CN109829876B - Chain plate defect online detection device and method based on machine vision - Google Patents

Chain plate defect online detection device and method based on machine vision Download PDF

Info

Publication number
CN109829876B
CN109829876B CN201810534908.9A CN201810534908A CN109829876B CN 109829876 B CN109829876 B CN 109829876B CN 201810534908 A CN201810534908 A CN 201810534908A CN 109829876 B CN109829876 B CN 109829876B
Authority
CN
China
Prior art keywords
image
edge
chain
gray
chain plates
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.)
Active
Application number
CN201810534908.9A
Other languages
Chinese (zh)
Other versions
CN109829876A (en
Inventor
周怡君
罗晨
孙润民
王玉立
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.)
Southeast University
Original Assignee
Southeast University
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 Southeast University filed Critical Southeast University
Priority to CN201810534908.9A priority Critical patent/CN109829876B/en
Publication of CN109829876A publication Critical patent/CN109829876A/en
Application granted granted Critical
Publication of CN109829876B publication Critical patent/CN109829876B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

The invention discloses a device and a method for detecting defects of chain plates on line based on machine vision, wherein the chain plates to be detected are placed in a vibrating hopper, and a conveying belt is started to convey the chain plates onto the conveying belt; triggering the industrial camera to be started after the trigger delay relay detects the chain plates on the conveyor belt, so as to acquire images of the chain plates on the conveyor belt and send the images to a computer for image preprocessing and chain plate defect detection; and the chain plates without the detected defects are qualified chain plates, the chain plates with the defects are unqualified chain plates, and the qualified chain plates and the unqualified chain plates are sorted and output by a sorting and discharging mechanism. After chain plate parts are punched and discharged, the chain plate parts in batches are detected in real time without omission, three types of defective chain plates can be accurately identified, and the chain plate parts are sorted and discharged, 3, the detection efficiency of a single detection line reaches 60-140 pieces/minute, and the detection accuracy rate is more than 95%.

