A machine vision-based black rubber module appearance detection method and system

Abstract

The application discloses a kind of black glue module appearance detection method and system based on machine vision, comprising the following steps: using imaging device obtains the image containing black glue and protective film gap;Image is separated to channel, and high-contrast single-channel image is obtained as measured image;The measured image is denoised and image enhanced, and enhanced image is obtained;Edge detection is carried out to enhanced image using improved edge gradient extraction method, and edge image is obtained;According to the texture features of edge image, combined with the gray scale features of enhanced image, extreme value method and least square method are used to fit two edge center lines respectively, and the distance of two center lines is calculated;Distance is compared with pre-specified qualified numerical range, and quality determination result of black glue module appearance is output according to comparison result.The application detects black glue module appearance using machine vision method based on improved edge gradient extraction method, which can effectively improve detection efficiency and detection rate.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of machine vision, and particularly relates to a black adhesive module appearance detection method and system based on machine vision. BACKGROUND

[0002] Most of the current electronic device screens use liquid crystals as light adjusting elements to display images. In the manufacturing process of liquid crystal display panels, due to the quality control problems in mass production, the produced display panels may have defects such as light leakage and uneven display. Therefore, panel manufacturers will detect the quality of the screen to eliminate defective products.

[0003] In recent years, in order to improve the screen ratio of the screen, mobile phone manufacturers have continuously proposed more innovative screen designs, such as drop screen, full screen, and round hole screen. All different backlights will use a light-shielding double-sided adhesive tape to connect and fix the glass plate and the plastic frame below. This double-sided adhesive tape has the functions of fixing and preventing light leakage, and is generally referred to as light-shielding double-sided adhesive tape, or black adhesive (because the middle substrate is a black PET film), or called mouth glue (because the frame structure of the display screen is mostly a rectangular or square mouth shape). The current screen appearance detection mainly detects the distance between the black adhesive and the screen edge, and the distance between the protective film and the screen edge. However, complex processes can also lead to new defects, such as black adhesive intrusion and side light leakage defects. Currently, there is no special defective product detection scheme in the field, and the original liquid crystal display panel detection scheme and manual inspection method are still used, which has low detection efficiency and low detection rate. SUMMARY

[0004] The present application provides a black adhesive module appearance detection method based on machine vision, which aims to solve the problem of low detection efficiency and low detection rate of the existing black adhesive module appearance detection.

[0005] To solve the above technical problems, the present application provides the following technical scheme:

[0006] A black adhesive module appearance detection method based on machine vision, comprising the following steps:

[0007] S1: using an imaging device to obtain an image containing a black adhesive and a protective film gap.

[0008] S2: performing channel separation on the image to obtain a plurality of single-channel images, detecting the contrast of all single-channel images, and selecting the single-channel image with the highest contrast as the to-be-detected image.

[0009] S3: performing Gaussian filter denoising on the to-be-detected image, and performing image enhancement on the denoised to-be-detected image in a histogram equalization manner to obtain an enhanced image.

[0010] S4: edge detection is performed on the enhanced image by using the improved edge gradient extraction method to detect edges of the black rubber and the screen and edges of the protective film and the screen in the enhanced image, and an edge image is obtained.

[0011] The improved edge gradient extraction method sets a 3*3 convolution kernel according to edge features of the enhanced image, calculates an average gray gradient of each pixel in the enhanced image using the convolution kernel, and replaces the original gray gradient of each pixel with the average gray gradient. An average gray gradient maximum value is selected, and a new high threshold value is recalculated by combining the average gray gradient of each pixel and a pre-specified original high threshold value. The new high threshold value is combined with a pre-specified original low threshold value, and the enhanced image is traversed pixel by pixel. The pixel points with a gray gradient between the new high threshold value and the original low threshold value are extracted, and a set of the extracted pixel points is the edges of the black rubber and the screen and the edges of the protective film and the screen, and the edge image is obtained.

[0012] S5: according to the edge image texture features and the enhanced image gray features, two extreme points are selected according to the gray gradient of each row or each column of pixels on the edge image by using the extreme value method, a double extreme point set is generated by traversing the image, and two edge center lines are fitted by using the least square method on the double extreme point set, and the two edge center lines are detected as a first edge center line and a second edge center line, a double edge line image is obtained, and the distance between the first edge center line and the second edge center line in the double edge line image is calculated.

