CN106053479A - System for visually detecting workpiece appearance defects based on image processing - Google Patents

Abstract

The invention discloses a visual detection system for workpiece appearance defects based on image processing, which includes an industrial computer, a coaxial light source, a CCD industrial camera, an image acquisition card and a rejecting mechanism; both the coaxial light source and the rejecting mechanism are connected with the industrial computer; the CCD The industrial camera is connected to the industrial computer through the image acquisition card; among them: the coaxial light source is used to provide the diffuse reflection light source for the workpiece to be detected; the CCD industrial camera is used to capture the image of the workpiece in the detection station; Eliminate defective workpieces through detection; the industrial computer has a defect detection module based on image processing; the system has high detection efficiency and is easy to implement.

Description

A Visual Inspection System for Workpiece Appearance Defects Based on Image Processing

technical field

The invention belongs to the field of automatic detection, in particular to a visual detection system for workpiece appearance defects based on image processing.

Background technique

The main production process of metal workpieces is machining, stamping, precision casting, powder metallurgy, metal injection molding, dimensional inspection, appearance defect inspection, etc. Affected by the manufacturing process throughout the production process, the size and appearance of the workpiece will be unqualified to a certain extent. Among them, appearance defects mainly include: gaps, sticky materials, cracks, indentations, pinholes, scratches and blisters, etc. If workpieces with appearance quality defects flow into the next production process, the assembly will be blocked and deformed, which will affect the quality of the assembled parts. potential economic losses and reputational risks.

Traditional appearance defect inspection methods include manual visual inspection and stroboscopic light inspection. The speed of the automated production line is very fast, and the human eye cannot quickly capture accurate defect information, especially some small defects, and the human naked eye cannot distinguish whether it is qualified or not, which leads to the problems of low defect detection accuracy and high false detection rate . Stroboscopic light detection is mainly based on the static response of the human retina to a certain pulse of light. The method is to combine a specific camera and a stroboscopic light source to determine the surface condition of the workpiece by fixedly observing the detector. The disadvantage is that the reliability of the detection results is low, and the degree of automatic detection is also low.

The automatic detection technology includes infrared, flood current and magnetic flux leakage detection technology. These three detection methods are also widely used in our country at present. The eddy current testing technology is mainly to detect the flow resistance defects of the lower layer of the workpiece surface, but it consumes a lot of power, which causes a waste of energy for the production enterprise. The eddy current testing method has relatively high requirements on the quality of the workpiece itself. The surface of the workpiece must be pure and free of impurities, the temperature must be uniform, and the speed of the conveyor belt is relatively slow. This results in limited production and testing, and cannot meet the high-speed, high-quality production requirements. In recent years, the continuous development of machine vision and image processing technology has made it possible for machine vision inspection to replace manual inspection on the production line. Visual detection and recognition of appearance defects is the most effective and promising method. High-resolution industrial cameras can provide rich workpiece appearance image information, and can accurately, efficiently and reliably complete the detection and identification of workpiece appearance defects.

At present, the main methods of visual detection of workpiece appearance defects: (1) The automatic detection of metal workpiece surface defects is realized through genetic algorithm and visual image processing morphology. (2) The existence of product defects is judged by using the grayscale feature of the image and the abnormal change of the grayscale value. However, due to the strong reflective properties of the metal surface, the system has a high false detection rate.

Therefore, it is necessary to design an efficient and high-precision visual inspection system for workpiece appearance defects based on image processing.

Contents of the invention

The technical problem to be solved by the present invention is to provide a visual inspection system for workpiece appearance defects based on image processing, which has high detection efficiency and is easy to implement.

The technical solution of the invention is as follows:

A visual detection system for workpiece appearance defects based on image processing, characterized in that it includes an industrial computer, a coaxial light source, a CCD industrial camera, an image acquisition card and a rejection mechanism;

Both the coaxial light source and the rejecting mechanism are connected with the industrial computer;

The CCD industrial camera is connected to the industrial computer through the image acquisition card;

Among them: the coaxial light source is used to provide diffuse reflection light source for the workpiece to be inspected; the CCD industrial camera is used to capture the image of the workpiece at the inspection station; the reject mechanism is used to remove the defective workpiece from the production line after passing the inspection;

The industrial computer has a defect detection module based on image processing;

The defect detection module based on image processing implements defect detection according to the following steps:

Step 1: Work piece image acquisition and preprocessing;

Step 2: Image segmentation and workpiece pose correction;

Step 3: Inspect for the following cosmetic defects: nicks, sticks, cracks, indentations, pinholes, scratches, and blisters.

In step 1, the workpiece image f(x, y) is collected by using the CCD industrial camera and the image acquisition card through the coaxial light source, and the workpiece image is a grayscale image, and then the workpiece image is sent to the industrial computer for preprocessing, and the preprocessing is Perform median filter processing on the collected workpiece images.

In step 2:

(1) Image segmentation:

Segment the preprocessed image based on the histogram method. The gray histogram of the workpiece image will show two peaks: one is the workpiece as the foreground, and the other is the background. The gray value of the valley is taken as the segmentation threshold to effectively segment the foreground. and background:

In the formula, F(x,y) is the segmented workpiece image, and Th _f is the segmentation threshold

(2) Image correction is to realize the translation and rotation angle correction of the workpiece in the image through affine transformation. Affine transformation is an existing mature technology.