Description

Chain plate defect online detection device and method based on machine vision
Technical Field
The invention relates to the technical field of chain plate defect detection in the production process of chains, in particular to a device and a method for detecting the defects of chain plates on line based on machine vision.
Background
Chains have gone through the development process from manual assembly to mechanical assembly to streamlined assembly since their birth. Chains with a pitch below 19.05mm have reached a high technical level. Various problems still exist in the whole production flow.
The conventional telescopic roller chain mainly comprises the following components: the chain comprises an inner chain plate, an outer chain plate, a sleeve, a roller and a pin shaft. The production and manufacturing of the chain are industrial processes of part production and overall assembly performed around the expansion of the five parts, and the production links of various parts comprise chain plate blanking, pin shaft slitting, sleeve pipe coiling, roller cold heading forming and part heat treatment. The heat treatment process of the parts is generally performed collectively after the parts are individually formed.
The polygon effect (uneven speed) and the dynamic load impact during the operation of the sleeve roller chain cause the potential defects of any part in the chain to possibly cause the abnormal failure of the finished chain, thereby causing production operation accidents. When the chain parts are produced, a small amount of unqualified defective parts are inevitably generated, so that the quality of the parts in each link is comprehensively monitored and timely detected in the chain production flow.
According to the above, before the chain parts are assembled, the produced chain parts should be comprehensively checked and screened, and defective parts are eliminated. At present, domestic enterprises mainly use a manual operation mode or a single machine semi-automatic operation mode to produce middle and low-end transmission chains, and part production lacks professional detection equipment to carry out real-time comprehensive detection. Part of parts still adopt artifical visual detection mode, and this kind of lower artifical detection link of reliability has leaded to the increase of cost of labor, and production efficiency has received the restriction, has also seriously hindered chain automation line's formation.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a device and a method for detecting the defects of a chain plate on line based on machine vision.
In order to achieve the purpose, the invention is implemented according to the following technical scheme:
a chain plate defect online detection device based on machine vision comprises a feeding module, a transmission module and a discharging and sorting module, wherein the feeding module comprises a vibrating hopper for vibrating feeding, chain plates are arranged in the vibrating hopper, the transmission module comprises a conveyor belt, the discharging and sorting module comprises a sorting and discharging mechanism, a discharge port of the vibrating hopper is connected with one end of the conveyor belt and used for conveying the chain plates onto the conveyor belt, a feed port of the sorting and discharging mechanism is connected with the other end of the conveyor belt and used for sorting and outputting the chain plates, the chain plate defect online detection device further comprises an image acquisition module, the image acquisition module comprises a trigger time delay relay, an industrial camera and a computer comprising software for preprocessing images of the chain plates and detecting defects of the chain plates, the industrial camera is fixed right above the conveyor belt, a lens of the industrial camera faces downwards and is perpendicular to the conveyor belt and used for acquiring images of the, the trigger delay relay is located on one side of the industrial camera and connected with the industrial camera to trigger the industrial camera to acquire images of the chain plates when the chain plates are detected to be conveyed on the conveying belt, and the industrial camera is connected with the computer to transmit the acquired images of the chain plates to the computer to perform image preprocessing and detect defects of the chain plates.
Further, the image acquisition module also comprises an illumination light source fixed below the lens of the industrial camera.
Further, the trigger delay relay is a proximity switch.
In addition, the invention also provides a machine vision-based chain plate defect online detection method, which comprises the following steps:
s1, placing the chain plates to be detected into a vibration hopper, and opening a conveyor belt to convey the chain plates onto the conveyor belt;
s2, triggering the industrial camera to be started after the trigger delay relay detects the chain plate on the conveyor belt, so as to acquire the image of the chain plate on the conveyor belt and send the image to the computer for image preprocessing and chain plate defect detection;
and S3, detecting the chain plates without defects to be qualified chain plates, detecting the chain plates with the defects to be unqualified chain plates, and sorting and outputting the qualified chain plates and the unqualified chain plates through a sorting and discharging mechanism.
Specifically, the image preprocessing and the chain plate defect detection performed by the computer in S2 specifically include:
s21, image preprocessing:
s211, image noise reduction: by two-dimensional Gaussian distribution
Figure GDA0001935709540000031
Carrying out smoothing processing on a chain plate image input into a computer;
s212, an image enhancement algorithm based on PALMSHE: improving and enhancing the image by using gray value transformation g (x, y) to T [ f (x, y) ], wherein f (x, y) is an input image, g (x, y) is an output image, and T is a gray operation defined on the neighborhood of the point (x, y) for the gray value;
s213, continuously processing the image through binary equalization of the near peak minimum value of the gray level histogram, wherein the specific principle is as follows:
Figure GDA0001935709540000032
Figure GDA0001935709540000041
wherein L is1For dividing point gray scale, selecting the gray minimum value point nearest to the gray peak value of the histogram as the dividing point in the algorithm, L1Its corresponding gray scale. p is a radical ofx1(k),px2(k) Is a pixel frequency statistical function of the divided sub-gray level histogram; correspondingly, cdfx1(k)，cdfx2(k) Is a cumulative function of the respective pixel frequency statistics; the low-frequency gray value interval and the high-frequency concentrated gray value interval are divided through a pixel frequency minimum value point L1 between high and low-frequency gray levels, the pixel frequency and the accumulated frequency of two independent sub-histograms are respectively counted and unified into the final accumulated gray frequency, the accumulated gray frequency based on equalization processing is changed into a piecewise function in the front and rear intervals after the gray level division, meanwhile, the gray level range after the equalization processing of each interval is limited during the equalization processing, so that pixel points belonging to different gray ratio intervals are subjected to equalized gray level conversion operation in each original gray level interval according to the histogram statistical frequency, cross-interval gray level extrusion and transfer cannot occur, the picture can be enhanced, and simultaneously, the picture can be more perfectRetaining image characteristic information therein;
s22, detecting defects of the chain plates:
s221, detecting incomplete contours of the chain plates: the preprocessed image is processed by a threshold segmentation algorithm, and the formula of the threshold segmentation algorithm is
Figure GDA0001935709540000042
Wherein f (x, y) is the gray level of the original pixel point, g (x, y) is the processed pixel area, and T is the selected segmentation threshold;
setting a threshold value T to divide the image into two parts C0And C1Then C is0、C1The probability, mean and variance of each of the two parts are:
Figure GDA0001935709540000051
Figure GDA0001935709540000052
Figure GDA0001935709540000053
where i is the number of gray levels, Pi is the probability of gray level correspondence, L is the upper limit of the number of gray levels of the image, and μ is the global mean of the image
Figure GDA0001935709540000054
μ (T) is a gray-scale average value of the area image C0
Figure GDA0001935709540000055
The variance between regions C0 and C1 is defined as: sigmaB 2＝ω0(μ0-μ)2+ω1(μ1-μ)2＝ω0ω1(μ1-μ0)2
Figure GDA0001935709540000056
When the variance value between classes is maximum, the difference between the foreground and the background is maximumThe corresponding T is the optimal threshold;