[0013] S6: the distance calculated in step S5 is compared with a pre-specified qualified value range, and a quality judgment result of the black rubber module appearance is output according to the comparison result.

[0014] Preferably, the formula for calculating the contrast in step S2 is:

[0015]

[0016] In the formula, Con is the contrast, L max is the maximum brightness value in each single-channel image, L min is the minimum brightness value in each single-channel image.

[0017] Preferably, a plane rectangular coordinate system is established with the center point of the convolution kernel as the origin, and is used to calculate four direction gradients: an x direction gradient, a 45° direction gradient, a y direction gradient and a 135° direction gradient.

[0018] The calculation formulas of the x direction gradient, the 45° direction gradient, the y direction gradient and the 135° direction gradient are respectively:

[0019] G x(i, j) = I(i + 1, j - 1) - I(i - 1, j - 1) + 2(I(i + 1, j) - I(i - 1, j)) + I(i + 1, j + 1) - I(i - 1, j + 1)

[0020] G 45 (i, j) = 2(I(i - 1, j + 1) - I(i + 1, j - 1)) + I(i - 1, j) - I(i, j - 1) + I(i, j + 1) - I(i + 1, j)

[0021] G y (i, j) = I(i - 1, j + 1) - I(i - 1, j - 1) + 2(I(i, j + 1) - I(i, j - 1)) + I(i + 1, j + 1) - I(i + 1, j - 1)

[0022] G 135 (i, j) = 2(I(i - 1, j + 1) - I(i + 1, j - 1)) + I(i - 1, j) - I(i, j - 1) + I(i, j + 1) - I(i + 1, j)

[0023] wherein G x is the x-direction gradient, G 45 is the 45°-direction gradient, G y is the y-direction gradient, G 135 is the 135°-direction gradient, I is the gray value of the corresponding pixel, i is the row number of a single pixel in the 3x3 pixel kernel, and j is the column number of a single pixel in the 3x3 pixel kernel.

[0024] Preferably, the calculation formula of the average gray gradient is:

[0025]

[0026] wherein G E (i,j) is the average gray gradient of the i-th row and j-th column pixel, G x (i,j) is the x-direction gradient of the i-th row and j-th column pixel, G 45 (i,j) is the 45°-direction gradient of the i-th row and j-th column pixel, G y (i,j) is the y-direction gradient of the i-th row and j-th column pixel, G 135 (i,j) is the 135°-direction gradient of the i-th row and j-th column pixel.

[0027] Preferably, the new high threshold calculation method is:

[0028]

[0029] wherein H is the new high threshold, H o is the set original high threshold, G maxGmax is the maximum gray gradient value of each enhanced image E G(i,j) is the average gray gradient of the i-th row and j-th column pixel.

[0030] Preferably, the least square method fits the set of bipolar points into two straight lines, and a straight line y = bx + a is set in a plane rectangular coordinate system, where b is the slope and a is the intercept. The least square method calculates the slope and intercept according to the principle that the perpendicular error of each point in the set of bipolar points to the straight line is minimized.

[0031] The method for calculating the optimal estimate of the slope and intercept is as follows:

[0032]

[0033]

[0034] In the formula, a is the optimal estimate of the intercept, b is the optimal estimate of the slope, N is the number of coordinate points to be fitted into a straight line, x i x is the x-coordinate value of point i, y i y is the y-coordinate value of point i.

[0035] Correspondingly, the present application also provides a black rubber module appearance detection system based on machine vision, comprising an imaging module, an image processing module, an image recognition module and a judgment module, and the detection system is configured to execute the above-mentioned black rubber module appearance detection method based on machine vision.

[0036] The imaging module is used to acquire an image containing the gap between the black rubber and the protective film.

[0037] The image processing module is used to perform channel separation, denoising and image enhancement on the image acquired by the imaging module.

[0038] The image recognition module uses an improved edge gradient extraction method to extract the edge gradient of the enhanced image generated by the image processing module, then obtains a set of bipolar points through an extreme value method, fits the set of bipolar points through a least square method, generates two edge center lines, and calculates the distance between the two edge center lines.

