects detection field, Lu etc. proposes a kind of method of quick detection axis class surface defect, passes through median filtering, OTSU threshold value Segmentation, mathematical morphology examine defect target, extract contour feature, are classified using SVM to defect.Sun Xuechen etc. devises defect Partitioning algorithm and defect area labeling algorithm complete the wound to camshaft surface, sand holes, grind the typical defects such as bad Determine.
In conclusion the defect detecting technique based on image especially on shaft-like workpiece Surface testing application also in rise Step section, existing detection method there are characteristics of image False Rates it is high, detection efficiency is not high the problems such as.
Summary of the invention
In order to improve the accuracy and robustness of axial workpiece defects detection, the invention proposes a kind of Axle Surfaces Quality determining method and device.
The invention adopts the following technical scheme:
One, a kind of Axle Surface quality determining method, method comprise the steps of:
Step 1) takes pictures to the surface of axial workpiece using industrial High-Speed Line Scanning Camera, obtains axial workpiece industry high speed Line scan image, axial workpiece be equipped with hole and key, and in image axial workpiece axially along image level direction；
In specific implementation, industrial High-Speed Line Scanning Camera is scanned axial workpiece with axial high-speed line is parallel to, Complete image is spliced to form after scanning.
Step 2) axial workpiece industry high-speed line scan image Threshold segmentation:
Using improved high temperature sensitivity threshold value iterative splitting approach according to tonal gradation to acquisition axial workpiece image Pixel be split, be divided into foreground and background and carry out binaryzation；
Step 3) axial workpiece defect image extracts:
Firstly, carrying out background removal to the image after segmentation；
Then, connected domain algorithm is sought by bianry image and obtains image connectivity domain, meet the image of following formula L < τ ∩ W < τ Connected domain is then judged as noise, and is removed to noise, in formula, τ be noise decision threshold, L, W be image connectivity domain most Small boundary rectangle is long and wide；
Finally, removing the interference sections in axial workpiece hole and key on image；If defect image is there is no this axis of surface Qualified product classifies to the defect image type of rejected product.
The classification of step 4) Axle Surface defect: the two-dimensional signal and grayscale image three-dimensional reconstruction of defect image are comprehensively utilized Four kinds of features of obtained three-dimensional information extraction carry out the classification of Axle Surface defect.
In the step 2), steps are as follows for improved high temperature sensitivity threshold value iterative splitting approach:
2.1) gray value of image is set as g (x, y), and x, y are the transverse and longitudinal coordinate of image slices vegetarian refreshments, find out minimum and maximum pixel Gray value LmaxAnd Lmin, take its intermediate value T1As the initial segmentation value of image,I is initially 0 in formula；
2.2) the partition value T of i-th iteration is utilizediDivide the image into g (x, y) < TiWith g (x, y) > TiProspect and back Two regions of scape calculate separately out the respective pixel number N in two regions1And N2And respective average gray value AoAnd Ab:
2.3) new partition value T is calculated againi+1=α Ao+βAb, α and β are the first, second weight coefficient, α ≠ β；
If | Ti-Ti+1Then iteration stopping, ε indicate iteration stopping threshold value, T to | < εi+ 1 is last threshold value, otherwise Ti=Ti+1, return Return step 2.1)；
In specific implementation, the weight of foreground and background in threshold value of the first, second weight coefficient α and β to adjust division. When more sensitive to prospect, β value is larger, and when more sensitive to background, α value is larger.
2.4) repeat the above steps continuous iterative processing, the final partition value T obtained with iteration stoppingiIt divides the image into Two regions of foreground and background, set 0 for prospect, set 1 for background, carry out binary conversion treatment.
The step 3) carries out background removal to the image after segmentation, specifically:
The binary map for constituting curve is drawn according to the ordinate accumulated value that the result after step 2) segmentation makes each column Pixel value after image same row all pixels binaryzation, is specifically added and tires out as the ordinate of the column by the cumulative figure of ordinate Value added, each column have an ordinate accumulated value, and it is cumulative that all ordinate accumulated values draw curve composition binary map ordinate Figure.The accumulated value for occurring background parts after cumulative is significantly higher than hole, key and rejected region accumulated value, and there are obvious boundary, Curve generation herein suddenly changes.Using two occurred at suddenly variation on curve vertical boundary lines as line of demarcation, in image from The abscissa in left-to-right two lines of demarcation is denoted as tlAnd t2, t is located to image abscissa1+ Δ t and t2Between Δ t part into Row retains, and Δ t is the abundant value of safety, and rest part is cast out as background.
The step 3) removes the interference sections in axial workpiece hole and key on image, specifically:
3.1) image after removing noise is denoted as original image img, and original image img is replicated to obtain referring to image img_m, Internal waviness is removed to closed operation is carried out referring to image img_m, operation removal defect is carried out out again later, finally to referring to image Img_m carries out expansion process；