taking a slightly larger value as a discrimination threshold value T1-1Counting the pixel points in the link plate area by CopComparison with a discrimination threshold, Cop<T1-1The link plates with complete outlines are larger than the threshold value, and the link plates with incomplete outlines are smaller than the threshold value;
s222, punching shift detection: smoothing the preprocessed image by using a two-dimensional Gaussian distribution G (x, y), wherein the formula of the two-dimensional Gaussian distribution G (x, y) is as follows: g (x, y) ═ G (x, y) × f (x, y), where f (x, y) is the original image, G (x, y) is the filtered image, and × is the convolution operation; obtaining a corresponding Gaussian filter transformation matrix through the formula of the two-dimensional Gaussian distribution G (x, y);
calculating the magnitude and direction of the gradient by using the finite difference of the first-order partial derivatives;
carrying out non-maximum suppression on the gradient amplitude;
edges are detected and connected with a dual threshold algorithm: the dual threshold algorithm applies two thresholds τ 1 and τ 2, 2 τ 1 ≈ τ 2, to the non-maximum suppressed image, thereby obtaining two threshold edge images N1[i,j]And N2[i，j]Due to N2[i，j]Obtained using a high threshold, and thus contains few false edges, but is discontinuous, i.e., not closed; the double threshold method is to be in N2[i，j]The algorithm is at N when the end points of the contour are reached1[i,j]8 neighbor location finding to connect to the edge on the contour, the algorithm is constantly at N1[i,j]Until N is reached2[i,j]Until the images are connected, the image edges meeting the canny detection standard can be obtained by extracting the connected domains which are connected by 8;
performing interpolation operation on the detected discrete edge points by adopting a Deriche method and applying a quadratic interpolation fitting technology, and connecting the edge points to obtain a sub-pixel level edge curve; the image edge set obtained through the processing comprises the edge information of the link plate outline edge, the punched edge and the peripheral defects of the link plate outline edge, and the punched edge of the link plate is extracted from the edge image through the length of the edge line and the head-tail linear distance of the edge line;
adopting a least square method:
Figure GDA0001935709540000061
wherein (x)c,yc) Is the coordinate of the center of the fitting circle, and R is the radius of the fitting circle; the edge has n edge points with the coordinate of (x)i,yi) (ii) a d is the sum of the squares of the distances from the points to the contour of the fitting circle, and a certain fitting circle is preset, so that the sum of the squares of the distances from all the points to the contour of the preset fitting circle is minimized, because
Figure GDA0001935709540000062
Without an analytical solution, it is expressed approximately as:
Figure GDA0001935709540000063
defining an auxiliary function: h (x, y) ═ xi-xc)2+(yi-yc)2-R2Then, the original formula is represented as:
Figure GDA0001935709540000071
since the sum of squared differences f is greater than 0, the maximum value is infinite when there is a minimum value equal to or greater than 0. The following conditions are satisfied when the minimum value is taken:
Figure GDA0001935709540000072
wherein
Figure GDA0001935709540000073
If the radius R is not 0, then there is ∑ h (x)i,yi) 0, additionally:
Figure GDA0001935709540000074
Figure GDA0001935709540000075
to obtain
∑xih(xi,yi)＝0
∑yih(xi,yi)＝0，
Is provided with
Figure GDA0001935709540000076
Derived by substituting into the preceding formula
∑uih(xi,yi)＝0
∑vih(xi,yi)＝0，
Is unfolded to obtain
∑ui((ui-uc)2+(vi-vc)2-R2)＝0
∑vi((ui-uc)2+(vi-vc)2-R2)＝0，
Further developed
∑(ui 3-2ui 2uc+uivi 2-2uivivc)＝0
∑(ui 2vi-2uiviuc+vi 3-2vi 2vc)＝0，
Defining partial expressions
Figure GDA0001935709540000081
Substituted into the above formula
Figure GDA0001935709540000082
Figure GDA0001935709540000083
Solve uc and vc
Figure GDA0001935709540000084
To further obtain
Figure GDA0001935709540000091
Obtaining the center coordinates and radius values of the fitting circle, and obtaining the center position coordinates (x) of the two punching circles after fitting the circlec1,yc1)(xc2,yc2) And radius r1 and r2, taking the midpoint of the connecting line between the centers of the two circles, and connecting the midpoint coordinate with the center coordinate (x) of the previously extracted chain plate connecting domain0,y0) Comparing, wherein the absolute value of the error is larger than the error threshold value, namely the transverse error threshold value T2-125, longitudinal threshold T2-255, namely the unqualified chain plate with punching deviation,
Figure GDA0001935709540000092
s223, detecting the punching edge defects: the circle center coordinate (x) obtained according to the step of punching and fitting a circlec1,yc1)(xc2,yc2) And the radius r1, r2 is used as a concentric circle with a slightly larger radius, and pictures in the concentric circle region are extracted to be used as an ROI for detecting punching edge defects; and (4) extracting the edge of the ROI picture area again, and screening lines which accord with the defect characteristics by using the line length in the edge extraction so as to obtain the number of the extracted defect lines>1 is taken as a main discrimination standard; number of saddle points detected in ROI (region of interest)>2 as the secondary criterion.
Compared with the prior art, the invention has the beneficial effects that:
1. after the chain plate parts are punched and discharged, detecting the chain plate parts in batches in real time without omission; 2. the chain plates with the three types of defects can be accurately identified, and the chain plate parts can be sorted and discharged; 3. the detection efficiency of a single detection line reaches 60-140 sheets/minute, and the detection accuracy rate is more than 95%.
Drawings
FIG. 1 is a schematic structural diagram of the present invention.
FIG. 2 is a flow chart of the detection according to the present invention.
FIG. 3 is a flow chart of an image pre-processing algorithm of the present invention.
Fig. 4 is a flow chart of binary equalization of the near-peak minimum value of the gray histogram in the image preprocessing process of the present invention.
Fig. 5 is a processing effect diagram of a link plate part picture acquired under the condition of insufficient illumination in the image preprocessing process of the invention, (a) is the acquisition of an original image, (b) is an original image gray level histogram, (c) is a traditional histogram equalization processing image, (d) is a traditional histogram equalization processing image gray level histogram, (e) is a PALMSHE processing image, and (f) is a PALMSHE processing image gray level histogram.
Fig. 6 is a processing effect diagram of a link plate part picture acquired under a strong illumination condition in the image preprocessing process of the invention, (a) is the acquisition of an original image, (b) is an original image gray level histogram, (c) is a traditional histogram equalization processing image, (d) is a traditional histogram equalization processing image gray level histogram, (e) is a palmsche processing image, and (f) is a palmsche processing image gray level histogram.
Fig. 7 is a flow chart of detecting the link plate profile deformity according to the present invention.
Fig. 8 is a graph showing the effect of threshold segmentation processing in the link plate contour defect detection process of the present invention, (a) is an original gray image, and (b) is an inter-class variance threshold segmentation processing image.
FIG. 9 is a flow chart of punch shift detection according to the present invention.
Fig. 10 is a diagram showing the effect of the Deriche operator in the punching shift detection process of the present invention, (a) is an original image in grayscale, and (b) is an image processed by the Deriche operator.
Fig. 11 is a drawing showing the effect of the Deriche operator processing on the edges of the sub-pixels during the punching shift detection of the present invention, (a) a gray image of the original image, and (b) a sub-pixel Deriche operator processing image.
Fig. 12 is a diagram showing the effect of circle fitting in the punching shift detection process according to the present invention.
FIG. 13 is a flow chart of the punching edge defect detection of the present invention.
FIG. 14 is a schematic diagram of saddle points in a punching edge defect detection process of the present invention, (a) saddle points in three-dimensional space, and (b) saddle points in a plane.
FIG. 15 is a diagram illustrating the effect of saddle point detection in the ROI area during the detection of edge defects in punched holes according to the present invention.
Detailed Description
The present invention will be further described with reference to specific examples, which are illustrative of the invention and are not to be construed as limiting the invention.