Perform mathematical morphological processing on the corrected image; the processing process is to apply a morphological opening operation to the image through the structural element B to remove the burr on the edge of the workpiece and smooth the edge of the workpiece, as follows:

In the formula, is the opening operator, is the erosion operator, ⊕ is the expansion operator, B is the structure element, the size is 3, and all elements are 1, which is a disc structure.

In step 3:

Mark the edge of the workpiece as the gap detection area, denoted as Reg _qk ;

Calibrate the entire surface area of the workpiece as the detection area of sticky material, cracking, indentation, pinhole, scratch and blister defect; where the sticky material and pinhole detection area is marked as Reg _nl and Reg _zy ; the scratch and cracking detection area is marked as Reg _hh And Reg _kl ; Indentation detection area is recorded as Reg _yh ; Bubble detection area is recorded as Reg _qp ;

Defect area judgment threshold:

In the formula, Th is the area judgment threshold of the defect; φ is the defect tolerance; W and H are the width and height of the workpiece in the image, in pixels; M and N are the actual length and width of the workpiece, in millimeters;

Local dynamic segmentation threshold determination method:

Firstly, the filter mask of (2D+1)×(2D+1) is used for smoothing, where D is the diameter of the extracted target; then the mean (x, y) and standard value of the smoothed image gray value are calculated. difference σ(x,y); when the extracted target appears as a bright pixel, select T=Mean(x,y)+γ·σ(x,y) as the segmentation threshold; when the extracted target appears as a dark pixel, Select T=Mean(x,y)-γ·σ(x,y) as the segmentation threshold, where γ is the standard deviation intensity.

① Gap detection:

1) Use the image segmentation threshold T _qk to segment the Blob candidate block of the gap in the region Reg _qk , and identify the Blob connected domain through the eight connected regions, which is denoted as B1 _qk ;

Determination of the segmentation threshold T _qk : use (2D _qk +1) × (2D _qk +1) filter mask for smoothing, where D _qk is the diameter of the notch defect; calculate the mean value Mean of the smoothed image gray value _qk (x, y) and standard deviation σ _qk (x, y); since gap defects appear as dark pixels, T _qk = Mean _qk (x, y)-γ _qk σ _qk (x, y) is selected as the segmentation Threshold, γ _qk is the standard deviation weight of the gap defect. The value range of the weight is [0,1], and the specific value needs to be determined according to prior knowledge.

2) Utilize the pixel counting method to extract the pixel area feature Area _qk of the Bl _qk connected domain; judge whether Bl _qk is a gap defect according to the following formula:

In the formula, the defect area judgment threshold Th _qk is determined by Equation 3, where the value range of φ is [0.0120,0.0130], YES and NO indicate the presence or absence of notch defects, respectively;

② Sticky material and needle hole detection:

1) Use the segmentation threshold T _zz to segment the Blob candidate blocks of sticky material and needle holes in the regions Reg _nl and Reg _zy , and identify the Blob connected domains through eight connected regions, which are denoted as Bl _nl and Bl _zy ;

Determination of the segmentation threshold T _zz :

T _zz = Mean' _zz (x,y)-δ _zz V' _zz (x,y)

In the formula, Mean' _zz (x, y) and V' _zz (x, y) are the mean and variance of pixel gray values in the detection area, and δ _zz is the variance weight of sticky material and pinhole defects;

The pixel area below the segmentation threshold in the image is a defect candidate block;

2) Using the pixel counting method to extract the pixel area feature Area _nl and roundness feature of the Bl _nl connected domain Pixel Area Feature Area _zy and Roundness Feature of Bl _zy Connected Domain According to the following formulas 5 and 6, judge whether Bl _nl is a sticky material defect and whether Bl _zy is a pinhole defect:

In the formula, the defect area judgment thresholds Th _nl1 and Th _zy1 are determined by formula 3, where the value ranges of φ are [0.0020,0.0021] and [0.0024,0.0025] respectively; the defect roundness judgment thresholds Th _nl2 and Th _zy2 are The ranges are [0.5,1] and [0.85,1] respectively; ∩ represents logical "AND"operation; YES and NO represent yes and no respectively;

The pixel area feature is the number of pixels in the area, and the roundness feature description is the ratio of the area of the target area to the area of the circumscribed circle. The closer the shape is to a circle, the closer the ratio is to 1. The value range of the roundness feature is: The calculation formula is Where r is the radius of the circumscribed circle of the extracted target, where the extracted target refers to sticky material and pinhole defects;

③Scratch and crack detection:

1) Use the local image variance strength algorithm to obtain the segmentation threshold T _hk , segment the scratched and cracked blob block candidates in the regions Reg _hh and Reg _kl , and identify the Blob connected domain through eight connected regions, which are recorded as Bl _hh and Bl _kl ;

Local image variance intensity is an extension of the concept of image local threshold. Since the detected workpiece will have an uneven background due to the influence of the production process, it is difficult to find a fixed threshold to completely segment the target defect from the background. Therefore, a local threshold detection method is proposed, that is, a method of combining local gray features with the whole; combined with the characteristics of local variance and variance, first use (2D _hk +1) × (2D _hk +1) filter mask for smoothing, In the formula, D _hk is the length of scratches and cracking defects; then calculate the standard deviation σ _hk (x, y) and variance V _hk (x, y) of the image gray value after smoothing; the segmentation threshold is determined by the following formula:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; ;</span>$

