CN102680494A - Real-time detecting method of metal arc plane flaw based on machine vision - Google Patents

Abstract

The present invention relates to a machine vision-based method for real-time detection of flaws on polished metal arcuate surfaces, which includes the following steps: Step 1. In an offline situation, acquire N first sample images and M second sample images, and perform data fusion Step 2, set up the image histogram after fusion, obtain the linear relationship between the image background and the corresponding gray value; Step 3, calculate the reflection component of the selected first sample image and the second sample image; Step 4, establish The corresponding relationship between the standard deviation of the reflection component of the qualified workpiece and the corresponding gray value; step 5, obtain the detection image of the polished metal arc-shaped workpiece online in real time, and calculate the first standard deviation and the second standard deviation; step 6, perform the threshold Segment to obtain the corresponding binary image; step 7, compare the area of the connected region in the binary image with the preset judgment threshold, and judge the flaw of the polished metal arc surface. The invention has the advantages of convenient operation, high detection precision, good detection adaptability, stability and reliability.

Description

Real-time detection method of polished metal arc-shaped surface defects based on machine vision

technical field

The invention relates to a detection method, in particular to a machine vision-based real-time detection method for flaws on polished metal arcuate surfaces, belonging to the technical field of polished metal arcuate surface detection.

Background technique

In industrial production, with the continuous improvement of industrial technology, the quality requirements of products also increase. The traditional monitoring of workpiece quality mainly depends on manual visual inspection. Due to the influence of the subjective factors of the inspectors, it is easy to have false detections and missed detections, and manual visual inspection has low efficiency, low accuracy and insufficient standardization. The detection results are closely related to the personal ability and mental state of the inspectors. , the stability and reliability are relatively poor. In addition, the detection data classification cannot be sent to the computer in real time for automatic quality management. In order to solve the problems of heavy workload, low efficiency and high missed detection rate of manual visual inspection, enterprises urgently need to introduce an automatic detection technology to replace manual operation, so as to achieve strict control of product quality while reducing labor costs. Therefore, the application of machine vision in workpiece quality monitoring is introduced. Product quality is monitored by machine vision instead of traditional human eyes, which improves production efficiency and product quality.

At present, many scholars at home and abroad are also studying the flaw detection of polished arc-shaped metal surfaces, such as cylindrical metal workpieces such as the outer surface of bearings. There are still several problems in the detection of such workpieces: (1) Due to the high reflectivity of the surface of the polished arc-shaped workpiece, the reflection angle of the reflected light received by the camera is different, so it is difficult to obtain an image with uniform illumination and a wide format; ( 2) The gray distribution of the collected image is uneven, and the effective area is too narrow. If coaxial light is used as the light source, the effective area is too narrow, and at least 60 images are required to detect a bearing through experimental analysis. In order to solve the problem of uneven illumination, the background image can be extracted, and then the image can be enhanced according to the background image to obtain an image with uniform gray scale. But when the workpiece has a large defect, the extraction of the background image will have errors.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art and provide a machine vision-based real-time detection method for polished metal arc-shaped surface flaws, which is easy to operate, high in detection accuracy, good in detection adaptability, stable and reliable.

According to the technical solution provided by the present invention, a machine vision-based method for real-time detection of flaws on polished metal arcuate surfaces, the method for real-time detection of flaws on polished metal arcuate surfaces includes the following steps:

Step 1. Offline, obtain the first sample images of N qualified workpieces under normal working light and the second sample images of M qualified workpieces under low light, and perform median Filter to obtain image sequence I _n , n=1,2,...,N, respectively perform median filtering on M second sample images to obtain image sequence I _m ,m=1,2,...,M, for image sequence I _n and the image sequence I _m respectively carry out data fusion to obtain images g(x, y), h(x, y);

Step 2. Count the corresponding gray value and the number of pixels corresponding to the gray value in the fused image g(x, y), h(x, y), and obtain the image g(x, y), h(x, y) respectively ) histogram, according to the gray value corresponding to the corresponding larger wave peak in the histogram, the linear relationship between the image background I(x, y) and the gray value corresponding to the larger wave peak in the histogram is established as

I(x,y)=a(x,y)*(Z _max -Z _h )+h(x,y)

Among them, Z _max represents the gray value corresponding to the larger wave peak in the histogram, and a(x, y) is the slope matrix;