As shown in fig. 1, the device for detecting defects of chain plates on line based on machine vision in this embodiment comprises a feeding module, a transmission module and a discharging and sorting module, wherein the feeding module comprises a vibrating hopper 1 for vibrating feeding, the chain plates 4 are arranged in the vibrating hopper 1, the transmission module comprises a conveyor belt 2, the discharging and sorting module comprises a sorting and discharging mechanism 3, a discharge port of the vibrating hopper 1 is connected with one end of the conveyor belt 2 and used for conveying the chain plates 4 onto the conveyor belt 2, a feed port of the sorting and discharging mechanism 3 is connected with the other end of the conveyor belt 2 and used for sorting and outputting the chain plates 4, the device further comprises an image acquisition module, the image acquisition module comprises a trigger delay relay 5, an industrial camera 6 and a computer 7 containing software for preprocessing images of the chain plates and detecting defects of the chain plates, the trigger delay relay is a proximity switch, the industrial camera 6 is fixed directly over the conveyer belt 2 and the camera lens of industrial camera 6 is down with the perpendicular image of the link joint 4 on the collection conveyer belt of using of conveyer belt 2, trigger delay relay 5 and be located one side of industrial camera 6 and trigger delay relay 5 and be connected with industrial camera 6 and be used for triggering industrial camera 6 and carry out image acquisition to the link joint when detecting that there is the link joint to carry on the conveyer belt, industrial camera 6 is connected with computer 7 and is used for carrying out image preprocessing and detection link joint defect with the link joint image transmission who gathers to the computer.
Further, the image acquisition module further comprises an illumination light source (not shown in the figure) fixed below the lens of the industrial camera 6, and the illumination light source is used for supplementing light when the industrial camera 6 acquires the chain plate image.
The method for detecting the defects of the chain plates on line based on the machine vision comprises the following steps:
s1, placing the chain plates to be detected into a vibration hopper, and opening a conveyor belt to convey the chain plates onto the conveyor belt;
s2, triggering the industrial camera to be started after the trigger delay relay detects the chain plate on the conveyor belt, so as to acquire the image of the chain plate on the conveyor belt and send the image to the computer for image preprocessing and chain plate defect detection;
and S3, detecting the chain plates without defects to be qualified chain plates, detecting the chain plates with the defects to be unqualified chain plates, and sorting and outputting the qualified chain plates and the unqualified chain plates through a sorting and discharging mechanism.
The image preprocessing and the chain plate defect detection in the S2 are specifically as follows:
and carrying out image preprocessing on the link plate image acquired in real time, and sequentially detecting the three defects. The chain plate detection algorithm flow is shown in fig. 2.
Image pre-processing
The image preprocessing algorithm flow is shown in fig. 3:
(1) image noise reduction
The gaussian filter is also a linear filter, and can effectively suppress noise and smooth images. The principle of action is similar to that of an averaging filter, and the average value of pixels in a filter window is taken as output.
Two-dimensional Gaussian distribution of
Figure GDA0001935709540000131
(2) Image enhancement algorithm based on PALMSHE
Gray histogram equalization is an image enhancement processing means widely used. The essence of the method is that the image is subjected to nonlinear stretching, and the pixel values of the image are redistributed according to the cumulative pixel probability density function of the gray level histogram.
Although gray histogram equalization can highlight a major part of the image gray concentration, there are some side effects that are difficult to ignore. On one hand, the processing method forcibly changes the illumination information of the image, so that the average gray value of the enhanced image is always fixed to the intermediate gray level of 128, and on the other hand, the transformation mode across gray levels causes large-range shift of the gray levels, which easily causes the loss of part of statistical features of the original gray level histogram, and further causes the coverage of the gray level region information in the non-pixel concentration in the image and the loss of the image features contained in the gray level region information.
To avoid the situation that the gray scale interval is squeezed when the gray histogram is equalized, the most direct method is to preset a certain gray scale range for the high-frequency gray scale number when the equalization is expanded, and simultaneously correspondingly enhance the image information contained in the low-frequency gray scale. Therefore, a histogram equalization method between divided regions is considered.
The paper proposes an improved enhancement method of 'near-peak minimum binary equalization of a gray histogram' (PALMSHE) on the basis of an image enhancement method of gray histogram equalization. The main principle is as follows:
Figure GDA0001935709540000141
Figure GDA0001935709540000142
Figure GDA0001935709540000143
Figure GDA0001935709540000144
f(X(i,j))＝Xmin+(L1-Xmin)cdfx1(k(i,j))
f(X(i,j))＝L1+(Xmax-L1)cdfx2(k(i,j))
wherein L is1For dividing point gray scale, selecting the gray minimum value point nearest to the gray peak value of the histogram as the dividing point in the algorithm, L1Its corresponding gray scale. p is a radical ofx1(k),px2(k) Is a pixel frequency statistical function of the divided sub-gray level histogram;correspondingly, cdfx1(k)，cdfx2(k) Is a cumulative function of the respective pixel frequency statistics.
Passing pixel frequency minima L between high and low frequency gray levels1Dividing the low-frequency gray value interval and the high-frequency concentrated gray value interval, respectively counting the pixel frequency and the accumulated frequency of the two independent sub-histograms, and unifying the pixel frequency and the accumulated frequency to the final accumulated gray frequency, so that the accumulated gray frequency according to the equalization processing becomes a piecewise function in the intervals before and after the division of the gray scale, and simultaneously, the gray scale range after the equalization processing of each interval is limited during the equalization processing. Therefore, pixels belonging to different gray scale ratio intervals are subjected to equalized gray scale conversion operation in the original gray scale intervals of the pixels according to the histogram statistical frequency, cross-interval gray scale extrusion and transfer are avoided, the image feature information can be kept more perfectly while the image is enhanced, and the specific flow is shown in fig. 4.
Fig. 5 and 6 show the processing effect of the link plate part picture collected under the extreme illumination condition, and it can be seen from the processed image and the corresponding gray histogram: the illumination information and the histogram distribution characteristics of the link plate images collected under different illumination conditions are reserved, and the required link plate characteristic information is highlighted; the contrast of the chain plate relative to the background is enhanced, so that the subsequent image segmentation and extraction are facilitated; meanwhile, the edge parts are extruded during the equalization operation in the high-gray scale interval, so that the edge blurring condition caused by dynamic acquisition is improved to a certain extent. And a better image enhancement effect is realized.
4.2 Defect one detection: incomplete link joint profile
The detection algorithm flow is shown in fig. 7:
threshold segmentation
The threshold segmentation is a segmentation method based on a pixel gray value and region division technology, and the algorithm operation can be defined as:
Figure GDA0001935709540000151
wherein f (x, y) is the gray level of the original pixel point, g (x, y) is the processed pixel area, and T is the selected segmentation threshold.
The threshold segmentation algorithm is a segmentation operation based on image gray value difference, so that the segmentation processing effect is better for an image with high foreground and background contrast and large difference between the interested region to be segmented and the background gray value. The link plate picture belongs to the category of pictures with strong foreground and background contrast, the calculated amount of a threshold segmentation algorithm is small, the performance is stable, and the requirement of real-time detection and rapid processing detection is met.