Where σ′ _hk (x, y) and V′ _hk (x, y) represent the standard deviation and variance of the entire image before smoothing;

2) Utilize the pixel counting method to extract the pixel area feature Area _hh and the inner longest diameter feature Diameter _hh of the Bl _hh connected domain and the pixel area feature Area _kl and the inner longest diameter feature Diameter _kl of the Bl _kl connected domain; according to formula 7,8 Determine whether Bl _hh is a scratch defect, and whether Bl _kl is a crack defect:

In the formula, the defect area judgment thresholds Th _hh1 and Th _kl1 are determined by formula 3, where the value ranges of φ are [0.0110,0.0120] and [0.0048,0.0049] respectively; the longest defect diameter judgment thresholds Th _hh2 and Th _kl2 The value range is determined by the empirical value; ∩ represents the logic "AND"operation;

The longest internal diameter is the distance between the two furthest pixels on the boundary of the area. The distance and area are both in pixels, that is, the number of pixels contained within the distance or within the area;

④ Indentation detection:

1) Segment the indented Blob candidate blocks through the Laplacian Gaussian transform algorithm and the local dynamic threshold T _yh ;

Determination of T _yh : use (2D _yh +1) × (2D _yh +1) filter mask for smoothing, where D _yh is the diameter of the indentation defect; calculate the average Mean _yh ( x, y) and standard deviation σ _yh (x, y); since indentation defects appear as bright pixels in the image after Laplacian-Gaussian transformation, T _yh = Mean _yh (x, y)+γ _yh σ _yh (x, y) is the segmentation threshold, and γ _yh is the standard deviation weight of indentation defects; Blob connected domains are identified through eight connected areas, denoted as B _yh ;

The value range of the weight is [0,1], and the specific value needs to be determined according to prior knowledge;

2) Use the pixel counting method to extract the pixel area feature Area _yh and the rectangularity feature Rectan _yh of the Bl _yh connected domain. The rectangularity describes the fullness of the extracted area to its circumscribed rectangle. The calculation formula is Among them, S _m is the area of the rectangular area circumscribing the extracted area; judge whether B _yh is an indentation defect according to formula 9:

In the formula, the area judgment threshold Th _yh1 is determined by formula 3, where the value range of φ is [0.0160,0.0170]; the value range of the rectangularity judgment threshold Th _yh2 is [0.7,1]; ∩ represents the logic "AND"operation;

Laplacian of Gauss algorithm:

The method is a combination of Gaussian filtering and Laplacian operator. The main steps of the algorithm are as follows:

(1) Filtering: First, smoothing and filtering the image F(x,y) is performed, and the filtering function is a Gaussian function, namely

$\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&pi;\&sigma;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; \&lsqb;</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&pi;\&sigma;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>$

Convolving the image F(x,y) with G(x,y) can get a smooth image, namely

g(x,y)=F(x,y)*G(x,y)

(2) Image enhancement: Laplacian operation is performed on the smooth image g(x,y), namely

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&dtri;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>$

Since the Laplacian operation on the smooth image g(x,y) is equivalent to the convolution of the Laplacian operation of g(x,y) and F(x,y), the above formula becomes:

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{<span\; class="notranslate">\; \&dtri;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>$

In the formula becomes the LOG filter, which is:

${<span\; class="notranslate">\; \&dtri;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{<span\; class="notranslate">\; \&part;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; G</span>}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; +</span>\frac{{<span\; class="notranslate">\; \&part;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; G</span>}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&pi;\&sigma;</span>}}^{\mathrm{<span\; class="notranslate">\; 4</span>}}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>$

Function: The Laplacian Gaussian operator combines the Gaussian smoothing filter and the Laplacian sharpening filter to flatten the noise first, and then enhance the edge of the image, so it can effectively highlight the indentation defect.

⑤ Blister detection:

1) Transform the image function from the spatial domain to the frequency domain through fast Fourier transform, use a low-pass filter to smooth the image, and then transform the image from the frequency domain to the spatial domain through inverse Fourier transform; according to the gray histogram of the image, select the valley gray The degree value is the blob candidate block of the segmentation threshold segmentation target segmentation foam, and the blob connected domain is identified by eight connected areas, which is denoted as Bl _qp ;

The histogram of the image has only one trough, because after Fourier transform and smoothing filter processing, the bubbling part is brighter than the overall workpiece background, so the histogram shows two peaks and one trough, and one of the two peaks belongs to the background of the workpiece. The other belongs to the bubble defect, so the background and the bubble defect can be separated by using the trough gray value;

2) Using the pixel counting method to extract the pixel area feature Area _qp and roundness feature of the Bl _qp connected domain Judging whether Bl _qp is a bubble defect according to formula 10:

In the formula, the area judgment threshold Th _qp1 is determined by formula 3, where the value range of φ is [0.0123,0.0124]; the value range of the roundness judgment threshold Th _qp2 is [0.5,1]; ∩ represents the logic "AND" operation . During the acquisition process of positioning and guiding images, the circular diffuse reflection light source is used for illumination; during the detection image acquisition process, the coaxial light source is used for illumination.