[0039] The judgment module compares the distance information output by the image recognition module with a pre-specified qualified numerical range, and outputs the quality judgment result of the black rubber module appearance according to the comparison result.

[0040] Preferably, the light source of the imaging module adopts a coaxial light source.

[0041] Compared with the prior art, the present application has the following technical effects:

[0042] 1. The machine vision-based black rubber module appearance detection method of the present application improves the traditional edge gradient extraction method, uses a 3*3 pixel convolution kernel for edge extraction, reduces the redundant information obtained in the edge gradient extraction process while ensuring that rich edge information can be extracted, effectively reduces the positioning error of the traditional edge gradient extraction method, and improves the detection rate of edge detection.

[0043] 2. The machine vision-based black rubber module appearance detection method of the present application uses machine vision to detect the appearance of the black rubber module, uses an improved edge gradient extraction method, re-calculates the high threshold, overcomes the problem of missing edge information caused by using the same high threshold for edge extraction on all pixels of an enhanced image in the conventional extraction method, retains more edge information, more comprehensively reflects the detail information of the enhanced image, and improves the accuracy of edge extraction.

[0044] 3. The machine vision-based black rubber module appearance detection system of the present application uses a coaxial light source for the imaging system, provides more uniform illumination than traditional light sources, can highlight the unevenness of the object surface, overcome the interference caused by surface reflection, and enhance the contrast of the black rubber and the screen edge, and the contrast of the protective film and the screen edge. The imaging system is more suitable for detecting planar objects with serious reflection, such as liquid crystal screen panels, and improves the accuracy and reproducibility of machine vision. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a flowchart of the machine vision-based black rubber module appearance detection method of the present application;

[0046] Figure 2 is a schematic diagram of the image to be detected of the machine vision-based black rubber module appearance detection method of the present application;

[0047] Figure 3 is a schematic diagram of the enhanced image of the machine vision-based black rubber module appearance detection method of the present application;

[0048] Figure 4 is a schematic diagram of the edge image of the machine vision-based black rubber module appearance detection method of the present application;

[0049] Figure 5 is a schematic diagram of the double-edge line image of the machine vision-based black rubber module appearance detection method of the present application;

[0050] Figure 6 is a schematic diagram of the convolution kernel of the machine vision-based black rubber module appearance detection method of the present application. DETAILED DESCRIPTION

[0051] In order to make the objects, technical solutions and advantages of the present application clearer, the following will combine the specific embodiments of the present application, and refer to the drawings for clearly and completely describing the technical solutions of the present application.

[0052] Please refer to Figure 1 A flow chart of a black rubber module appearance detection method based on machine vision is provided for the present application, including the following steps:

[0053] S1: using an imaging device to obtain an image containing the gap between black rubber and protective film. An optical imaging device is provided on the detection device, and the optical imaging device is used to obtain a photo containing the gap between black rubber and protective film on the black rubber module on the detection device.

[0054] S2: channel separation is performed on the image to obtain a plurality of single-channel images, the contrast of all single-channel images is detected, the single-channel image with the highest contrast is selected as the to-be-detected image, and the to-be-detected image is as shown in Figure 2 .

[0055] The image obtained by the imaging device is a multi-channel color image, which is R, G and B channels respectively. The RGB image can be channel separated to obtain three single-channel images. In this embodiment, the channel separation function split() of OpenCV is selected to perform channel separation on the image: first, the function receives a multi-channel source image as input; then, the function creates a corresponding number of output images according to the number of channels of the source image, for example, if the source image has 3 channels (RGB color model), 3 single-channel output images will be created; then, the function iterates through each pixel in the source image, and stores the channel value of each pixel in the corresponding output image; finally, the function returns a vector containing all output images.

[0056] The contrast represents the stretching contrast of light and dark of an image, which refers to the measurement of different brightness levels between the brightest white and the darkest black in the light and dark areas of an image. The greater the difference range represents the greater the contrast, and the smaller the difference range represents the smaller the contrast. In order to improve the efficiency and accuracy of subsequent detection, the single-channel image with the highest contrast is selected as the to-be-detected image for subsequent processing. The contrast calculation method of each single-channel image is as follows:

[0057]

[0058] In the formula, Con is the contrast, L max is the maximum brightness value in each single-channel image, L min is the minimum brightness value in each single-channel image.