The subject adopts the maximum variance method among classes in the global threshold segmentation method. The maximum variance method between classes was proposed by Otsu university in Japan in 1979, and is an adaptive threshold determination method, also called Otsu method (OTSU method). The method is derived on the basis of decision analysis or least square principle. The core idea is to divide the image into two parts of a foreground and a background, and when the difference between the two parts of the foreground and the background reaches the maximum value by the threshold value, the threshold value is the optimal threshold value. Assuming that the image is divided into two parts, C0 and C1, by a threshold T, the probabilities, means, and variances of the two parts, C0 and C1, are:
Figure GDA0001935709540000161
Figure GDA0001935709540000162
Figure GDA0001935709540000163
where i is the number of gray levels, Pi is the probability of gray level correspondence, L is the upper limit of the number of gray levels of the image, and μ is the global mean of the image
Figure GDA0001935709540000164
μ (T) is a gray-scale average value of the area image C0
Figure GDA0001935709540000165
The variance between regions C0 and C1 is defined as:
σB 2＝ω0(μ0-μ)2+ω1(μ1-μ)2＝ω0ω1(μ1-μ0)2
Figure GDA0001935709540000166
and when the inter-class variance value is maximum, the difference between the foreground and the background is maximum, and the corresponding T is the optimal threshold. The inter-class maximum variance method has higher operation speed compared with other threshold selection methods, and has better effect on the condition that double peaks or multiple peaks appear in the gray level histogram statistical chart, namely the contrast of the foreground and the background of the image is not clear enough. The effect graph of the inter-class variance threshold segmentation process is shown in fig. 8.
The acquisition distance between the industrial camera and the lens is kept unchanged in the image acquisition process. On the premise, the mathematical statistical information of the chain plate area in the picture and the actual chain plate size are always kept in a consistent proportion, so that the statistical information of the chain plate image area can be directly used as a judgment standard of chain plate defects.
Determining the number of pixel points of the largest incomplete chain plate image through a pre-experiment, and taking a slightly larger value as a discrimination threshold T1-1(the threshold value of the scheme is set to 165000), and the statistics C of pixel points in the chain plate area is calculatedopAnd comparing the data with a discrimination threshold, wherein the data which is larger than the discrimination threshold is a complete-contour chain plate, and the data which is smaller than the discrimination threshold is a defective chain plate.
Cop<T1-1
4.3 Defect two detection: displacement of punched hole
The detection algorithm flow is shown in fig. 9:
edge detection and circle fitting
The image edge in the acquired image means local step change of the image gray scale, and outlines the shape of the shot object. The edge detection can obviously reduce the data volume of the picture to be processed while keeping important characteristic information in the original picture, can improve the processing speed and meets the requirement of real-time detection.
The gray value variation along the edge direction is moderate, while the pixel gray value variation perpendicular to the edge direction is significant, which is a gray characteristic of edge detection. The corresponding detection method is to traverse and detect the mathematical operator according to the parallel and vertical preset directions, wherein the mathematical operator accords with the edge gray scale change characteristic.
The Deriche operator was proposed by Rachid Deriche in 1987. The theory behind the operator is still mainly the edge detection theory proposed by john Canny, and strictly complies with the relevant standards of Canny optimal edge extraction: extracting edge quality, accuracy and definition; and the operation processing steps are also consistent with the Canny operator steps. The Deriche operator is therefore often also referred to as the Canny-Deriche operator. The operation steps are as follows:
(1) denoising: smoothing images with gaussian filters
The two-dimensional Gaussian distribution G (x, y) is shown in equation (3.15)
g(x,y)＝G(x,y)*f(x,y)
Wherein f (x, y) is the original image, g (x, y) is the filtered image, and x is the convolution operation.
The corresponding gaussian transform matrix can be obtained by the above formula, for example, when σ is 1.4, the gaussian transform matrix of 5 × 5 is:
Figure GDA0001935709540000181
thus, a single pixel noise becomes almost unaffected on the Gaussian smoothed image.
(2) The magnitude and direction of the gradient is calculated using the finite difference of the first order partial derivatives.
(3) Non-maximum suppression is performed on the gradient amplitudes.
Obtaining a global gradient is not sufficient to determine the edge, and the point where the local gradient is the largest needs to be preserved, so the direction of the gradient needs to be used to suppress non-maxima therein.
(4) Edges are detected and connected using a dual threshold algorithm.
The dual threshold algorithm applies two thresholds τ 1 and τ 2, 2 τ 1 ≈ τ 2, to the non-maximum suppressed image, thereby obtaining two threshold edge images N1[i,j]And N2[i，j]. Due to N2[i，j]Obtained using a high threshold, and thus contains few false edges, but discontinuities (non-closures). The double threshold method is to be in N2[i，j]The algorithm is at N when the end points of the contour are reached1[i,j]Find edges that can be connected to the contour at the 8 neighbor locations, so the algorithm is constantly at N1[i,j]Until N is reached2[i,j]Is connected until[44]. And extracting the connected domain which is connected by 8 to obtain the image edge which meets the canny detection standard.
The main difference between the Deriche operator and the canny operator is that not gaussian filtering but an infinite impulse response filter is used in the filtering step. Its convolution mask becomes:
g(x)＝sxe-α|x|
where s is the transform constant coefficient and α is the adjustable parameter.
The infinite impulse response filter has the advantages that adaptive operation can be carried out on different pictures only by adjusting one parameter α, and when the noise of the pictures to be processed is high and more smoothing processing is needed, the operation speed of the Dericche operator is obviously superior to that of the Canny operator.
The Deriche operator processing effect graph is shown in FIG. 10, and the boundary and the defect edge are clear.
The detection precision of the traditional edge detection method can only reach one pixel level, but with the continuous improvement of the industrial detection precision requirement of modern high-end manufacturing industry, the pixel level detection method can not meet the requirement of actual measurement in some occasions, and the concept of sub-pixel edge detection is developed. A sub-pixel can be understood as an image processing technique that improves the processing accuracy by means of software algorithms, without changing the image acquisition hardware.
The application of the sub-pixel localization technique has certain preconditions: 1. the detected target is not an individual pixel point, but is composed of a plurality of pixel point clusters with obvious distribution characteristics; 2. the image characteristics of the detected target image can be analyzed and recognized, and the accurate position of the target can be determined.
The edge detection algorithms currently studied at the sub-pixel level can be summarized in 3 types: moment methods, interpolation methods and fitting methods.
The moment method has the advantages of simple calculation and capability of obtaining an analytic solution. However, the moment method is sensitive to image noise, and if a blurred edge model is considered, model parameters are increased, so that determination of an analytic solution becomes difficult.
The core of the interpolation method is to interpolate the gray value of the pixel point or the derivative of the gray value, and add information to realize sub-pixel edge detection. Among them, the methods which are more studied include quadratic interpolation, B-spline interpolation, chebyshev polynomial interpolation, and the like. The interpolation operation time is short, the quadratic interpolation algorithm is simple, the method can be realized by hardware, and the method is suitable for online detection. When the line spread function of the optical system is symmetrical, the accuracy of interpolation edge detection is higher. Interpolation is similar to the moment-based method in its characteristics, and is simple in calculation process, but is easily affected by noise.