3.2) each of reference image img_m obtained to step 3.1) processing connected domain calculates minimum circumscribed rectangle Long L, width W；To meet | Ls_k-L | < λ Ls_k ∩ | Ws_k-W | the image connectivity domain of < λ Ws_k condition is connected to as hole Domain will meet | Ls_j-L | < λ Ls_j ∩ | Ws_j-W | the image connectivity domain of < λ Ws_j condition as key connected domain, wherein The boundary rectangle that Ls_k and Ws_k respectively indicates the hole connected domain of qualified axis in image is long and wide, and Ls_j and Ws_j respectively indicate figure The boundary rectangle of the hole connected domain of qualified key is long and wide as in, and λ indicates shape decision acceptance；
According in image same row the Diff N of hole and the center of strong connected domain as interference vertical direction on Period Δ y is extended above and below image img_m, is expanded height and is Δ y and is filled with 0 pixel column, ginseng It does not extend according to the left and right sides of image img_m, is completely interfered completely being interfered referring to the top in image img_m with bottom Distinguish moving distance Δ y up and down, the top, which is completely interfered, refers to that the maximum multiple connected domains of ordinate, bottom are complete Interference refers to the smallest multiple connected domains of ordinate；Last original image img is subtracted referring to all connected domains in image img_m Part obtains defect image.
The step 4), specifically:
Combined axis class surface defects of parts feature passes through bianry image for the axial workpiece image after extraction defect and is connected to Domain seeks algorithm and obtains defect connected domain, constructs following three two dimensional characters: connected domain area S, connected domain area and its minimum The area ratio S/ (LW) and tubbiness degree W/L, W, L of boundary rectangle are the short side and long side of minimum circumscribed rectangle；
Since two-dimensional image feature cannot accurately identify this water stain false defect, thus connected by finding out defect in binary map Lead to the corresponding former gray level image in domain, the depth Z (x, y) of each pixel in defect connected domain, formula are obtained by three-dimensionalreconstruction algorithm Middle x, y are the transverse and longitudinal coordinate of image in defect connected domain, false defects are waited using its depth discrimination is water stain, according to water stain three-dimensional reconstruction Figure feature proposes the depth reflectance signature that defect connected domain is obtained using following formula:
In formula, γ is jump function, and T is depth threshold, and r is the minimum circumscribed rectangle flexibility of connected domain, i.e. long side ratio The ratio of short side, S are connected domain area, and D indicates depth reflectance signature；
With the corresponding all connected domain area S of the up-to-standard axial workpiece of known surface, connected domain area and its minimum Area ratio S/ (LW), tubbiness degree W/L, the depth reflectance signature D of boundary rectangle are input to classifier and are trained, and use is trained Classifier Surface Quality axial workpiece to be detected carries out defect classification and Detection.
Two, a kind of Axle Surface quality detection device:
Including image-forming module, drive module, control module, computing module and display module；
The image-forming module for obtaining the Axle Surface image of high quality, including uses light source and camera, light source Axial workpiece side is placed on camera；
The drive module for driving axial workpiece rotation to be detected to rotate around own central axis line, and is responsible for axis class zero The upper dress of part and lower dress；
The control module acquires shooting to camera and obtains axial workpiece for controlling drive module and image-forming module Image is spliced, and high fidelity visual is obtained；
The computing module, for carrying out Threshold segmentation to high fidelity visual, defect image extracts, Axle Surface lacks Fall into classification；
The display module, for receive and show computing module as a result, display Axle Surface quality testing knot Fruit.
The light source is linear light sorurce.
The camera is industrial line scanning CCD camera.
The invention has the benefit that
1, the present invention completes Threshold segmentation by improving threshold iterative method, obtains high temperature sensitivity, mentions to the greatest extent Defect is taken.Image segmentation is completed by background, noise and interference elimination, extracts defect image.
2, the present invention is by extracting defect image two dimensional character: area, area accounting, tubbiness degree, and combines three-dimensionalreconstruction Obtained depth reflectance signature carries out defect classification, so that Classifcation of flaws has strong robustness.
3, the present invention and traditional artificial method ratio, the present invention use method high degree of automation, non-contact inspection may be implemented It surveys, stable working state, detection speed is fast.
Detailed description of the invention
In order to which the present invention is further explained, described content, with reference to the accompanying drawing makees a specific embodiment of the invention Further details of explanation.It should be appreciated that these attached drawings are only used as typical case, and it is not to be taken as to the scope of the present invention It limits.
Fig. 1 is the overall flow figure of the method for the present invention.
Fig. 2 is the schematic illustration of detection device of the invention.
Fig. 3 is the schematic diagram of the acquisition image of detection device of the present invention.
Fig. 4 is Threshold segmentation of the present invention and negates rear schematic diagram.
Fig. 5 is binary map ordinate accumulated value coordinate diagram of the present invention.
Fig. 6 is schematic diagram after background removal of the present invention.