Beneficial effect:

The visual inspection system for workpiece appearance defects based on image processing of the present invention adopts a coaxial light source to provide illumination for the workpiece, uses a camera and an image acquisition card to obtain images of the workpiece, and uses an industrial computer to realize defect detection based on image processing, and detects defective workpieces by The elimination mechanism is used for elimination; in the process of defect detection, the robot is first guided by the vision system, and the target workpiece pose is accurately located according to the template matching algorithm based on the gray value, and then the appearance defect detection of the workpiece is carried out. The steps are: (1) Obtain the image of the workpiece , using median filter for preprocessing; (2) using the global threshold to segment the target workpiece, and correcting the workpiece pose; (3) removing the burr interference on the edge of the workpiece through mathematical morphology opening operation; (4) detecting gaps, sticky materials, Cracks, indentations, pinholes, scratches and blistered appearance defects.

The advantages of the present invention are:

1. Fast positioning speed and high precision. Template matching based on gray value, using normalized cross-correlation algorithm, and using image pyramid to achieve multi-level matching, improving matching accuracy and speed;

2. Defect detection is highly targeted and fast.

Using a simple and effective Blob algorithm, firstly preprocess the acquired original image, suppress noise interference and enhance the expressive tension of useful information in the image. The present invention adopts a targeted detection method for different defects, which is easy to implement and the method is ingenious. Specifically, it includes calibrating the detection area and segmenting the ROI through binarization, and extracting the defect features of the gap based on the local gray threshold based on the ROI area; the local dynamic threshold algorithm Extract sticky and pinhole defect features; local image variance intensity algorithm extracts scratch and crack defect features; Laplacian-Gaussian algorithm extracts surface indentation defect features; fast Fourier transform algorithm extracts surface blister defect features. Finally, the extracted defect features are analyzed and judged according to the judgment rules in the judgment rule base, and the detection results are output;

3. Wide adaptability, strong transplantability, and more comprehensive types of detectable defects. The detection algorithm can be widely used in the visual detection of the appearance quality of precision workpieces on high-speed production lines, and can be transplanted to the visual detection production lines of electronic components, small parts, etc. It has strong adaptability and is a highly versatile and accuracy of the visual inspection system.

The system of the invention can solve the problems of slow manual detection speed, low efficiency and poor precision; can overcome the problems of single defect type, poor imaging quality and high false detection rate in the current visual detection, and improve the automation degree and product quality of precision workpiece production.

Description of drawings

Figure 1 is a flow chart of visual detection of workpiece appearance defects based on image processing;

Figure 2 is the detected normal image and defect image, where Figures a-h correspond to normal, sticky, indentation, notch, crack, scratch, blister and pinhole images respectively;

Fig. 3 is a block diagram of the overall structure of the visual inspection system for workpiece appearance defects based on image processing.

detailed description

In order to facilitate the understanding of the present invention, the invention will be described more comprehensively and in detail below in conjunction with the accompanying drawings and preferred embodiments, but the protection scope of the present invention is not limited to the following specific embodiments.

Unless otherwise defined, all technical terms used hereinafter have the same meanings as commonly understood by those skilled in the art. The terminology used herein is only for the purpose of describing specific embodiments, and is not intended to limit the protection scope of the present invention.

Example 1:

As shown in Figure 1-3, a visual inspection system for workpiece appearance defects based on image processing is characterized in that it includes an industrial computer, a coaxial light source, a CCD industrial camera, an image acquisition card and a rejection mechanism;

Both the coaxial light source and the rejecting mechanism are connected with the industrial computer;

The CCD industrial camera is connected to the industrial computer through the image acquisition card;

Among them: the coaxial light source is used to provide diffuse reflection light source for the workpiece to be detected; the CCD industrial camera is used to photograph the

The image of the workpiece; the removal mechanism is used to remove the defective workpiece from the production line;

The industrial computer has a defect detection module based on image processing;

The defect detection module based on image processing implements defect detection according to the following steps:

Step 1: Work piece image acquisition and preprocessing;

Step 2: Image segmentation and workpiece pose correction;

Step 3: Inspect for the following cosmetic defects: nicks, sticks, cracks, indentations, pinholes, scratches, and blisters.

In step 1, the workpiece image f(x, y) is collected by using the CCD industrial camera and the image acquisition card through the coaxial light source, and the workpiece image is a grayscale image, and then the workpiece image is sent to the industrial computer for preprocessing, and the preprocessing is Perform median filter processing on the collected workpiece images to remove the noise that may be caused during image capture and transmission, and improve the image signal-to-noise ratio.

In step 2:

(1) Image segmentation:

Segment the preprocessed image based on the histogram method. The gray histogram of the workpiece image will show two peaks: one is the workpiece as the foreground, and the other is the background. The gray value of the valley is taken as the segmentation threshold to effectively segment the foreground. and background:

In the formula, F(x,y) is the segmented workpiece image, and Th _f is the segmentation threshold

(2) Image correction is to realize the translation and rotation angle correction of the workpiece in the image through affine transformation. Affine transformation is an existing mature technology.

Perform mathematical morphological processing on the corrected image; the processing process is to apply a morphological opening operation to the image through the structural element B to remove the burr on the edge of the workpiece and smooth the edge of the workpiece, as follows:

In the formula, ° is the opening operator, is the erosion operator, ⊕ is the expansion operator, B is the structural element, the size is 3,

The elements are all 1, which is a disc structure.