The fitting method is to obtain the edge location of the sub-pixels by fitting the gray values of the assumed edge model. Because the fitting does not need numerical differentiation, and the fitting is carried out according to the minimum distance between each gray value and the fitting curve, the gray value with errors is reasonably utilized, and the influence of the gray value errors can be reduced, so that the fitting method is insensitive to noise. But the solving speed is slow due to the complex model.
Through test comparison, a Deriche method with the best effect is selected from the traditional edge detection methods, then the secondary interpolation fitting technology is used for carrying out interpolation operation on the detected discrete edge points, and each edge point is connected to obtain a sub-pixel level edge curve.
The effect graph of the Deriche operator processing for sub-pixel edge detection is shown in fig. 11.
The image edge set obtained through the processing comprises edge information of the link plate contour edge, the punching edge and the peripheral defects of the link plate contour edge and the punching edge. And then screening out the target edge line through the statistics of the relevant data of the edge line. The screening standard of selecting for use in this scheme is "edge line length" and "edge line head and the tail linear distance (closure)," draws out the edge of punching a hole of link joint from the edge picture.
Fitting of circles
The punched edge lines extracted by screening need to be fitted into a circle or a circular arc to further determine the position parameters of the punched edge lines. Similar to the method of line fitting, the most straightforward is the least squares method: a certain fitting circle is preset, and the sum of squares of distances from all points to the outline of the preset fitting circle is minimized.
Figure GDA0001935709540000211
Wherein (x)c,yc) Is the coordinate of the center of the fitting circle, and R is the radius of the fitting circle; the edge has n edge points with the coordinate of (x)i,yi) (ii) a d is the sum of the squares of the distances of the points to the fitted circle profile.
The equation has no analytic solution, and is approximately expressed as:
Figure GDA0001935709540000212
defining an auxiliary function:
h(x,y)＝(xi-xc)2+(yi-yc)2-R2
the original formula can be expressed as:
Figure GDA0001935709540000213
since the sum of squared differences f is greater than 0, the maximum value is infinite when there is a minimum value equal to or greater than 0. The following conditions are satisfied when the minimum value is taken:
Figure GDA0001935709540000214
wherein
Figure GDA0001935709540000221
Radius R is not 0, then
∑h(xi,yi)＝0
Another two formulas
Figure GDA0001935709540000222
Figure GDA0001935709540000223
To obtain
∑xih(xi,yi)＝0
∑yih(xi,yi)＝0
Is provided with
Figure GDA0001935709540000224
Figure GDA0001935709540000225
Figure GDA0001935709540000226
Figure GDA0001935709540000227
Figure GDA0001935709540000228
Figure GDA0001935709540000229
Derived by substituting into the preceding formula
∑uih(xi,yi)＝0
∑vih(xi,yi)＝0
Is unfolded to obtain
∑ui((ui-uc)2+(vi-vc)2-R2)＝0
∑vi((ui-uc)2+(vi-vc)2-R2)＝0
Further developed
∑(ui 3-2ui 2uc+uivi 2-2uivivc)＝0
∑(ui 2vi-2uiviuc+vi 3-2vi 2vc)＝0
Defining partial expressions
Suuu＝∑ui 3
Svvv＝∑vi 3
Suu＝∑ui 2
Svv＝∑vi 2
Suv＝∑uivi
Suuv＝∑ui 2vi
Suvv＝∑uivi 2
Substituted into the above formula
Figure GDA0001935709540000231
Figure GDA0001935709540000232
Solve uc and vc
Figure GDA0001935709540000233
Figure GDA0001935709540000234
To further obtain
Figure GDA0001935709540000235
Figure GDA0001935709540000236
R2＝∑((xi-xc)2+(yi-yc)2)
And obtaining the center coordinates and the radius values of the fitting circles.
The least squares circle fitting is greatly influenced by outlier noise, and the improved method is to introduce weight values to reduce the interference of outliers. An initial circle is fitted by least squares fitting, and then the distances from each contour point to the initial circle are used to calculate the proportion of each point in subsequent iterations. Thus, the circle fitting algorithm with strong robustness is obtained.
Since the concentration occurrence region of the third type of the defects of the link plate is in the vicinity of the edge of the punching circle, when the link plate has the third type of defects, excessive deformation in an extreme case affects the edge shape of the punching circle. It is therefore desirable to approximate the circle fitting method with the least squares of the dividing weights to enhance robustness.
The circle and the center position obtained by circle fitting are shown in fig. 12.
After fitting the circles, the center position coordinates (x) of the two punching circles are obtainedc1,yc1)(xc2,yc2) And radius r1 and r2, taking the midpoint of the connecting line between the centers of the two circles, and connecting the midpoint coordinate with the center coordinate (x) of the previously extracted chain plate connecting domain0,y0) Comparing the absolute value of the error with the error threshold (transverse error threshold T)2-125, longitudinal thresholdValue T2-255) is the unqualified chain plate with punching hole offset.
Figure GDA0001935709540000241
Figure GDA0001935709540000242
d1>T2-1
d2>T2-2
4.4 detection of defect three: edge defect of punched hole
The detection algorithm flow is shown in fig. 13:
ROI extraction and saddle point detection
ROI (region of interest) refers to a region of interest, and the ROI is extracted independently and then processed, so that the image processing calculation amount and the processing time can be effectively reduced. Common ROI extraction methods include threshold segmentation, edge extraction, and the like.
As mentioned above, the third defect is that the punching nail slightly inclines to press the link plate material at the punching edge to cause convex wrinkles and fine cracks, and the defect position is limited to the punching edge part. The two circular areas of the ring punch edge can thus be used as ROIs for defect three detection.
The two ROI areas are concentric circles of a punching fitting circle, and the center coordinates (x) are obtained according to the punching fitting circle stepc1,yc1)(xc2,yc2) And the radius r1, r2 is used as a concentric circle with a slightly larger radius, and pictures in the concentric circle region are extracted as the ROI for defect three detection.
And (4) performing edge extraction on the ROI picture area again, screening a defect curve from the extracted curve set, namely performing defect three detection by using an edge detection method. However, there is a possibility that edges of small-sized defects may be overlooked due to the influence of illumination and motion blur during edge detection. Other methods are needed to aid detection.
The saddle point, in three-dimensional space, is literally understood to be a middle point on a curved surface shaped like a saddle. Further extension is defined as: the points that are local maxima and local minima in the two perpendicular directions, respectively, are saddle points. The definition is popularized to a two-dimensional plane space, namely a matrix space for image display, namely: at a certain point in the matrix, the local maximum value of the row number column is located, and the local minimum value of the column number column is also located, so that the point is a saddle point. A schematic diagram of the saddle point is shown in fig. 14.
If the associated cracks around the extrusion defect are compared in a way of being vertical to the crack trend, each point on the crack is a local minimum point of the line where the associated crack is located, and therefore the local maximum point on the crack along the crack trend can be known as a saddle point according to the definition. In the ROI region where the gray value variation is gentle, the presence of the saddle point indicates the existence of such a secondary crack.
In order to improve the detection efficiency and avoid the interference of other chain plate miscellaneous points, a certain ROI interested detection area is specified and selected according to the central coordinates of a punching circle obtained by fitting the previous pitch circle in advance, and then a saddle point detection process is carried out in the selected detection area, so that the detection efficiency and the accuracy can be improved. In addition, since the saddle point detection needs to traverse the ROI, in order to reduce the detection time, the subject is to use the saddle point detection as an auxiliary detection means for edge detection to detect the third defect. The saddle point detection effect map is shown in fig. 15.
And (3) defect judgment standard: 1. the method comprises the following steps of (1) edge extraction, screening lines which accord with defect characteristics according to line length, and taking the number (> < 1) of extracted defect lines as a main judgment standard; and 2, the number (>2) of saddle points detected by the ROI area saddle points is used as an auxiliary judgment criterion.
The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims (3)