Fig. 7 is that the present invention completes schematic diagram after defect is extracted.
Fig. 8 is pit defect schematic diagram and three-dimensional reconstruction figure.
Fig. 8 (a) is pit defect grayscale image.
Fig. 8 (b) is pit three-dimensional reconstruction figure.
Fig. 9 is crack defect schematic diagram and three-dimensional reconstruction figure.
Fig. 9 (a) is crack defect grayscale image.
Fig. 9 (b) is crackle three-dimensional reconstruction figure.
Figure 10 is point defect schematic diagram and three-dimensional reconstruction figure.
Figure 10 (a) is point defect grayscale image.
Figure 10 (b) is point three-dimensional reconstruction figure.
Figure 11 is the water stain schematic diagram of false defect and three-dimensional reconstruction figure.
Figure 11 (a) is the water stain grayscale image of false defect.
Figure 11 (b) is water stain three-dimensional reconstruction figure.
Figure 12 is minimum circumscribed rectangle schematic diagram.
Figure 13 is defect shaft detection result figure.
Figure 14 is normal shaft detection result figure.
Specific embodiment
Illustrate embodiments of the present invention below by way of specific specific example, those skilled in the art can be by this specification Other advantages and efficacy of the present invention can be easily understood for the content made known.The present invention can also pass through in addition different specific realities The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints and application, without departing from Various modifications and alterations are carried out under spirit of the invention.
As shown in Figure 1, the embodiment of the invention patent the following steps are included:
As shown in Fig. 2, the device of specific implementation includes image-forming module, for obtaining the Axle Surface figure of high quality As, using suitable light source and camera, light source is placed on axial workpiece side, and camera is placed on the other side.Drive module, For driving axial workpiece rotation specified angle to be detected, rotate it along axis, and be responsible for the upper dress and lower dress of axial workpiece. Control module obtains image to camera and splices, obtain high fidelity visual for controlling drive module and image-forming module.It takes Detection device experiment porch is built, it is as shown in Figure 3 to obtain image.
Implementation process is broadly divided into three frame steps: Threshold segmentation, defect image extract, defect is classified.
Step 1, industry High-Speed Line Scanning Camera CA-HL02MX and its matching component take pictures to Axle Surface, obtain Axial workpiece industry high-speed line scan image.
Step 2, Threshold segmentation
2.1) gray value of image is set as g (x, y), and x, y are the transverse and longitudinal coordinate of image slices vegetarian refreshments, find out minimum and maximum pixel Gray value LmaxAnd Lmin, take its intermediate value T1As the initial segmentation value of image,I is initially 0 in formula；
2.2) the partition value T of i-th iteration is utilizediDivide the image into g (x, y) < TiWith g (x, y) > TiProspect and back Two regions of scape calculate separately out the respective pixel number N in two regions1And N2And respective average gray value AoAnd Ab:
2.3) new partition value T is calculated againi+1=α Ao+βAb, α and β are the first, second weight coefficient, α ≠ β；
If | Ti-Ti+1Then iteration stopping, ε indicate iteration stopping threshold value, T to | < εi+ 1 is last threshold value, otherwise Ti=Ti+1, return Return step 2.1)；
2.4) repeat the above steps continuous iterative processing, the final partition value T obtained with iteration stoppingiIt divides the image into Two regions of foreground and background, set 0 for prospect, set 1 for background, carry out binary conversion treatment.
In this example, β is specially 0.995, α 0.005.It is 0.001 that iteration stopping threshold value, which takes ε,.
Step 3, defect image extract
3.1) binary map after segmentation is negated, as shown in Figure 4.
Both sides white strip is acquisition system both ends retaining part, is background to be removed.Binary map ordinate accumulated value As shown in figure 5, background parts accumulated value is significantly higher than hole, key and rejected region accumulated value, and there are obvious boundary, curve herein Generation suddenly changes.Using two occurred at suddenly variation on curve vertical boundary lines as line of demarcation, in image from left to right The abscissa in two lines of demarcation is denoted as tlAnd t2, t is located to image abscissa1+ Δ t and t2Partially retained between Δ t, Δ T is the abundant value of safety, and rest part is cast out as background.Δ t=20 in this example, after background removal Fig. 6.
3.2) connected domain algorithm is sought by bianry image, the low defect area with area very little of gray value is noise, will The image connectivity domain for meeting following formula L < τ ∩ W < τ is then judged as noise, and is removed to noise, and in formula, τ is noise decision threshold Value, L, W are that the minimum circumscribed rectangle in image connectivity domain is long and wide；This example τ=15.
3.3) followed by rule of surface interference elimination:
3.3.1 the image after) removing noise is denoted as original image img, and original image img replicates to obtain referring to image img_ M removes internal waviness to closed operation is carried out referring to image img_m, and disk structural element, size 8 are chosen in closed operation.Later again into Row opens operation removal defect, chooses disk structural element, size 30.Finally expansion process, choosing are carried out to referring to image img_m Take disk structural element, size 5；