In step 3:

Mark the edge of the workpiece as the gap detection area, denoted as Reg _qk ;

Calibrate the entire surface area of the workpiece as the detection area of sticky material, cracking, indentation, pinhole, scratch and blister defect; where the sticky material and pinhole detection area is marked as Reg _nl and Reg _zy ; the scratch and cracking detection area is marked as Reg _hh And Reg _kl ; Indentation detection area is recorded as Reg _yh ; Bubble detection area is recorded as Reg _qp ;

Defect area judgment threshold:

In the formula, Th is the judgment threshold of the defect area; φ is the defect tolerance; W and H are the width and height of the workpiece in the image, in pixels; M and N are the actual length and width of the workpiece, in millimeters; Local dynamic segmentation threshold determination method:

Firstly, the filter mask of (2D+1)×(2D+1) is used for smoothing, where D is the diameter of the extracted target; then the mean (x, y) and standard value of the smoothed image gray value are calculated. difference σ(x,y); when the extracted target appears as a bright pixel, select T=Mean(x,y)+γ·σ(x,y) as the segmentation threshold; when the extracted target appears as a dark pixel, Select T=Mean(x,y)-γ·σ(x,y) as the segmentation threshold, where γ is the standard deviation intensity.

① Gap detection:

1) Use the image segmentation threshold T _qk to segment the Blob candidate block of the gap in the region Reg _qk , and identify the Blob connected domain through the eight connected regions, which is denoted as B1 _qk ;

Determination of the segmentation threshold T _qk : use (2D _qk +1) × (2D _qk +1) filter mask for smoothing, where D _qk is the diameter of the notch defect; calculate the mean value Mean of the smoothed image gray value _qk (x, y) and standard deviation σ _qk (x, y); since gap defects appear as dark pixels, T _qk = Mean _qk (x, y)-γ _qk σ _qk (x, y) is selected as the segmentation Threshold, γ _qk is the standard deviation weight of the gap defect. The value range of the weight is [0,1], and the specific value needs to be determined according to prior knowledge.

2) Utilize the pixel counting method to extract the pixel area feature Area _qk of the Bl _qk connected domain; judge whether Bl _qk is a gap defect according to the following formula:

In the formula, the defect area judgment threshold Th _qk is determined by Equation 3, where the value range of φ is [0.0120,0.0130], YES and NO indicate the presence or absence of notch defects, respectively;

② Sticky material and needle hole detection:

1) Use the segmentation threshold T _zz to segment the Blob candidate blocks of sticky material and needle holes in the regions Reg _nl and Reg _zy , and identify the Blob connected domains through eight connected regions, which are denoted as Bl _nl and Bl _zy ;

Determination of the segmentation threshold T _zz :

T _zz = Mean' _zz (x,y)-δ _zz V' _zz (x,y)

In the formula, Mean' _zz (x, y) and V' _zz (x, y) are the mean and variance of pixel gray values in the detection area, and δ _zz is the variance weight of sticky material and pinhole defects;

The pixel area below the segmentation threshold in the image is a defect candidate block;

2) Using the pixel counting method to extract the pixel area feature Area _nl and roundness feature of the Bl _nl connected domain Pixel Area Feature Area _zy and Roundness Feature of Bl _zy Connected Domain According to the following formulas 5 and 6, judge whether Bl _nl is a sticky material defect and whether Bl _zy is a pinhole defect:

In the formula, the defect area judgment thresholds Th _nl1 and Th _zy1 are determined by formula 3, where the value ranges of φ are [0.0020,0.0021] and [0.0024,0.0025] respectively; the defect roundness judgment thresholds Th _nl2 and Th _zy2 are The ranges are [0.5,1] and [0.85,1] respectively; ∩ means logic "AND"operation; YES and NO;

The pixel area feature is the number of pixels in the area, and the roundness feature description is the ratio of the area of the target area to the area of the circumscribed circle. The closer the shape is to a circle, the closer the ratio is to 1. The value range of the roundness feature is: The calculation formula is Where r is the radius of the circumscribed circle of the extracted target, where the extracted target refers to sticky material and pinhole defects;

③Scratch and crack detection:

1) Use the local image variance strength algorithm to obtain the segmentation threshold T _hk , segment the scratched and cracked blob block candidates in the regions Reg _hh and Reg _kl , and identify the Blob connected domain through eight connected regions, which are recorded as Bl _hh and Bl _kl ;

Local image variance intensity is an extension of the concept of image local threshold. Since the detected workpiece will have an uneven background due to the influence of the production process, it is difficult to find a fixed threshold to completely segment the target defect from the background. Therefore, a local threshold detection method is proposed, that is, a method of combining local gray features with the whole; combined with the characteristics of local variance and variance, first use (2D _hk +1) × (2D _hk +1) filter mask for smoothing, In the formula, D _hk is the length of scratches and cracking defects; then calculate the standard deviation σ _hk (x, y) and variance V _hk (x, y) of the image gray value after smoothing; the segmentation threshold is determined by the following formula:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; ;</span>$

Where σ′ _hk (x, y) and V′ _hk (x, y) represent the standard deviation and variance of the entire image before smoothing;

2) Utilize the pixel counting method to extract the pixel area feature Area _hh and the inner longest diameter feature Diameter _hh of the Bl _hh connected domain and the pixel area feature Area _kl and the inner longest diameter feature Diameter _kl of the Bl _kl connected domain; according to formula 7,8 Determine whether Bl _hh is a scratch defect, and whether Bl _kl is a crack defect:

In the formula, the defect area judgment thresholds Th _hh1 and Th _kl1 are determined by formula 3, where the value ranges of φ are [0.0110,0.0120] and [0.0048,0.0049] respectively; the longest defect diameter judgment thresholds Th _hh2 and Th _kl2 The value range is determined by the empirical value; ∩ represents the logic "AND"operation;

The longest internal diameter is the distance between the two furthest pixels on the boundary of the area. The distance and area are both in pixels, that is, the number of pixels contained within the distance or within the area;

④ Indentation detection:

1) Segment the indented Blob candidate blocks through the Laplacian Gaussian transform algorithm and the local dynamic threshold T _yh ;

Determination of T _yh : use (2D _yh +1) × (2D _yh +1) filter mask for smoothing, where D _yh is the diameter of the indentation defect; calculate the average Mean _yh ( x, y) and standard deviation σ _yh (x, y); since indentation defects appear as bright pixels in the image after Laplacian-Gaussian transformation, T _yh = Mean _yh (x, y)+γ _yh σ _yh (x, y) is the segmentation threshold, and γ _yh is the standard deviation weight of indentation defects; Blob connected domains are identified through eight connected areas, denoted as B _yh ;

The value range of the weight is [0,1], and the specific value needs to be determined according to prior knowledge;

2) Use the pixel counting method to extract the pixel area feature Area _yh and the rectangularity feature Rectan _yh of the Bl _yh connected domain. The rectangularity describes the fullness of the extracted area to its circumscribed rectangle. The calculation formula is Among them, S _m is the area of the rectangular area circumscribing the extracted area; judge whether B _yh is an indentation defect according to formula 9:

In the formula, the area judgment threshold Th _yh1 is determined by formula 3, where the value range of φ is [0.0160,0.0170]; the value range of the rectangularity judgment threshold Th _yh2 is [0.7,1]; ∩ represents the logic "AND"operation;

⑤ Blister detection:

1) Transform the image function from the spatial domain to the frequency domain through fast Fourier transform, use a low-pass filter to smooth the image, and then transform the image from the frequency domain to the spatial domain through inverse Fourier transform; according to the gray histogram of the image, select the valley gray The degree value is the blob candidate block of the segmentation threshold segmentation target segmentation foam, and the blob connected domain is identified by eight connected areas, which is denoted as Bl _qp ;

The histogram of the image has only one trough, because after Fourier transform and smoothing filter processing, the bubbling part is brighter than the overall workpiece background, so the histogram shows two peaks and one trough, and one of the two peaks belongs to the background of the workpiece. The other belongs to the bubble defect, so the background and the bubble defect can be separated by using the trough gray value;

2) Using the pixel counting method to extract the pixel area feature Area _qp and roundness feature of the Bl _qp connected domain Judging whether Bl _qp is a bubble defect according to formula 10:

In the formula, the area judgment threshold Th _qp1 is determined by formula 3, where the value range of φ is [0.0123,0.0124]; the value range of the roundness judgment threshold Th _qp2 is [0.5,1]; ∩ represents the logic "AND" operation .

Image preprocessing also includes:

A. Workpiece positioning:

The workpiece template image is Temp(x,y), and the direction is based on the positive direction of the X axis at 0 degrees. Illuminated by the circular diffuse reflection light source, using the CCD industrial camera and the image acquisition card to collect the image of the workpiece on the conveyor belt, and then searching for the target area matching the known template Temp(x,y) according to the template matching technology based on the gray value, and calculating Its center of gravity coordinates (x _c , y _c ) and deflection angle θ. The template matching adopts the normalized cross-correlation algorithm (NCC), and uses the image pyramid to achieve multi-level matching, which improves the matching accuracy and speed. The NCC algorithm formula is as follows:

$\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\mathrm{<span\; class="notranslate">\; T</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}{\sqrt{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; .</span>\frac{\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}{\sqrt{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}$

In the formula, n is the number of pixels in the region of interest of the template; Temp(x, y) is the template image, and the size of the template is W _T × H _T . The template image is the target workpiece image that is segmented, and the direction is specified based on the positive direction of the X-axis at 0 degrees; the area of interest is the target workpiece area, because the contrast between the workpiece area and the background area is very obvious, so the global grayscale threshold is used for segmentation The target artifact, that is, the region of interest, can be extracted using this method. m _T is the average gray value of the template, is the variance of the gray value of all pixels in the template, m _f (a,b) and is the average gray value and variance of all pixels in the image in the region of interest of the template translated to the current position of the image, ncc(a,b) represents the matching similarity, and the value range is -1≤ncc(a,b)≤1 . x, y represent pixel coordinates in the image. a, b are the image pixel coordinate translation amount.

Matching is to find the target workpiece in the field of view and quickly and accurately calculate the pose information of the workpiece. It is a prerequisite for subsequent robotics and vision processing.

B. Robot vision guidance:

The vision guidance system is mainly for the precise positioning of the target workpiece on the conveyor belt. For the same workpiece, the robot only needs to teach once and record the teaching position as the zero position. During production, the camera captures the target workpiece on the conveyor belt and calculates the coordinates of the central feature point of the workpiece on the image, that is, the pose information (x _c ,y _c ) and θ. The vision system calculates the deviation between the coordinates of the current target workpiece and the zero position in the X, Y, R _Z directions, and the robot plans the grasping path and action according to the deviation, and completes the task of grasping the target workpiece. The R _Z direction is the rotation direction of the workpiece on the plane.