Step 3. Select a sample image from the above-mentioned first sample image and the second sample image, and establish the histogram of the selected first sample image and the second sample image to obtain the selected first sample image , the gray value corresponding to the larger wave peak value in the histogram of the second sample image, according to the gray value and the linear relational expression in step 2, respectively obtain the reflection components of the selected first sample image and the second sample image;

Step 4. Obtain the reflection components of the selected first sample image and the second sample image according to step 3, and establish the corresponding relationship between the standard deviation of the reflection component of the qualified workpiece and the gray value corresponding to the larger wave peak in the histogram:

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

Among them, σ(Z _max ) represents the standard deviation of the reflection component of a qualified workpiece, Z _{max_n} represents the gray value corresponding to the maximum peak value in the histogram of the selected first sample image, and Z _{max_0} represents the gray value in the histogram of the selected second sample image The gray value corresponding to the maximum peak value, σ _n is the standard deviation of the selected first sample image, and σ ₀ is the standard deviation of the selected second sample image;

Step 5. Obtain the detection image of the polished metal arc-shaped workpiece in real time online under the working light, perform median filtering on the detection image, and establish a histogram of the detection image; according to the gray value corresponding to the larger wave peak in the histogram and the above The first step is to obtain the reflection component of the detection image; Gaussian filtering is performed on the reflection component of the detection image to obtain the second reflection component, and the second standard deviation of the second reflection component is calculated, and the detection image is calculated according to step 4 at the corresponding gray level The first standard deviation of the qualified workpiece image;

Step 6. Comparing the first standard deviation and the second standard deviation to select a segmentation threshold, and performing threshold segmentation on the second reflection component through the segmentation threshold to obtain a corresponding binary image;

Step 7. Scanning the above binary image, marking different connected areas in the binary image, counting the area of the connected area, comparing the area of the connected area with a preset judgment threshold, and judging the flaws of the polished metal arc surface.

Described step 1 comprises the following steps:

Step 1.1. In the offline situation, collect the first sample images of N qualified workpieces under working light, and then reduce to the required light intensity, and collect the second sample images of M qualified workpieces under low light intensity;

Step 1.2, performing median filtering on the first sample image and the second sample image to obtain image sequences I _n , I _m respectively; where, n=1, 2,..., N, m=1, 2,...M;

将图像序列I_n进行数据融合，其中，(x,y)表示图像序列内的像素位置；Step 1.3, according to

Carry out data fusion with the image sequence I _n , wherein, (x, y) represents the pixel position in the image sequence;

将图像序列I_m进行数据融合。Step 1.4, according to

The image sequence I _m is subjected to data fusion.

Described step 2 comprises the following steps:

Step 2.1, the grayscale of the image is 0~255, the number of pixels corresponding to the grayscale value of the image g(x,y) is counted, and then divided by the total number of pixels of the image g(x,y) to obtain the image g(x,y) y) histogram p(z), and calculate the gray value Z _g that reaches the larger peak value;

Step 2.2, according to step 2.1, obtain the histogram q(z) of the image h(x, y), and calculate the gray value Z _h that reaches a larger peak value;

Step 2.3, establish the linear slope matrix between the image background I (x, y) and the corresponding gray value of the larger wave peak in the histogram, and obtain

$\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}<span\; class="notranslate">\; ;</span>$

Step 2.4, obtain the linear relationship between the image background I (x, y) and the larger wave peak in the histogram according to step 2.3, as

I(x,y)=a(x,y)*(Z _max -Z _h )+h(x,y),

Wherein, Z _max represents the gray value corresponding to the larger wave peak in the histogram.

Described step 3 comprises the steps:

Step 3.1, choose a sample image from the above N first sample images and M second sample images, count the number of pixels corresponding to the gray value in the selected first sample image and the second sample image, To establish the histogram of the selected first sample image and the second sample image, obtain the gray value corresponding to the larger wave peak value in the selected first sample image and the second sample image histogram;

Step 3.2, according to step 2 and the gray value corresponding to the larger wave peak value, obtain the background component I (x, y) respectively corresponding to the selected first sample image and the second sample image;

求取所选第一样本图像、第二样本图像的反射分量，其中，k为常数，f(x,y)为所选第一样本图像或第二样本图像。Step 3.3, according to

Calculating reflection components of the selected first sample image and the second sample image, where k is a constant, and f(x, y) is the selected first sample image or the second sample image.