[0059] S3: Gaussian filter denoising is performed on the to-be-tested image, and the denoised to-be-tested image is subjected to image enhancement in a histogram equalization manner to obtain an enhanced image, as shown in FIG. 3. Figure 3 The selected to-be-tested image also has a large amount of noise. In this embodiment, Gaussian filter is used to denoise the to-be-tested image. A convolution kernel is used to scan each pixel in the to-be-tested image, and a weighted average gray value of pixels in a neighborhood determined by the convolution kernel is used to replace the value of the center pixel point of the template. The value of each pixel point in the to-be-tested image is obtained by weighted average of the value of the pixel point itself and other pixel values in the neighborhood. The denoised to-be-tested image may not be clear enough due to the fact that the gray distribution of the to-be-tested image may be concentrated in a narrow interval. Therefore, the denoised to-be-tested image is subjected to image enhancement in a histogram equalization manner again, that is, the histogram of the denoised to-be-tested image is transformed into a uniformly distributed form, the dynamic range of the difference between the gray values of pixels is increased, and the overall contrast of the image is enhanced, so that an enhanced image with greater contrast is obtained.

[0060] S4: An improved edge gradient extraction method is used to perform edge detection on the enhanced image to detect the edges of the black rubber and the screen and the edges of the protective film and the screen, and an edge image is obtained.

[0061] The improved edge gradient extraction method sets a convolution kernel with a size of 3x3 according to the edge features of the enhanced image, calculates the average gray gradient of each pixel in the enhanced image using the convolution kernel, and replaces the original gray gradient of each pixel with the average gray gradient. An average gray gradient maximum value is selected, and a new high threshold value is recalculated by combining the average gray gradient of each pixel and a pre-specified original high threshold value. The new high threshold value is combined with a pre-specified original low threshold value, the enhanced image is traversed pixel by pixel, and the pixel points with a gray gradient between the new high threshold value and the original low threshold value are extracted. The set of extracted pixel points is the edges of the black rubber and the screen and the edges of the protective film and the screen, and an edge image is obtained, as shown in FIG. 4. Figure 4

[0062] For each pixel, the gray gradients of two adjacent pixels along the gradient direction are checked and compared with the gray gradient of the pixel. If the gray gradient of the pixel is the largest among the gray gradients of the four adjacent pixels along the gradient direction, the pixel is retained as a possible edge point, otherwise it is discarded.

[0063] ​For the remaining possible edge points, a dual thresholding process is applied. Pixels are classified into three categories based on two pre-specified thresholds (a high threshold and a low threshold): strong edges, weak edges, and non-edges. If a pixel's gradient strength exceeds the high threshold, it is marked as a strong edge; if its gradient strength is below the low threshold, it is marked as a non-edge; and if its gradient strength is between the high and low thresholds, it is marked as a weak edge.

[0064] Finally, for all pixels marked as weak edges, if any one of its eight surrounding pixels is marked as a strong edge, then that pixel is marked as a strong edge.

[0065] Finally, all pixels marked as strong edges are returned as output, generating an edge image.

[0066] like Figure 6 The diagram illustrates the use of a 3×3 convolutional kernel to calculate the average gradient in the appearance detection method proposed in this embodiment. The gradients in four directions are calculated: the x-axis gradient, the 45° gradient, the y-axis gradient, and the 135° gradient. Then, the average grayscale gradient of the corresponding pixel is calculated using these four directional gradients. The specific calculation method is as follows:

[0067] G x (i,j)=I(i+1,j-1)-I(i-1,j-1)+2(I(i+1,j)-I(i-1,j))+I(i+1,j+1)-I(i-1,j+1)

[0068] G 45 (i,j)=2(I(i-1,j+1)-I(i+1,j-1))+I(i-1,j)-I(i,j-1)+I(i,j+1)-I(i+1,j)