3.3.2) each of reference image img_m obtained to step 3.1) processing connected domain calculates minimum external square Long L, the width W of shape；To meet | Ls_k-L | < λ Ls_k ∩ | Ws_k-W | the image connectivity domain of < λ Ws_k condition is as Kong Lian Logical domain, will meet | Ls_j-L | < λ Ls_j ∩ | Ws_j-W | the image connectivity domain of < λ Ws_j condition as key connected domain, The boundary rectangle that middle Ls_k and Ws_k respectively indicates the hole connected domain of qualified axis in image is long and wide, and Ls_j and Ws_j are respectively indicated The boundary rectangle of the hole connected domain of qualified key is long and wide in image, and λ indicates shape decision acceptance；This example Ls_k=118, Ws_k =118, Ls_j=674, Ws_j=214, λ=0.2.
According in image same row the Diff N of hole and the center of strong connected domain as interference vertical direction on Period Δ y, this example Δ y=833 are extended above and below image img_m, and expanding height is Δ y and filling For 0 pixel column, do not extended referring to the left and right sides of image img_m, will completely interfere referring to the top in image img_m and Bottom completely interferes respectively moving distance Δ y up and down, and the top, which is completely interfered, refers to the maximum multiple companies of ordinate Logical domain, bottom, which is completely interfered, refers to the smallest multiple connected domains of ordinate；Last original image img is subtracted referring to image img_ All connected domain parts in m obtain defect image.As shown in Figure 7.If defect image is qualified product there is no this axis of surface, Classify to the defect image type of rejected product.
Step 4, defect image classification
Common deficiency type and its three-dimensional reconstruction figure such as pit Fig. 8, crackle Fig. 9, point Figure 10, in addition there are false defect water Stain Figure 11, (a) is defect image in figure, (b) is reconstructed results.
It needs to extract defect characteristic to realize that defect is correctly classified.Combined axis class surface defects of parts feature, for extraction Axial workpiece image after defect seeks algorithm by bianry image connected domain and obtains defect connected domain, further constructs following two Dimensional feature: connected domain area: the area ratio (area accounting) of S, connected domain and its minimum circumscribed rectangle: S/ (LW), tubbiness degree: W/ L.W, L is the short side and long side of minimum circumscribed rectangle.
The corresponding former gray level image of each defect connected domain in binary map is found out, minimum circumscribed rectangle is as shown in figure 12, short Side W, long side L.The depth Z (x, y) of each pixel in defect connected domain is obtained by three-dimensionalreconstruction algorithm.X in formula, y are defect Image transverse and longitudinal coordinate in connected domain.Specifically using document (Tsai P S, Shah M.Shape From Shading Using Linear Approximation [J] .Image Vision Comp, 1995,12 (8): 487-498) the method carry out weight Structure.According to water stain three-dimensional reconstruction figure feature, following calculating defect connected domain depth reflectance signature is proposed:
In formula, γ is jump function, and T is depth threshold, and it is the minimum circumscribed rectangle flexibility of connected domain that this example, which takes 2, r, I.e. long side is than the ratio of short side, and S is connected domain area, and D indicates depth reflectance signature.
With the area ratio S/ (LW) of each connected domain area S of defect image, connected domain area and its minimum circumscribed rectangle, slightly Short degree W/L, depth reflectance signature D are input to logistic regression classifier and are trained, with trained logistic regression classifier pair Surface quality axial workpiece to be detected carries out defect classification and Detection.
For the accuracy for verifying proposed method, 1000 axial workpiece production images are taken, will have pit, crackle, point The axis of defect problem is defined as rejected product.Figure 13 and 14 is the shaft member classification results of rejected product and qualified product.Weighting identification Rate are as follows:
In formula, P is precision ratio, and R is recall ratio.
In formula, TP is the number that qualified product is identified as qualified product, and FN is the number that qualified product is identified as rejected product, and FP is Rejected product is identified as the number of qualified product, and TN is the number that rejected product is identified as rejected product.
As shown in table 1, axial workpiece qualified product is identified as qualified product 955, and qualified product is identified as rejected product 22, no Qualified product is identified as qualified product 1, and rejected product is identified as 21 of rejected product, weights discrimination F1It is 98.86%, it is average It is 3.69 seconds time-consuming, reduced 63.1% compared with artificial detection average 10 seconds, greatly improves the detection efficiency of axial workpiece.
1 defect axis recognition result of table
Defect classification takes 100 image composition training sets return the training of classifier.To trained network, benefit It is tested with 20 image composition test sets.As shown in table 2.Training set accuracy 82%, test set accuracy 75%.
2 classifier experimental result of table
Thus above-mentioned implementation is as it can be seen that energy detection axis class surface quality of workpieces of the present invention, and can automatic identification axial workpiece Surface defect classification, defect recognition rate with higher has preferable robustness to water stain equal false defects.
Only example formula illustrates the principle of the present invention and its effect to above-mentioned specific embodiment, is not intended to limit the present invention.Appoint What those skilled in the art all without departing from the spirit and scope of the present invention, modifies above-described embodiment or is changed Become.Therefore, all equivalent modifications or change completed without departing from the spirit and technical ideas disclosed in the present invention, still It should be covered by the claims of the present invention.