Described step 4 comprises the steps:

Step 4.1, calculate the mean value μ of the reflection component of the selected first sample image, and obtain

$\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Among them, n*m represents the size of the reflection component image;

Step 4.2, calculate the standard deviation σ _n of the reflection component of the selected first sample image according to step 4.1, and obtain

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ;</span>$

Step 4.3, calculate the standard deviation σ ₀ of the reflection component of the selected second sample image according to steps 4.1 and 4.2;

Step 4.4, establish the corresponding relationship between the standard deviation of the reflection component of the qualified workpiece and the gray value Z _max of the larger wave peak in the histogram, and obtain,

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

Among them, σ(Z _max ) represents the standard deviation of the reflection component of a qualified workpiece, Z _{max_n} represents the gray value corresponding to the maximum peak value in the histogram of the selected first sample image, and Z _{max_0} represents the gray value in the histogram of the selected second sample image The gray value corresponding to the maximum peak value.

Described step 5 comprises the steps:

Step 5.1. Under the working light conditions, collect the workpiece detection image t(x, y), and perform median filtering on the detection image t(x, y);

Step 5.2, after establishing the median filter, detect the histogram of the image, and obtain the gray value Z _t corresponding to the larger wave peak in the histogram according to the histogram of the detected image;

Step 5.3, according to the gray value Z _t and step 2, obtain the background component I _t (x, y) of the detected image;

Step 5.4, according to step 5.3 and step 3, obtain the reflection component r _t (x, y) of the detected image;

Step 5.5, performing Gaussian filtering on the reflection component r _t (x, y) to generate a second reflection component R(x, y);

Step 5.6, obtain the first standard deviation σ(Z _t ) of the detected image according to the gray value Z _t and step 4.4;

Step 5.7. Calculate the second standard deviation σ _t of the second reflection component R(x, y) according to Step 4.1 and Step 4.2.

Described step 6 comprises the steps:

Step 6.1, compare the relationship between the first standard deviation σ(Z _t ) and the second standard deviation σ _t to get

$\mathrm{<span\; class="notranslate">\; sigma</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\end{array}\right.$

Step 6.2. Set the segmentation upper threshold T ₁ and lower threshold T ₂ to obtain

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; sigma</span>}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; sigma</span>}\end{array}\right.$

Among them, λ is the adjustment coefficient;

Step 6.3, using the upper threshold T ₁ and the lower threshold T ₂ to perform threshold segmentation on the second background component R(x,y) to obtain a binary image b(x,y), and obtain

$\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right.$

Among them, 0 is the background area, and 1 is the area to be processed.

Described step 7 comprises the steps:

Step 7.1, scan the binary image b(x, y), and mark the different connected regions in the binary image b(x, y);

Step 7.2, count the areas of different connected regions in the binary image b(x, y) respectively, and set the required judgment threshold S; when the area of connected regions in the binary image b(x, y) is greater than the judgment threshold S, then corresponding The connected area is a defective area; when the area of the connected area in the binary image b(x, y) is smaller than the judgment threshold S, the corresponding connected area is a normal area.

In the step 5.5, the convolution template of the Gaussian filter is

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 16</span>}}<span\; class="notranslate">\; *</span>\left[\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\\ \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 4</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]<span\; class="notranslate">\; .</span>$

The adjustment coefficient λ is 3-4.

Advantages of the present invention: first learn and analyze the brightness distribution of the measured workpiece in the working state in the offline state, and the statistical characteristics of the qualified workpiece; Filtering, threshold segmentation and other operations are used to perform real-time and accurate defect detection on workpieces, with convenient operation, high detection accuracy, good detection adaptability, and stability and reliability.

Description of drawings

FIG. 1 is a schematic diagram of the structure of the present invention for collecting images.

Fig. 2 is a flowchart of the present invention.

Explanation of reference numerals: 100-industrial camera, 110-support frame, 120-measured workpiece, 130-diffused plane light, 140-base and 150-suspension frame.

Detailed ways

The present invention will be further described below in conjunction with specific drawings and embodiments.

The purpose of the present invention is to carry out real-time detection of flaws on polished metal arc-shaped surfaces. Since the surface of polished metal arc-shaped surfaces is smooth, the diffuse reflectance is low, and the surface is arc-shaped, the intensity of light emitted by surfaces at different spatial positions is different, so it is difficult to obtain Wide image with uniform brightness. According to this problem, a method based on offline learning knowledge for real-time detection of polished metal arc-shaped surfaces and online real-time detection of defects is proposed.