1. A chain plate defect online detection device based on machine vision comprises a feeding module, a transmission module and a discharging and sorting module, wherein the feeding module comprises a vibrating hopper for vibrating feeding, chain plates are arranged in the vibrating hopper, the transmission module comprises a conveyor belt, the discharging and sorting module comprises a sorting and discharging mechanism, a discharge port of the vibrating hopper is connected with one end of the conveyor belt and used for conveying the chain plates onto the conveyor belt, a feed port of the sorting and discharging mechanism is connected with the other end of the conveyor belt and used for sorting and outputting the chain plates, the chain plate defect online detection device is characterized by further comprising an image acquisition module, the image acquisition module comprises a trigger delay relay, an industrial camera and a computer comprising software for preprocessing images of the chain plates and detecting defects of the chain plates, the industrial camera is fixed right above the conveyor belt, and a lens of the industrial camera faces downwards and is perpendicular to the conveyor belt and used for acquiring the images of the, the trigger delay relay is positioned on one side of the industrial camera, is connected with the industrial camera and is used for triggering the industrial camera to acquire images of the chain plates when the chain plates are detected to be conveyed on the conveying belt, and the industrial camera is connected with the computer and is used for transmitting the acquired images of the chain plates to the computer for image preprocessing and detecting defects of the chain plates;
the method comprises the following steps:
s1, placing the chain plates to be detected into a vibration hopper, and opening a conveyor belt to convey the chain plates onto the conveyor belt;
s2, triggering the industrial camera to be started after the trigger delay relay detects the chain plate on the conveyor belt, so as to acquire the image of the chain plate on the conveyor belt and send the image to the computer for image preprocessing and chain plate defect detection;
s3, detecting that the chain plates without defects are qualified chain plates, detecting that the chain plates with defects are unqualified chain plates, and sorting and outputting the qualified chain plates and the unqualified chain plates through a sorting and discharging mechanism;
the image preprocessing and the chain plate defect detection performed by the computer in the step S2 are specifically as follows:
s21, image preprocessing:
s211, image denoising: by two-dimensional Gaussian distribution
Figure FDA0002374912700000011
Carrying out smoothing processing on a chain plate image input into a computer;
s212, an image enhancement algorithm based on PALMSHE: improving and enhancing the image by using gray value transformation g (x, y) to T [ f (x, y) ], wherein f (x, y) is an input image, g (x, y) is an output image, and T is a gray operation defined on the neighborhood of the point (x, y) for the gray value;
s213, continuously processing the image through binary equalization of the near peak minimum value of the gray level histogram, wherein the specific principle is as follows:
Figure FDA0002374912700000012
Figure FDA0002374912700000013
Figure FDA0002374912700000014
Figure FDA0002374912700000021
f(X(i,j))＝Xmin+(L1-Xmin)cdfx1(k(i,j))
f(X(i,j))＝L1+(Xmax-L1)cdfx2(k(i,j)) Wherein L is1For dividing point gray scale, selecting the gray minimum value point nearest to the gray peak value of the histogram as the dividing point in the algorithm, L1For its corresponding gray scale, px1(k),px2(k) Is a pixel frequency statistical function of the divided sub-gray level histogram; correspondingly, cdfx1(k)，cdfx2(k) Is a cumulative function of the respective pixel frequency statistics; dividing the low-frequency gray value interval and the high-frequency concentrated gray value interval by the pixel frequency minimum value point L1 between the high and low-frequency gray levels, further respectively counting the pixel frequency and the accumulation frequency of the two independent sub-histograms, and unifying the pixel frequency and the accumulation frequency into the final accumulation gray frequency, so that the accumulation gray frequency according to the equalization processing is before the division of the gray levelsThe back interval is changed into a piecewise function, and the gray level range after the equalization processing of each interval is limited during the equalization processing, so that pixels belonging to different gray scale ratio intervals perform equalized gray level conversion operation in each original gray level interval according to histogram statistical frequency, cross-interval gray level extrusion and transfer are avoided, and the image characteristic information can be more perfectly kept while the image is enhanced;
s22, detecting defects of the chain plates:
s221, detecting incomplete contours of the chain plates: the preprocessed image is processed by a threshold segmentation algorithm, and the formula of the threshold segmentation algorithm is
Figure FDA0002374912700000022
Wherein f (x, y) is the gray level of the original pixel point, g (x, y) is the processed pixel area, and T is the selected segmentation threshold;
setting a threshold value T to divide the image into two parts C0And C1Then C is0、C1The probability, mean and variance of each of the two parts are:
Figure FDA0002374912700000023
Figure FDA0002374912700000024
Figure FDA0002374912700000025
wherein i is the number of gray levels, PiIs the corresponding probability of gray level, L is the upper limit of the image gray level, mu is the global average value of the image
Figure FDA0002374912700000026
μ (T) is the area image C0Mean value of gray scale of
Figure FDA0002374912700000027
Region C0And C1The variance between is defined as:
σB 2＝ω0(μ0-μ)2+ω1(μ1-μ)2＝ω0ω1(μ1-μ0)2
Figure FDA0002374912700000031
when the inter-class variance value is maximum, the difference between the foreground and the background is maximum, and the corresponding T is the optimal threshold;
taking a slightly larger value as a discrimination threshold value T1-1Counting the pixel points in the link plate area by CopComparing the detected data with a discrimination threshold, wherein the data which is larger than the discrimination threshold is a complete-contour chain plate, and the data which is smaller than the discrimination threshold is a incomplete chain plate;
s222, punching shift detection: smoothing the preprocessed image by using a two-dimensional Gaussian distribution G (x, y), wherein the formula of the two-dimensional Gaussian distribution G (x, y) is as follows: g (x, y) ═ G (x, y) × f (x, y), where f (x, y) is the original image, G (x, y) is the filtered image, and × is the convolution operation; obtaining a corresponding Gaussian filter transformation matrix through the formula of the two-dimensional Gaussian distribution G (x, y);
calculating the magnitude and direction of the gradient by using the finite difference of the first-order partial derivatives;
carrying out non-maximum suppression on the gradient amplitude;
edges are detected and connected with a dual threshold algorithm: the dual threshold algorithm applies two thresholds τ 1 and τ 2, 2 τ 1 ≈ τ 2, to the non-maximum suppressed image, thereby obtaining two threshold edge images N1[i,j]And N2[i，j]Due to N2[i，j]Obtained using a high threshold, and thus contains few false edges, but is discontinuous, i.e., not closed; the double threshold method is to be in N2[i，j]The algorithm is at N when the end points of the contour are reached1[i,j]8 neighbor location finding to connect to the edge on the contour, the algorithm is constantly at N1[i,j]Until N is reached2[i,j]Until being connected, the method can be obtained by extracting the connected domain with 8 connectedThe image edge meeting the canny detection standard is obtained;
performing interpolation operation on the detected discrete edge points by adopting a Deriche method and applying a quadratic interpolation fitting technology, and connecting the edge points to obtain a sub-pixel level edge curve; the image edge set obtained through the processing comprises the edge information of the link plate outline edge, the punched edge and the peripheral defects of the link plate outline edge, and the punched edge of the link plate is extracted from the edge image through the length of the edge line and the head-tail linear distance of the edge line;
adopting a least square method:
Figure FDA0002374912700000032
wherein (x)c,yc) Is the coordinate of the center of the fitting circle, and R is the radius of the fitting circle; the edge has n edge points with the coordinate of (x)i,yi) (ii) a d is the sum of the squares of the distances from the points to the contour of the fitting circle, and a certain fitting circle is preset, so that the sum of the squares of the distances from all the points to the contour of the preset fitting circle is minimized, because
Figure FDA0002374912700000033
Without an analytical solution, it is expressed approximately as:
Figure FDA0002374912700000034
defining an auxiliary function: h (x, y) ═ xi-xc)2+(yi-yc)2-R2Then, the original formula is represented as:
Figure FDA0002374912700000041
the sum of the squared differences f is greater than 0, so that when a minimum value greater than or equal to 0 exists, the maximum value is infinite, and the minimum value satisfies the following condition:
Figure FDA0002374912700000042
wherein
Figure FDA0002374912700000043
Radius R is not 0, then∑h(xi,yi) 0, additionally:
Figure FDA0002374912700000044
Figure FDA0002374912700000045
to obtain
∑xih(xi,yi)＝0
∑yih(xi,yi)＝0，
Is provided with
Figure FDA0002374912700000046
Derived by substituting into the preceding formula
∑uih(xi,yi)＝0
∑vih(xi,yi)＝0，
Is unfolded to obtain
∑ui((ui-uc)2+(vi-vc)2-R2)＝0
∑vi((ui-uc)2+(vi-vc)2-R2)＝0，
Further developed
∑(ui 3-2ui 2uc+uivi 2-2uivivc)＝0
∑(ui 2vi-2uiviuc+vi 3-2vi 2vc)＝0，
Defining partial expressions
Figure FDA0002374912700000051
Substituted into the above formula
Figure FDA0002374912700000052
Figure FDA0002374912700000053
Get u from solutioncAnd vc
Figure FDA0002374912700000054
To further obtain
Figure FDA0002374912700000055
Obtaining the center coordinates and radius values of the fitting circle, and obtaining the center position coordinates (x) of the two punching circles after fitting the circlec1,yc1)(xc2,yc2) And radius r1 and r2, taking the midpoint of the connecting line between the centers of the two circles, and connecting the midpoint coordinate with the center coordinate (x) of the previously extracted chain plate connecting domain0,y0) Comparing, wherein the absolute value of the error is larger than the error threshold value, namely the transverse error threshold value T2-125, longitudinal threshold T2-255, namely the unqualified chain plate with punching deviation,
Figure FDA0002374912700000061
s223, detecting the punching edge defects: the circle center coordinate (x) obtained according to the step of punching and fitting a circlec1,yc1)(xc2,yc2) And the radius r1, r2 is used as a concentric circle with a slightly larger radius, and pictures in the concentric circle region are extracted to be used as an ROI for detecting punching edge defects; and (4) extracting the edge of the ROI picture area again, and screening lines which accord with the defect characteristics by using the line length in the edge extraction so as to obtain the number of the extracted defect lines>1 is taken as a main discrimination standard; ROI areaNumber of saddle points for saddle point detection>2 as the secondary criterion.
2. The machine vision-based chain plate defect online detection device as claimed in claim 1, wherein: the image acquisition module also comprises an illumination light source fixed below the lens of the industrial camera.
3. The machine vision-based chain plate defect online detection device as claimed in claim 1, wherein: the trigger delay relay is a proximity switch.
CN201810534908.9A 2018-05-30 2018-05-30 Chain plate defect online detection device and method based on machine vision Active CN109829876B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810534908.9A CN109829876B (en) 2018-05-30 2018-05-30 Chain plate defect online detection device and method based on machine vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810534908.9A CN109829876B (en) 2018-05-30 2018-05-30 Chain plate defect online detection device and method based on machine vision