Description of the overall process: The workpiece to be inspected first enters the upper surface inspection station, the photoelectric sensor triggers the camera to take a picture and collects a frame of image, detects defects through the appearance inspection algorithm, and transmits the inspection results to the lower computer through the visual inspection system. Unqualified products are removed by the rejecting device, and qualified products will enter the robot inspection station. Using vision to guide the robot to accurately pick up the target workpiece on the conveyor belt and place it in the detection position, the robot sequentially triggers multiple cameras to take pictures to detect the appearance defects of the front and rear, left and right sides and bottom surface of the workpiece. The industrial control computer comprehensively analyzes the multi-camera processing results and transmits them to the lower computer through the visual inspection system, and finally realizes the intelligent sorting of workpieces.

Claims (5)

1. A visual detection system based on image processing workpiece appearance defect, is characterized in that, comprises industrial computer, coaxial light source, CCD industrial camera, image acquisition card and rejecting mechanism; Both the coaxial light source and the rejecting mechanism are connected with the industrial computer; The CCD industrial camera is connected to the industrial computer through the image acquisition card; Among them: the coaxial light source is used to provide diffuse reflection light source for the workpiece to be inspected; the CCD industrial camera is used to capture the image of the workpiece at the inspection station; the reject mechanism is used to remove the defective workpiece from the production line after passing the inspection; The industrial computer has a defect detection module based on image processing; The defect detection module based on image processing implements defect detection according to the following steps: Step 1: Work piece image acquisition and preprocessing; Step 2: Image segmentation and workpiece pose correction; Step 3: Inspect for the following cosmetic defects: nicks, sticks, cracks, indentations, pinholes, scratches, and blisters. 2. the visual detection system of the workpiece appearance defect based on image processing according to claim 1, is characterized in that, in step 1, by coaxial light source illumination, utilizes CCD industrial camera and image acquisition card to gather workpiece image f(x, y), the workpiece image is a grayscale image, and then the workpiece image is sent to the industrial computer for preprocessing, and the preprocessing is to perform median filtering on the collected workpiece image. 3. the visual detection system of the workpiece appearance defect based on image processing according to claim 2, is characterized in that, in step 2: (1) Image segmentation: Segment the preprocessed image based on the histogram method. The gray histogram of the workpiece image will show two peaks: one is the workpiece as the foreground, and the other is the background. The gray value of the valley is taken as the segmentation threshold to effectively segment the foreground. and background: In the formula, F(x,y) is the segmented workpiece image, and Th _f is the segmentation threshold (2) Image correction is to realize the translation and rotation angle correction of the workpiece in the image through affine transformation. 4. the visual detection system of the workpiece appearance defect based on image processing according to claim 3, is characterized in that, carries out mathematical morphology processing to the image after correction; The calculation removes the burr on the edge of the workpiece and smoothes the edge of the workpiece, including: In the formula, is the opening operator, is the corrosion operator, is the inflation operator, B is the structuring element with a size of 3, The elements are all 1, which is a disc structure. 5. The visual detection system of the workpiece appearance defect based on image processing according to claim 3 or 4, characterized in that, in step 3: Mark the edge of the workpiece as the gap detection area, denoted as Reg _qk ; Calibrate the entire surface area of the workpiece as the detection area of sticky material, cracking, indentation, pinhole, scratch and blister defect; where the sticky material and pinhole detection area is marked as Reg _nl and Reg _zy ; the scratch and cracking detection area is marked as Reg _hh And Reg _kl ; Indentation detection area is recorded as Reg _yh ; Bubble detection area is recorded as Reg _qp ; Defect area judgment threshold: In the formula, Th is the judgment threshold of the defect area; φ is the defect tolerance; W and H are the width and height of the workpiece in the image, Take pixels as the unit; M and N are the actual length and width of the workpiece, in millimeters; ① Gap detection: 1) Use the image segmentation threshold T _qk to segment the Blob candidate block of the gap in the region Reg _qk , and identify the Blob connected domain through the eight connected regions, which is denoted as B1 _qk ; Determination of the segmentation threshold T _qk : use (2D _qk +1) × (2D _qk +1) filter mask for smoothing, where D _qk is the diameter of the notch defect; calculate the mean value Mean of the smoothed image gray value _qk (x, y) and standard deviation σ _qk (x, y); since gap defects appear as dark pixels, T _qk = Mean _qk (x, y)-γ _qk σ _qk (x, y) is selected as the segmentation Threshold, γ _qk is the standard deviation weight of the gap defect. 