As shown in Figure 1, when collecting images, the present invention includes a base 140, the base 140 is provided with a vertically distributed support frame 110, and the metal arc-shaped surface can be supported by the support frame 110; the base 140 is also provided with a suspension frame 150 The industrial camera 100 can be suspended through the suspension frame 150; a diffuse plane light 130 is provided directly above the support frame 110, and the diffuse plane light 130 illuminates the metal arc-shaped surface to be measured, so that the industrial camera 100 can collect the required image. In the embodiment of the present invention, the X-Sight SV4-30m industrial camera 100 is used to collect the arc-shaped side surface image of the cylindrical workpiece. The sampling unit of the industrial camera 100 is 1/3 inch CMOS, and the resolution is 640*480 (pixels). Due to the high reflectivity of the polished metal arc-shaped surface, the detection surface is not on the same spatial plane, and only a very narrow effective area can be obtained by direct irradiation or coaxial irradiation. A blue diffuse plane LED light source is used to illuminate the measured surface vertically; because the shorter the wavelength, the better the scattering of the surface, so a blue light source with a wavelength of 470nm is selected.

As shown in Figure 2: the present invention comprises the following steps to the real-time detection method of polished metal arc-shaped surface defect:

Step 1. Offline, obtain the first sample images of N qualified workpieces under normal working light and the second sample images of M qualified workpieces under low light, and perform median Filter to obtain image sequence I _n , n=1,2,...,N, respectively perform median filtering on M second sample images to obtain image sequence I _m ,m=1,2,...,M, for image sequence I _n and the image sequence I _m respectively carry out data fusion to obtain images g(x, y), h(x, y);

The offline situation means that the image of the qualified workpiece is first collected by the industrial camera, and the required conclusion is obtained by analyzing a large amount of data, which can be used for online detection. The basis; the second is to extract features, which are used for threshold segmentation and decision-making during detection. Acquiring N first sample images and M second sample images is to perform data fusion on the sample images, so as to obtain more reliable background components, where N may be equal to M. The range of low light is best when the average gray value is around 20, and the average gray value of normal light is about 40.

Step 1 includes the following specific steps:

Step 1.1. In the offline situation, collect the first sample images of N qualified workpieces under working light, and then reduce to the required light intensity, and collect the second sample images of M qualified workpieces under low light intensity;

Step 1.2, performing median filtering on the first sample image and the second sample image to obtain image sequences I _n , I _m respectively; where, n=1, 2,..., N, m=1, 2,...M;

During median filtering, the gray value of each pixel is set to the median value of all pixel gray values in a certain neighborhood window at that point. The default window is a 25*25 neighborhood window. According to the collected sample image The size can be adjusted accordingly. Median filtering is a conventional filtering method, which is well known to those skilled in the art and will not be described in detail here;

将图像序列I_n进行数据融合，其中，(x,y)表示图像序列内的像素位置；Step 1.3, according to

Carry out data fusion with the image sequence I _n , wherein, (x, y) represents the pixel position in the image sequence;

将图像序列I_m进行数据融合。Step 1.4, according to

The image sequence I _m is subjected to data fusion.

Step 2. Count the corresponding gray value and the number of pixels corresponding to the gray value in the fused image g(x, y), h(x, y), and obtain the image g(x, y), h(x, y) respectively ) histogram, according to the gray value corresponding to the corresponding larger wave peak in the histogram, the linear relationship between the image background I(x, y) and the gray value corresponding to the larger wave peak in the histogram is established as

I(x,y)=a(x,y)*(Z _max -Z _h )+h(x,y)

Among them, Z _max represents the gray value corresponding to the larger wave peak in the histogram, and a(x, y) is the slope matrix;

Described step 2 comprises the following steps:

Step 2.1, the grayscale of the image is 0~255, the number of pixels corresponding to the grayscale value of the image g(x,y) is counted, and then divided by the total number of pixels of the image g(x,y) to obtain the image g(x,y) y) histogram p(z), and calculate the gray value Z _g that reaches the larger peak value;

Establishing a histogram is a commonly used technical means in this technical field. The abscissa of the histogram is the gray value of the image, and the ordinate is the ratio of the number of pixels of a certain gray value to the total number of phases of the image, which will not be detailed here. described; z is the gray value of the image;

Step 2.2, according to step 2.1, obtain the histogram q(z) of the image h(x, y), and calculate the gray value Z _h that reaches a larger peak value;

Step 2.3, establish the linear slope matrix between the image background I (x, y) and the corresponding gray value of the larger wave peak in the histogram, and obtain

$\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}<span\; class="notranslate">\; ;</span>$

Step 2.4, obtain the linear relationship between the image background I (x, y) and the larger wave peak in the histogram according to step 2.3, as

I(x,y)=a(x,y)*(Z _max -Z _h )+h(x,y),

Wherein, Z _max represents the gray value corresponding to the larger wave peak in the histogram.