Claims (6)

1. a kind of Axle Surface quality determining method, which is characterized in that comprise the steps of:
Step 1) takes pictures to the surface of axial workpiece using industrial High-Speed Line Scanning Camera, obtains axial workpiece industry high-speed line and sweeps Trace designs picture, axial workpiece be equipped with hole and key, and in image axial workpiece axially along image level direction；
Step 2) axial workpiece industry high-speed line scan image Threshold segmentation: using improved high temperature sensitivity threshold value iteration point Segmentation method is split the pixel for obtaining axial workpiece image, is divided into foreground and background and carries out binaryzation, specifically:
2.1) gray value of image is set as g (x, y), and x, y are the transverse and longitudinal coordinate of image slices vegetarian refreshments, find out minimum and maximum pixel grey scale Value LmaxAnd Lmin, take its intermediate value T1As the initial segmentation value of image,
2.2) the partition value T of i-th iteration is utilizediDivide the image into g (x, y) < TiWith g (x, y) > TiForeground and background two A region calculates separately out the respective pixel number N in two regions1And N2And respective average gray value AoAnd Ab:
2.3) new partition value T is calculated againi+1=α Ao+βAb, α and β are the first, second weight coefficient, α ≠ β；
If | Ti-Ti+1Then iteration stopping, ε indicate iteration stopping threshold value, T to | < εi+ 1 is last threshold value, otherwise Ti=Ti+1, return to step It is rapid 2.1)；
2.4) repeat the above steps continuous iterative processing, the final partition value T obtained with iteration stoppingiDivide the image into prospect With two regions of background, 0 is set by prospect, 1 is set by background, carries out binary conversion treatment；
Step 3) axial workpiece defect image extracts:
Firstly, carrying out background removal to the image after segmentation；
Then, connected domain algorithm is sought by bianry image and obtains image connectivity domain, meet the image connectivity of following formula L < τ ∩ W < τ Domain is then judged as noise, and is removed to noise, and in formula, τ is noise decision threshold, and L, W are that the minimum in image connectivity domain is outer It is long and wide to connect rectangle；
Finally, removing the interference sections in axial workpiece hole and key on image；
The classification of step 4) Axle Surface defect: the two-dimensional signal and grayscale image three-dimensional reconstruction for comprehensively utilizing defect image obtain Four kinds of features of three-dimensional information extraction carry out the classification of Axle Surface defect, specifically:
Algorithm is sought by bianry image connected domain for the axial workpiece image after extraction defect and obtains defect connected domain, is constructed Three two dimensional characters below: the area ratio S/ (LW) and tubbiness degree of connected domain area S, connected domain area and its minimum circumscribed rectangle W/L, W, L are the short side and long side of minimum circumscribed rectangle；
By finding out the corresponding former gray level image of defect connected domain in binary map, defect connected domain is obtained by three-dimensionalreconstruction algorithm In each pixel depth Z (x, y), x in formula, y are the transverse and longitudinal coordinate of image in defect connected domain, and proposition is obtained using following formula Obtain the depth reflectance signature of defect connected domain:
In formula, γ is jump function, and T is depth threshold, and r is the minimum circumscribed rectangle flexibility of connected domain, i.e. long side compares short side Ratio, S be connected domain area, D indicate depth reflectance signature；
It is external with the corresponding all connected domain area S of the up-to-standard axial workpiece of known surface, connected domain area and its minimum Area ratio S/ (LW), tubbiness degree W/L, the depth reflectance signature D of rectangle are input to classifier and are trained, with trained classification Device Surface Quality axial workpiece to be detected carries out defect classification and Detection.
2. a kind of Axle Surface quality determining method according to claim 1, it is characterised in that:
The step 3) carries out background removal to the image after segmentation, specifically: it has been made according to the result after step 2) segmentation The ordinate accumulated value of each column draws the cumulative figure of binary map ordinate for constituting curve, will occur at suddenly variation on curve As line of demarcation, the abscissa in two lines of demarcation in image from left to right is denoted as t in two vertical boundary lineslAnd t2, to the horizontal seat of image Mark is in t1+ Δ t and t2Partially retained between Δ t, Δ t is the abundant value of safety, and rest part is cast out as background.
3. a kind of Axle Surface quality determining method according to claim 1, it is characterised in that:
The step 3) removes the interference sections in axial workpiece hole and key on image, specifically:
3.1) image after removing noise is denoted as original image img, and original image img replicates to obtain referring to image img_m, to ginseng Closed operation is carried out according to image img_m and removes internal waviness, operation removal defect is carried out out again later, finally to referring to image img_m Carry out expansion process；
3.2) each of reference image img_m obtained to step 3.1) processing connected domain calculates the length of minimum circumscribed rectangle L, width W；To meet | Ls_k-L | < λ Ls_k ∩ | Ws_k-W | it the image connectivity domain of < λ Ws_k condition, will as hole connected domain Meet | Ls_j-L | < λ Ls_j ∩ | Ws_j-W | the image connectivity domain of < λ Ws_j condition be used as key connected domain, wherein Ls_k with The boundary rectangle that Ws_k respectively indicates the hole connected domain of qualified axis in image is long and wide, and Ls_j and Ws_j are respectively indicated in image and closed The boundary rectangle of the hole connected domain of lattice key is long and wide, and λ indicates shape decision acceptance；
According in image same row the Diff N of hole and the center of strong connected domain as interference vertical direction on the period Δ y is extended above and below image img_m, is expanded height and is Δ y and is filled with 0 pixel column, by reference The top in image img_m completely interfere with bottom completely interfere distinguish moving distance Δ y, the top are complete up and down Whole interference refers to the maximum multiple connected domains of ordinate, and bottom, which is completely interfered, refers to the smallest multiple connected domains of ordinate；Most Original image img, which is subtracted, afterwards obtains defect image referring to image img_m.
4. a kind of Axle Surface quality detection device, it is characterised in that:
Including image-forming module, drive module, control module, computing module and display module；
The image-forming module for obtaining the Axle Surface image of high quality, including uses light source and camera, light source and phase Machine is placed on axial workpiece side；
The drive module for driving axial workpiece rotation to be detected to rotate around own central axis line, and is responsible for axial workpiece Upper dress and lower dress；
The control module acquires the image that shooting obtains axial workpiece to camera for controlling drive module and image-forming module Spliced, obtains high fidelity visual；
The computing module, for carrying out Threshold segmentation to high fidelity visual, defect image extracts, Axle Surface defect point Class；
The display module, for receive and show computing module as a result, display Axle Surface quality measurements.
5. a kind of Axle Surface quality detection device according to claim 4, it is characterised in that: the light source is line Property light source.
6. a kind of Axle Surface quality detection device according to claim 4, it is characterised in that:
The camera is industrial line scanning CCD camera.
CN201910261832.1A 2019-04-02 2019-04-02 Shaft part surface quality detection method and device Active CN110009618B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910261832.1A CN110009618B (en) 2019-04-02 2019-04-02 Shaft part surface quality detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910261832.1A CN110009618B (en) 2019-04-02 2019-04-02 Shaft part surface quality detection method and device