2) Utilize the pixel counting method to extract the pixel area feature Area _qk of the Bl _qk connected domain; judge whether Bl _qk is a gap defect according to the following formula: In the formula, the defect area judgment threshold Th _qk is determined by Equation 3, where the value range of φ is [0.0120,0.0130], YES and NO indicate the presence or absence of notch defects, respectively; ② Sticky material and needle hole detection: 1) Use the segmentation threshold T _zz to segment the Blob candidate blocks of sticky material and needle holes in the regions Reg _nl and Reg _zy , and identify the Blob connected domains through eight connected regions, which are denoted as Bl _nl and Bl _zy ; Determination of the segmentation threshold T _zz : T _zz = Mean' _zz (x,y)-δ _zz V' _zz (x,y) In the formula, Mean' _zz (x, y) and V' _zz (x, y) are the mean and variance of pixel gray values in the detection area, and δ _zz is the variance weight of sticky material and pinhole defects; The pixel area below the segmentation threshold in the image is a defect candidate block; 2) Using the pixel counting method to extract the pixel area feature Area _nl and roundness feature of the Bl _nl connected domain Pixel Area Feature Area _zy and Roundness Feature of Bl _zy Connected Domain According to the following formulas 5 and 6, judge whether Bl _nl is a sticky material defect and whether Bl _zy is a pinhole defect: In the formula, the defect area judgment thresholds Th _nl1 and Th _zy1 are determined by formula 3, where the value ranges of φ are [0.0020,0.0021] and [0.0024,0.0025] respectively; the defect roundness judgment thresholds Th _nl2 and Th _zy2 are The ranges are [0.5,1] and [0.85,1] respectively; ∩ represents logical "AND"operation; YES and NO represent yes and no respectively; The pixel area feature is the number of pixels in the area, and the roundness feature description is the ratio of the area of the target area to the area of the circumscribed circle. The closer the shape is to a circle, the closer the ratio is to 1. The value range of the roundness feature is: The calculation formula is Where r is the radius of the circumscribed circle of the extracted target, where the extracted target refers to sticky material and pinhole defects; ③Scratch and crack detection: 1) Use the local image variance strength algorithm to obtain the segmentation threshold T _hk , segment the scratched and cracked blob block candidates in the regions Reg _hh and Reg _kl , and identify the Blob connected domain through eight connected regions, which are recorded as Bl _hh and Bl _kl ; Local image variance intensity is an extension of the concept of image local threshold. Since the detected workpiece will have an uneven background due to the influence of the production process, it is difficult to find a fixed threshold to completely segment the target defect from the background. Therefore, a local threshold detection method is proposed, that is, a method of combining local gray features with the whole; combined with the characteristics of local variance and variance, first use (2D _hk +1) × (2D _hk +1) filter mask for smoothing, In the formula, D _hk is the length of scratches and cracking defects; then calculate the standard deviation σ _hk (x, y) and variance V _hk (x, y) of the image gray value after smoothing; the segmentation threshold is determined by the following formula: ${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; ;</span>$ Where σ′ _hk (x, y) and V′ _hk (x, y) represent the standard deviation and variance of the entire image before smoothing; 2) Utilize the pixel counting method to extract the pixel area feature Area _hh and the inner longest diameter feature Diameter _hh of the Bl _hh connected domain and the pixel area feature Area _kl and the inner longest diameter feature Diameter _kl of the Bl _kl connected domain; according to formula 7,8 Determine whether Bl _hh is a scratch defect, and whether Bl _kl is a crack defect: In the formula, the defect area judgment thresholds Th _hh1 and Th _kl1 are determined by formula 3, where the value ranges of φ are [0.0110,0.0120] and [0.0048,0.0049] respectively; the longest defect diameter judgment thresholds Th _hh2 and Th _kl2 The value range is determined by the empirical value; ∩ represents the logic "AND"operation; The longest internal diameter is the distance between the two furthest pixels on the boundary of the area. The distance and area are both in pixels, that is, the number of pixels contained within the distance or within the area; ④ Indentation detection: 1) Segment the indented Blob candidate blocks through the Laplacian Gaussian transform algorithm and the local dynamic threshold T _yh ; Determination of T _yh : use (2D _yh +1) × (2D _yh +1) filter mask for smoothing, where D _yh is the diameter of the indentation defect; calculate the average Mean _yh ( x, y) and standard deviation σ _yh (x, y); since indentation defects appear as bright pixels in the image after Laplacian-Gaussian transformation, T _yh = Mean _yh (x, y)+γ _yh σ _yh (x, y) is the segmentation threshold, and γ _yh is the standard deviation weight of indentation defects; Blob connected domains are identified through eight connected areas, denoted as B _yh ; 2) Use the pixel counting method to extract the pixel area feature Area _yh and the rectangularity feature Rectan _yh of the Bl _yh connected domain. The rectangularity describes the fullness of the extracted area to its circumscribed rectangle. The calculation formula is Among them, S _m is the area of the rectangular area circumscribing the extracted area; judge whether B _yh is an indentation defect according to formula 9: In the formula, the area judgment threshold Th _yh1 is determined by formula 3, where the value range of φ is [0.0160,0.0170]; the value range of the rectangularity judgment threshold Th _yh2 is [0.7,1]; ∩ represents the logic "AND"operation; ⑤ Blister detection: 1) Transform the image function from the spatial domain to the frequency domain through fast Fourier transform, use a low-pass filter to smooth the image, and then transform the image from the frequency domain to the spatial domain through inverse Fourier transform; according to the gray histogram of the image, select the valley gray The degree value is the blob candidate block of the segmentation threshold segmentation target segmentation foam, and the blob connected domain is identified by eight connected areas, which is denoted as Bl _qp ; 2) Using the pixel counting method to extract the pixel area feature Area _qp and roundness feature of the Bl _qp connected domain Judging whether Bl _qp is a bubble defect according to formula 10: In the formula, the area judgment threshold Th _qp1 is determined by formula 3, where the value range of φ is [0.0123,0.0124]; the value range of the roundness judgment threshold Th _qp2 is [0.5,1]; ∩ represents the logic "AND" operation .