Step 3. Select a sample image from the above-mentioned first sample image and the second sample image, and establish the histogram of the selected first sample image and the second sample image to obtain the selected first sample image , the gray value corresponding to the larger wave peak value in the histogram of the second sample image, according to the gray value and the linear relational expression in step 2, respectively obtain the reflection components of the selected first sample image and the second sample image;

Described step 3 comprises the steps:

Step 3.1, choose a sample image from the above N first sample images and M second sample images, count the number of pixels corresponding to the gray value in the selected first sample image and the second sample image, To establish the histogram of the selected first sample image and the second sample image, obtain the gray value corresponding to the larger wave peak value in the selected first sample image and the second sample image histogram;

Step 3.2, according to step 2 and the gray value corresponding to the larger wave peak value, obtain the background component I (x, y) respectively corresponding to the selected first sample image and the second sample image;

求取所选第一样本图像、第二样本图像的反射分量，其中，k为常数，f(x,y)为所选第一样本图像或第二样本图像；k是取决于反射分量的常量，为了使反射分量的均值在0.5左右，一般k取0.5。Step 3.3, according to

Find the reflection component of the selected first sample image and the second sample image, where k is a constant, f(x, y) is the selected first sample image or the second sample image; k depends on the reflection component The constant of , in order to make the average value of the reflection component around 0.5, generally k is 0.5.

Step 4. Obtain the reflection components of the selected first sample image and the second sample image according to step 3, and establish the corresponding relationship between the standard deviation of the reflection component of the qualified workpiece and the gray value corresponding to the larger wave peak in the histogram:

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

Among them, σ(Z _max ) represents the standard deviation of the reflection component of a qualified workpiece, Z _{max_n} represents the gray value corresponding to the maximum peak value in the histogram of the selected first sample image, and Z _{max_0} represents the gray value in the histogram of the selected second sample image The gray value corresponding to the maximum peak value, σ _n is the standard deviation of the selected first sample image, and σ ₀ is the standard deviation of the selected second sample image;

Described step 4 comprises the steps:

Step 4.1, calculate the mean value μ of the reflection component of the selected first sample image, and obtain

$\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Among them, n*m represents the size of the reflection component image;

Step 4.2, calculate the standard deviation σ _n of the reflection component of the selected first sample image according to step 4.1, and obtain

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ;</span>$

Step 4.3, calculate the standard deviation σ ₀ of the reflection component of the selected second sample image according to steps 4.1 and 4.2;

Step 4.4, establish the corresponding relationship between the standard deviation of the reflection component of the qualified workpiece and the gray value Z _max of the larger wave peak in the histogram, and obtain,

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

Among them, σ(Z _max ) represents the standard deviation of the reflection component of a qualified workpiece, Z _{max_n} represents the gray value corresponding to the maximum peak value in the histogram of the selected first sample image, and Z _{max_0} represents the gray value in the histogram of the selected second sample image The gray value corresponding to the maximum peak value.

Step 5. Obtain the detection image of the polished metal arc-shaped workpiece in real time online under the working light, perform median filtering on the detection image, and establish a histogram of the detection image; according to the gray value corresponding to the larger wave peak in the histogram and the above The first step is to obtain the reflection component of the detection image; Gaussian filtering is performed on the reflection component of the detection image to obtain the second reflection component, and the second standard deviation of the second reflection component is calculated, and according to step 4, the detection image is obtained at the corresponding gray level The first standard deviation of the qualified workpiece image;

On-line detection means that after the detection image is obtained, the detection and analysis of the detection image is used to determine whether there is a flaw on the metal arc-shaped surface; the online detection needs to use the conclusion and data of the offline detection. When the technical means used in the online detection are consistent with the offline detection, you can refer to The above calculation formula and method for off-line detection.