Publications (2)

Publication Number Publication Date
CN110009618A true CN110009618A (en) 2019-07-12
CN110009618B CN110009618B (en) 2021-04-20

Family

ID=67169488

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910261832.1A Active CN110009618B (en) 2019-04-02 2019-04-02 Shaft part surface quality detection method and device

Country Status (1)

Country Link
CN (1) CN110009618B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110426180A (en) * 2019-08-05 2019-11-08 浙江大学 A kind of three-dimensional scenic drafting delay measurement system and method based on high-speed line scanning
CN110658213A (en) * 2019-09-30 2020-01-07 合肥公共安全技术研究院 Nondestructive testing device for detecting three-dimensional defects on surface of seamless steel pipe
CN111487192A (en) * 2020-04-26 2020-08-04 天津海融科技有限公司 Machine vision surface defect detection device and method based on artificial intelligence
CN111882537A (en) * 2020-07-28 2020-11-03 研祥智能科技股份有限公司 Visual inspection method and system
CN111915581A (en) * 2020-07-27 2020-11-10 青岛大学 Method for detecting defects of smooth metal cambered surface
CN112122175A (en) * 2020-08-12 2020-12-25 浙江大学 Material enhanced feature recognition and selection method of color sorter
CN112200805A (en) * 2020-11-11 2021-01-08 北京平恒智能科技有限公司 Industrial product image target extraction and defect judgment method
CN112308832A (en) * 2020-10-29 2021-02-02 常熟理工学院 Bearing quality detection method based on machine vision
CN113096123A (en) * 2021-05-07 2021-07-09 湘潭大学 Multi-fuzzy reasoning cascaded primer side defect classification and damage degree analysis method
CN114419038A (en) * 2022-03-28 2022-04-29 深圳市信润富联数字科技有限公司 Method and device for identifying surface defects of hub, storage medium and electronic equipment

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN201331508Y (en) * 2009-01-14 2009-10-21 张颖 Automatic detection device for flaw detection of magnetic powder on metal shaft part surface
CN102680478A (en) * 2012-04-25 2012-09-19 华南农业大学 Detection method and device of surface defect of mechanical part based on machine vision
CN102854191A (en) * 2012-07-18 2013-01-02 湖南大学 Real-time visual detection and identification method for high speed rail surface defect
CN103729842A (en) * 2013-12-20 2014-04-16 中原工学院 Fabric defect detection method based on local statistical characteristics and overall significance analysis
CN104063873A (en) * 2014-07-08 2014-09-24 华东交通大学 Shaft sleeve part surface defect on-line detection method based on compressed sensing
US20140319340A1 (en) * 2012-03-27 2014-10-30 Kla-Tencor Corporation Method and apparatus for detecting buried defects
CN105092598A (en) * 2015-09-28 2015-11-25 深圳大学 Method and system for rapidly recognizing defects of big-breadth PCB on basis of connected areas
CN107884413A (en) * 2017-10-24 2018-04-06 华东交通大学 A kind of device and detection method of automatic detection bearing roller missing
CN107876425A (en) * 2017-11-10 2018-04-06 华东交通大学 A kind of bearing defect detecting system device of view-based access control model
CN108021938A (en) * 2017-11-29 2018-05-11 中冶南方工程技术有限公司 A kind of Cold-strip Steel Surface defect online detection method and detecting system
CN108714701A (en) * 2018-06-15 2018-10-30 广州大学 A kind of Machining of Shaft-type Parts device
US20180322623A1 (en) * 2017-05-08 2018-11-08 Aquifi, Inc. Systems and methods for inspection and defect detection using 3-d scanning
CN108898585A (en) * 2018-06-15 2018-11-27 广州大学 A kind of axial workpiece detection method and its device
CN108921846A (en) * 2018-07-17 2018-11-30 北京航空航天大学 A kind of rail tread defect identification method combined based on gray level image and depth image

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN201331508Y (en) * 2009-01-14 2009-10-21 张颖 Automatic detection device for flaw detection of magnetic powder on metal shaft part surface
US20140319340A1 (en) * 2012-03-27 2014-10-30 Kla-Tencor Corporation Method and apparatus for detecting buried defects
CN102680478A (en) * 2012-04-25 2012-09-19 华南农业大学 Detection method and device of surface defect of mechanical part based on machine vision
CN102854191A (en) * 2012-07-18 2013-01-02 湖南大学 Real-time visual detection and identification method for high speed rail surface defect
CN103729842A (en) * 2013-12-20 2014-04-16 中原工学院 Fabric defect detection method based on local statistical characteristics and overall significance analysis
CN104063873A (en) * 2014-07-08 2014-09-24 华东交通大学 Shaft sleeve part surface defect on-line detection method based on compressed sensing
CN105092598A (en) * 2015-09-28 2015-11-25 深圳大学 Method and system for rapidly recognizing defects of big-breadth PCB on basis of connected areas
US20180322623A1 (en) * 2017-05-08 2018-11-08 Aquifi, Inc. Systems and methods for inspection and defect detection using 3-d scanning
CN107884413A (en) * 2017-10-24 2018-04-06 华东交通大学 A kind of device and detection method of automatic detection bearing roller missing
CN107876425A (en) * 2017-11-10 2018-04-06 华东交通大学 A kind of bearing defect detecting system device of view-based access control model
CN108021938A (en) * 2017-11-29 2018-05-11 中冶南方工程技术有限公司 A kind of Cold-strip Steel Surface defect online detection method and detecting system
CN108714701A (en) * 2018-06-15 2018-10-30 广州大学 A kind of Machining of Shaft-type Parts device
CN108898585A (en) * 2018-06-15 2018-11-27 广州大学 A kind of axial workpiece detection method and its device
CN108921846A (en) * 2018-07-17 2018-11-30 北京航空航天大学 A kind of rail tread defect identification method combined based on gray level image and depth image