Publications (2)

Publication Number Publication Date
CN109829876A CN109829876A (en) 2019-05-31
CN109829876B true CN109829876B (en) 2020-04-07

Family

ID=66859498

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810534908.9A Active CN109829876B (en) 2018-05-30 2018-05-30 Chain plate defect online detection device and method based on machine vision

Country Status (1)

Country Link
CN (1) CN109829876B (en)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110672617B (en) * 2019-09-14 2022-06-14 华南理工大学 Method for detecting defects of silk-screen area of glass cover plate of smart phone based on machine vision
CN110930359A (en) * 2019-10-21 2020-03-27 浙江科技学院 Method and system for detecting automobile shifting fork
CN111982921B (en) * 2020-05-21 2023-11-03 北京安视中电科技有限公司 Method and device for detecting hole defects, conveying platform and storage medium
TWI742733B (en) * 2020-06-19 2021-10-11 倍利科技股份有限公司 Image conversion method
CN111981990B (en) * 2020-08-03 2022-02-01 无锡鼎茂机械制造有限公司 Machine vision-based self-sealing zipper quality online detection method
CN112629424B (en) * 2020-11-24 2021-09-24 江南大学 High-precision sandwich valve outer diameter dynamic vision measurement system and method
CN112691926A (en) * 2020-12-10 2021-04-23 苏州阿甘机器人有限公司 Automatic quality detection device based on machine vision and working method thereof
CN113008894A (en) * 2021-01-26 2021-06-22 宁波诺视智能科技有限公司 Chain appearance detection device and method
CN112792450B (en) * 2021-02-01 2021-12-07 深圳大学 Optical fiber automatic focusing method and automatic focusing system for laser processing
CN112927218B (en) * 2021-03-23 2023-07-25 上海晨兴希姆通电子科技有限公司 Method and system for detecting defects of CPU (Central processing Unit) heat dissipation plate
CN113139451A (en) * 2021-04-16 2021-07-20 北京京东乾石科技有限公司 Abnormal information generation method and device, electronic equipment and computer readable medium
CN113129304B (en) * 2021-05-18 2022-09-16 郑州轻工业大学 Part detection method based on machine vision
CN113379744B (en) * 2021-08-12 2021-11-19 山东大拇指喷雾设备有限公司 Nozzle device surface defect detection method and system based on image processing
CN113777033A (en) * 2021-08-18 2021-12-10 长沙长泰机器人有限公司 Raw sliver defect detection method and device based on machine vision
CN113916892B (en) * 2021-09-28 2022-10-28 南京理工大学 Multi-view vision-based brake disc gluing defect detection device and method
CN113899752B (en) * 2021-09-30 2023-01-10 湖北天宜机械股份有限公司 Method for detecting crack of chain plate by adopting visual detection device for crack of inner chain plate of double-row chain scraper conveyor
CN114359260B (en) * 2022-01-18 2023-11-03 成都理工大学 Method and device for detecting defects on surface of tobacco rod
CN114742757A (en) * 2022-03-09 2022-07-12 红塔烟草(集团)有限责任公司 Non-blind-area visual detection method for glue line of filter rod forming paper
CN114693652B (en) * 2022-04-01 2023-08-25 深圳市科宝图像技术有限公司 Fabric Defect Detection Method Based on Gaussian Mixture Model
CN114998311A (en) * 2022-07-11 2022-09-02 江苏名欧精密机械有限公司 Part precision detection method based on homomorphic filtering
CN116363136B (en) * 2023-06-01 2023-08-11 山东创元智能设备制造有限责任公司 On-line screening method and system for automatic production of motor vehicle parts
CN116612112B (en) * 2023-07-17 2023-09-22 鑫脉(山东)工贸有限公司 Visual inspection method for surface defects of bucket
CN116758057A (en) * 2023-08-10 2023-09-15 山东贺铭电气有限公司 Communication equipment defect detection method based on artificial intelligence
CN117708615B (en) * 2024-02-04 2024-05-17 芯联集成电路制造股份有限公司 Product detection method and processing equipment monitoring method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8103085B1 (en) * 2007-09-25 2012-01-24 Cognex Corporation System and method for detecting flaws in objects using machine vision
CN103454285A (en) * 2013-08-28 2013-12-18 南京师范大学 Transmission chain quality detection system based on machine vision
CN104128314A (en) * 2014-07-17 2014-11-05 大连集智立方科技有限公司 Material sorting device
CN206557124U (en) * 2017-03-28 2017-10-13 苏满调 A kind of carrier bar defect detecting device
CN206849687U (en) * 2017-06-30 2018-01-05 淮安市政旺电子有限公司 The forming and cutting sorting dress alms bowl integrated device of vertical inductance core
CN107991093A (en) * 2017-11-28 2018-05-04 绩溪山合机械有限公司 A kind of detection device that can detect carrier bar Short Item automatically

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8103085B1 (en) * 2007-09-25 2012-01-24 Cognex Corporation System and method for detecting flaws in objects using machine vision
CN103454285A (en) * 2013-08-28 2013-12-18 南京师范大学 Transmission chain quality detection system based on machine vision
CN104128314A (en) * 2014-07-17 2014-11-05 大连集智立方科技有限公司 Material sorting device
CN206557124U (en) * 2017-03-28 2017-10-13 苏满调 A kind of carrier bar defect detecting device
CN206849687U (en) * 2017-06-30 2018-01-05 淮安市政旺电子有限公司 The forming and cutting sorting dress alms bowl integrated device of vertical inductance core
CN107991093A (en) * 2017-11-28 2018-05-04 绩溪山合机械有限公司 A kind of detection device that can detect carrier bar Short Item automatically

Also Published As

Publication number Publication date
CN109829876A (en) 2019-05-31

Similar Documents

Publication Publication Date Title
CN109829876B (en) Chain plate defect online detection device and method based on machine vision
CN114820773B (en) Silo transport vehicle carriage position detection method based on computer vision
CN109242853B (en) PCB defect intelligent detection method based on image processing
CN116758083B (en) Quick detection method for metal wash basin defects based on computer vision
CN106780486A (en) A kind of Surface Defects in Steel Plate image extraction method
CN109781737B (en) Detection method and detection system for surface defects of hose
CN112378350B (en) Flatness detection method for PIN PIN of network transformer
CN109377450A (en) A kind of edge-protected denoising method
CN114972343A (en) Method for detecting production abnormity of laser welding product
CN110648330B (en) Defect detection method for camera glass
CN116091504B8 (en) Connecting pipe connector quality detection method based on image processing
CN113506246B (en) Concrete 3D printing component fine detection method based on machine vision
CN115100174B (en) Ship sheet metal part paint surface defect detection method
CN114926407A (en) Steel surface defect detection system based on deep learning
CN108020554A (en) A kind of steel strip surface defect recognition detection method
CN115049651B (en) Metal plate stamping abnormity detection method
CN106780437B (en) A kind of quick QFN chip plastic packaging image obtains and amplification method
CN115797361B (en) Aluminum template surface defect detection method
CN114549441A (en) Sucker defect detection method based on image processing
CN115330758A (en) Welding quality detection method based on denoising processing
CN115953398A (en) Defect identification method for strip steel surface
CN116168025A (en) Oil curtain type fried peanut production system
CN115953373A (en) Glass defect detection method and device, electronic equipment and storage medium
CN114155226A (en) Micro defect edge calculation method
CN108805854B (en) Method for rapidly counting tablets and detecting completeness of tablets in complex environment

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