Described step 5 comprises the following steps:

Step 5.1. Under the working light conditions, collect the workpiece detection image t(x, y), and perform median filtering on the detection image t(x, y);

Step 5.2, after establishing the median filter, detect the histogram of the image, and obtain the gray value Z _t corresponding to the larger wave peak in the histogram according to the histogram of the detected image;

Step 5.3, according to the gray value Z _t and step 2, obtain the background component I _t (x, y) of the detected image;

Step 5.4, according to step 5.3 and step 3, obtain the reflection component r _t (x, y) of the detected image;

Step 5.5. Perform Gaussian filtering on the reflection component r _t (x, y) to generate the second reflection component R(x, y), where the convolution template of the Gaussian filtering is

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 16</span>}}<span\; class="notranslate">\; *</span>\left[\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\\ \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 4</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 2</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]<span\; class="notranslate">\; .</span>$

Step 5.6, obtain the first standard deviation σ(Z _t ) of the detected image according to the gray value Z _t and step 4.4;

Step 5.7. Calculate the second standard deviation σ _t of the second reflection component R(x, y) according to Step 4.1 and Step 4.2.

Step 6. Comparing the first standard deviation and the second standard deviation to select a segmentation threshold, and performing threshold segmentation on the second reflection component through the segmentation threshold to obtain a corresponding binary image;

Described step 6 comprises the steps:

Step 6.1, compare the relationship between the first standard deviation σ(Z _t ) and the second standard deviation σ _t to get

$\mathrm{<span\; class="notranslate">\; sigma</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\end{array}\right.$

Step 6.2. Set the segmentation upper threshold T ₁ and lower threshold T ₂ to obtain

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; sigma</span>}\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; sigma</span>}\end{array}\right.$

Wherein, λ is an adjustment coefficient; the adjustment coefficient λ is 3-4.

Step 6.3, using the upper threshold T ₁ and the lower threshold T ₂ to perform threshold segmentation on the second background component R(x,y) to obtain a binary image b(x,y), and obtain

$\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right.$

Among them, 0 is the background area, and 1 is the area to be processed.

Step 7. Scanning the above binary image, marking different connected areas in the binary image, counting the area of the connected area, comparing the area of the connected area with a preset judgment threshold, and judging the flaws of the polished metal arc surface.

Described step 7 comprises the steps:

Step 7.1, scan the binary image b(x, y), and mark the different connected regions in the binary image b(x, y);

In the embodiment of the present invention, a recursive method can be used for region marking;

Step 7.2: Count the areas of different connected regions in the binary image b(x, y) respectively, and set the required judgment threshold S; when the area of connected regions in the binary image b(x, y) is greater than the judgment threshold S, then corresponding The connected area is a defective area; when the area of the connected area in the binary image b(x, y) is smaller than the judgment threshold S, the corresponding connected area is a normal area. By judging whether the connected area in the binary image b(x, y) is a flawed area, it is judged whether there is a flaw on the metal arc surface. Due to the metal polishing surface and the processing requirements of different industries, the judgment threshold S can be set accordingly.

The invention first learns and analyzes the brightness distribution of the measured workpiece in the working state in the off-line state, and the statistical characteristics of the qualified workpiece; then, during the online detection, it can effectively extract the reflection component of the measured workpiece, and through filtering, threshold Segmentation and other operations are used to perform real-time and accurate defect detection on workpieces.

Claims (10)

1. polishing metal arcuation face flaw real-time detection method based on machine vision, it is characterized in that: said polishing metal arcuation face flaw real-time detection method comprises the steps:

Under step 1, the off-line case, obtain N opening and closing lattice workpiece, N is opened first sample image carry out medium filtering acquisition image sequence I respectively at first sample image under the operate as normal illumination and second sample image of M opening and closing lattice workpiece under low light shines _n, n=1,2 ..., N opens second sample image to M and carries out medium filtering acquisition image sequence I respectively _m, m=1,2 ..., M is to image sequence I _n, image sequence I _mCarry out respectively obtaining after the data fusion image g (x, y), h (x, y);

Step 2, statistics fused image g (x; Y), h (x, y) in the corresponding number of pixels of corresponding gray-scale value and gray-scale value, obtain image g (x respectively; Y), h (x; Y) histogram, (x y) and in the histogram than the linear relation between big crest value corresponding gray does to set up image background I according to the corresponding big crest value corresponding gray of histogram

I(x,y)＝a(x,y)*(Z _max-Z _h)+h(x,y)

Wherein, Z _MaxBig crest value corresponding gray in the expression histogram, (x y) is the slope matrix to a;

Step 3, from above-mentioned first sample image, second sample image all optional sample image; And set up the histogram of selected first sample image, second sample image; Obtain the big crest value corresponding gray in selected first sample image, the second sample image histogram; According to the linear relation of said gray-scale value and step 2, obtain the reflecting component of selected first sample image, second sample image respectively;