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
TONG TONG 等: "Defects Detection Of Weld Image Based On Mathematical Morphology and Thresholding Segmentation", 《2012 8TH INTERNATIONAL CONFERENCE ON WIRELESS COMMUNICATIONS,NETWORKING AND MOBILE COMPUTING》 *
何再兴 等: "基于缺陷特征与种子填充的轮毂缺陷检测方法", 《浙江大学学报（工学版）》 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110426180A (en) * 2019-08-05 2019-11-08 浙江大学 A kind of three-dimensional scenic drafting delay measurement system and method based on high-speed line scanning
CN110658213A (en) * 2019-09-30 2020-01-07 合肥公共安全技术研究院 Nondestructive testing device for detecting three-dimensional defects on surface of seamless steel pipe
CN110658213B (en) * 2019-09-30 2021-11-26 合肥公共安全技术研究院 Nondestructive testing device for detecting three-dimensional defects on surface of seamless steel pipe
CN111487192A (en) * 2020-04-26 2020-08-04 天津海融科技有限公司 Machine vision surface defect detection device and method based on artificial intelligence
CN111915581A (en) * 2020-07-27 2020-11-10 青岛大学 Method for detecting defects of smooth metal cambered surface
CN111882537A (en) * 2020-07-28 2020-11-03 研祥智能科技股份有限公司 Visual inspection method and system
CN111882537B (en) * 2020-07-28 2023-12-15 研祥智能科技股份有限公司 Visual detection method and system
CN112122175A (en) * 2020-08-12 2020-12-25 浙江大学 Material enhanced feature recognition and selection method of color sorter
CN112308832A (en) * 2020-10-29 2021-02-02 常熟理工学院 Bearing quality detection method based on machine vision
CN112200805A (en) * 2020-11-11 2021-01-08 北京平恒智能科技有限公司 Industrial product image target extraction and defect judgment method
CN113096123A (en) * 2021-05-07 2021-07-09 湘潭大学 Multi-fuzzy reasoning cascaded primer side defect classification and damage degree analysis method
CN114419038A (en) * 2022-03-28 2022-04-29 深圳市信润富联数字科技有限公司 Method and device for identifying surface defects of hub, storage medium and electronic equipment

Also Published As

Publication number Publication date
CN110009618B (en) 2021-04-20

Similar Documents

Publication Publication Date Title
CN110009618A (en) A kind of Axle Surface quality determining method and device
CN108961217B (en) Surface defect detection method based on regular training
CN110349126B (en) Convolutional neural network-based marked steel plate surface defect detection method
CN109377485B (en) Machine vision detection method for instant noodle packaging defects
CN111862064B (en) Silver wire surface flaw identification method based on deep learning
CN104458755B (en) Multi-type material surface defect detection method based on machine vision
CN103048329B (en) A kind of road surface crack detection method based on active contour model
CN104008553B (en) Crack detection method with image gradient information and watershed method conflated
CN111310558A (en) Pavement disease intelligent extraction method based on deep learning and image processing method
CN102305798B (en) Method for detecting and classifying glass defects based on machine vision
CN101706445B (en) Image processing mthod for beef marbling grain grade scoring
CN108765402B (en) Non-woven fabric defect detection and classification method
CN111047655B (en) High-definition camera cloth defect detection method based on convolutional neural network
CN106093066A (en) A kind of magnetic tile surface defect detection method based on the machine vision attention mechanism improved
CN102496161B (en) Method for extracting contour of image of printed circuit board (PCB)
CN101852768A (en) Workpiece flaw identification method based on compound characteristics in magnaflux powder inspection environment
CN104504388A (en) Pavement crack identification and feature extraction algorithm and system
CN103593670A (en) Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine
CN109840483B (en) Landslide crack detection and identification method and device
CN103729856B (en) A kind of Fabric Defects Inspection detection method utilizing S-transformation signal extraction
CN105447512A (en) Coarse-fine optical surface defect detection method and coarse-fine optical surface defect detection device
CN103778627B (en) A kind of marine site based on SAR image oil spilling detection method
CN109087286A (en) A kind of detection method and application based on Computer Image Processing and pattern-recognition
CN112734761B (en) Industrial product image boundary contour extraction method
CN110473201A (en) A kind of automatic testing method and device of disc surface defect

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
GR01 Patent grant
GR01 Patent grant