Step 4, obtain the reflecting component of selected first sample image, second sample image, set up that the corresponding relation formula between big crest value corresponding gray does in qualified workpiece reflecting component standard deviation and the histogram according to step 3

$\mathrm{\&sigma;}\left(Z_{\max }\right)\&ap;\frac{{\mathrm{\&sigma;}}_n-{\mathrm{\&sigma;}}_0}{Z_{\max \_n}-Z_{\max \_0}}*(Z_{\max }-Z_{\max \_0})+{\mathrm{\&sigma;}}_0$

Wherein, σ (Z _Max) expression qualified workpiece the reflecting component standard deviation, Z _{Max_n}Represent maxima of waves peak value corresponding gray in the selected first sample image histogram, Z _{Max_0}Represent maxima of waves peak value corresponding gray in the selected second sample image histogram, σ _nBe the standard deviation of selected first sample image, σ ₀Standard deviation for selected second sample image;

Step 5, online in real time are obtained the detected image of polishing metal arcuation face workpiece under work illumination, and said detected image is carried out medium filtering, set up the histogram of detected image; Obtain the reflecting component of detected image according to big crest value corresponding gray and above-mentioned steps in the histogram; Reflecting component to obtaining detected image carries out gaussian filtering, obtains second reflecting component, and calculates second standard deviation of second reflecting component, and calculate first standard deviation of detected image at the qualified workpiece image of corresponding grey scale level according to step 4;

Step 6, comparison first standard deviation and second standard deviation to choose segmentation threshold, are carried out Threshold Segmentation through segmentation threshold to second reflecting component, obtain corresponding bianry image;

Step 7, scan above-mentioned bianry image, and different connected regions in the bianry image are carried out mark, statistics connected region area, with connected region area and preset judgment threshold relatively, the flaw of judgement polishing metal arcuation face.

2. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 1 is characterized in that said step 1 comprises the steps:

Step 1.1, under off-line case, gather N opening and closing lattice workpiece at work illumination first sample image down, be reduced to required intensity of illumination then, gather second sample image of M opening and closing lattice workpiece under low light photograph intensity;

Step 1.2, first sample image, second sample image are carried out medium filtering obtain image sequence I respectively _n, I _mWherein, n=1,2 ..., N, m=1,2 ... M;

Step 1.3, basis

With image sequence I _nCarry out data fusion, wherein, (x, y) location of pixels in the presentation video sequence;

Step 1.4, basis

With image sequence I _mCarry out data fusion.

3. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 1 is characterized in that said step 2 comprises the steps:

Step 2.1, gradation of image get 0 ~ 255, and ((x, total number of pixels y) obtain image g (x, histogram p (z) y), and obtain the gray-scale value Z that reaches big crest value to statistical picture g divided by image g then for x, the y) number of pixels of corresponding grey scale value _g

Step 2.2, according to step 2.1, obtain image h (x, histogram q (z) y), and obtain the gray-scale value Z that reaches big peak value _h

Step 2.3, (x, the linear gradient matrix between big crest value corresponding grey scale value y) and in the histogram obtains to set up image background I

$a(x,y)=\frac{g(x,y)-h(x,y)}{Z_g-Z_h};$

Step 2.4, according to step 2.3 obtain image background I (x, the linear relationship between big crest value y) and in the histogram, for

I(x,y)＝a(x,y)*(Z _max-Z _h)+h(x,y)，

Wherein, Z _MaxBig crest value corresponding gray in the expression histogram.

4. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 1 is characterized in that said step 3 comprises the steps:

Step 3.1, open first sample image, M from above-mentioned N and open all optional sample image second sample image; Add up the pixel count of corresponding grey scale value in selected first sample image, second sample image; To set up the histogram of selected first sample image, second sample image, obtain the big crest value corresponding gray in selected first sample image, the second sample image histogram;

Step 3.2, according to step 2 and big crest value corresponding gray, obtain the corresponding respectively background component I of selected first sample image, second sample image (x, y);

Step 3.3, basis

are asked for the reflecting component of selected first sample image, second sample image; Wherein, K is a constant; (x y) is selected first sample image or second sample image to f.

5. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 1 is characterized in that said step 4 comprises the steps:

The average μ of the reflecting component of step 4.1, selected first sample image of calculating obtains

$\mathrm{\&mu;}=\frac{1}{n*m}\underset{x=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}r(x,y),$

Wherein, n*m representes the size of reflecting component image;

Step 4.2, calculate the standard deviation sigma of the selected first sample image reflecting component according to step 4.1 _n, obtain

${\mathrm{\&sigma;}}_n=\sqrt{\frac{1}{n*m}\underset{x=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}{(r(x,y)-\mathrm{\&mu;}))}^2};$

Step 4.3, calculate the standard deviation sigma of the selected second sample image reflecting component according to step 4.1, step 4.2 ₀

Step 4.4, set up the gray-scale value Z of big crest value in qualified workpiece reflecting component standard deviation and the histogram _MaxThe corresponding relation formula obtains,

$\mathrm{\&sigma;}\left(Z_{\max }\right)\&ap;\frac{{\mathrm{\&sigma;}}_n-{\mathrm{\&sigma;}}_0}{Z_{\max \_n}-Z_{\max \_0}}*(Z_{\max }-Z_{\max \_0})+{\mathrm{\&sigma;}}_0$

Wherein, σ (Z _Max) expression qualified workpiece the reflecting component standard deviation, Z _{Max_n}Represent maxima of waves peak value corresponding gray in the selected first sample image histogram, Z _{Max_0}Represent maxima of waves peak value corresponding gray in the selected second sample image histogram.

6. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 5 is characterized in that said step 5 comprises the steps:

Step 5.1, under the work illumination condition, gather workpiece sensing image t (x, y), and to detected image t (x y) carries out medium filtering;

Step 5.2, set up medium filtering after, the histogram of detected image according to the histogram of detected image, obtains big crest value corresponding gray Z in the histogram _t

Step 5.3, according to gray-scale value Z _tReach step 2, obtain the background component I of detected image _t(x, y);

Step 5.4, according to step 5.3 and step 3, obtain the reflecting component r of detected image _t(x, y);

Step 5.5, to reflecting component r _t(x y) carries out gaussian filtering, generate the second reflecting component R (x, y);

Step 5.6, according to gray-scale value Z _tAnd step 4.4 obtains the first standard deviation sigma (Z of detected image _t);

Step 5.7, calculate the second reflecting component R (x, second standard deviation sigma y) according to step 4.1 and step 4.2 _t

7. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 6 is characterized in that said step 6 comprises the steps:

Step 6.1, the comparison first standard deviation sigma (Z _t) and second standard deviation sigma _tRelation, obtain

$\mathrm{sigma}=\left\{\begin{array}{c}2\mathrm{\&sigma;}\left(Z_t\right),{\mathrm{\&sigma;}}_t>2\mathrm{\&sigma;}\left(Z_t\right)\\ {\mathrm{\&sigma;}}_t,{\mathrm{\&sigma;}}_t\&le;2\mathrm{\&sigma;}\left(Z_t\right)\end{array}\right.$

Upper threshold value T is cut apart in step 6.2, setting ₁, lower threshold value T ₂, obtain

$\left\{\begin{array}{c}{T}_1=Z_t+\mathrm{\&lambda;}*\mathrm{sigma}\\ T_2=Z_t-\mathrm{\&lambda;}*\mathrm{sigma}\end{array}\right.$

Wherein, λ is an adjustment factor;

Step 6.3, utilize upper threshold value T ₁, lower threshold value T ₂(x y) carries out Threshold Segmentation, and (x y), obtains to obtain bianry image b to the second background component R

$b(x,y)=\left\{\begin{array}{c}0,T_2\&le;R(x,y)\&le;T_1\\ 1\end{array}\right.$

Wherein, 0 is the background area, and 1 is pending zone.

8. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 7 is characterized in that said step 7 comprises the steps:

Step 7.1, scanning bianry image b (x, y), and to bianry image b (x, the different connected regions in y) are carried out mark;

Step 7.2, respectively add up bianry image b (x, y) in the area of different connected regions, required judgment threshold S is set; Bianry image b (x, when y) interior connected region area was greater than judgment threshold S, then corresponding connected region was a defect areas; (x, when y) interior connected region area was less than judgment threshold S, then corresponding connected region was the normal region as bianry image b.

9. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 6, it is characterized in that: in the said step 5.5, the convolution template of gaussian filtering does

$h=\frac{1}{16}*\left[\begin{array}{ccc}1& 2& 1\\ 2& 4& 2\\ 1& 2& 1\end{array}\right].$

10. the polishing metal arcuation face flaw real-time detection method based on machine vision according to claim 7, it is characterized in that: said adjustment factor λ is 3 ~ 4.

Images