[0069] G y (i,j)=I(i-1,j+1)-I(i-1,j-1)+2(I(i,j+1)-I(i,j-1))+I(i+1,j+1)-I(i+1,j-1)

[0070] G 135 (i,j)=2(I(i-1,j+1)-I(i+1,j-1))+I(i-1,j)-I(i,j-1)+I(i,j+1)-I(i+1,j)

[0071] Based on the four directional gradients mentioned above, the grayscale gradient of the current pixel can be calculated:

[0072]

[0073] In the formula, Gx (i,j) the x-direction gradient of the pixel in the ith row and jth column, G 45 (i,j) the 45°-direction gradient of the pixel in the ith row and jth column, G y (i,j) the y-direction gradient of the pixel in the ith row and jth column, G 135 (i,j) the 135°-direction gradient of the pixel in the ith row and jth column, I is the gray value of the corresponding pixel, i is the row number of a single pixel in the 3x3 pixel kernel, j is the column number of a single pixel in the 3x3 pixel kernel, G E (i,j) is the average gray gradient of the pixel in the ith row and jth column.

[0074] After obtaining the average gray gradient of each pixel in the enhanced image, the original gray gradient of each pixel is replaced by the average gray gradient, and then the maximum value of the average gray gradient is searched in the enhanced image.

[0075] The new high threshold calculation method is:

[0076]

[0077] In the formula, H is the new high threshold, H o is the set original high threshold, G max is the maximum gray gradient value of each enhanced image, G 月 (i,j) is the average gray gradient of the pixel in the ith row and jth column.

[0078] By recalculating the high threshold, the problem of losing edge information caused by using the same high threshold for edge extraction for all pixels in an enhanced image in the conventional extraction method is overcome, more edge information is retained, the details of the enhanced image are more comprehensively reflected, and the accuracy of edge extraction is improved.

[0079] By improving the traditional edge gradient extraction method, a 3x3 pixel convolution kernel is used for edge extraction, which can extract rich edge information while reducing redundant information obtained in the process of edge gradient extraction, effectively reduces the positioning error of the traditional edge gradient extraction method, and improves the detection rate of edge detection.

[0080] S5: According to the edge image texture feature combined with the enhanced image gray feature, two extreme points are selected for each row or each column of pixels on the edge image according to the gray gradient using the extreme value method, a double extreme point set is obtained by traversing the image, and then two edge center lines are fitted using the least square method, and the two edge center lines are detected as the first edge center line and the second edge center line, and a double edge line graph is obtained. According to the pixel size and pixel distribution of the double edge line graph, the distance between the first edge center line and the second edge center line in the double edge line graph is calculated. The double edge line graph is as followsFigure 5 The minimum value point set is fitted as two straight lines, and a straight line y = bx + a is set in a plane rectangular coordinate system, where b is the slope and a is the intercept. The least square method calculates the slope and intercept according to the principle that the vertical error of each point in the minimum value point set to the straight line is minimized.

[0081] The least square method calculates the slope and intercept according to the principle that the vertical error of each point in the minimum value point set to the straight line is minimized.

[0082] The best estimate value of the slope and intercept is calculated as follows:

[0083]

[0084]

[0085] In the formula, a is the best estimate value of the intercept, b is the best estimate value of the slope, N is the number of coordinate points to be fitted as a straight line, x i xi is the x coordinate value of point i, y i yi is the y coordinate value of point i.

[0086] S6: According to the distance calculated in step S5, the distance is compared with the pre-specified qualified value range, and the quality judgment result of the appearance of the black rubber module is output according to the comparison result. If the distance of the two center lines calculated is within the pre-specified qualified value range, it is determined that the appearance detection of the black rubber module passes; if the distance of the two center lines calculated is outside the pre-specified qualified value range, it is determined that the appearance detection of the black rubber module fails.

[0087] Correspondingly, the embodiment also proposes a black rubber module appearance detection system based on machine vision, which comprises an imaging module, an image processing module, an image recognition module and a judgment module. The detection system is configured to execute the above-mentioned black rubber module appearance detection method based on machine vision.

[0088] The imaging module is used to acquire an image containing the gap between the black rubber and the protective film.

[0089] The image processing module is used to separate the image acquired by the imaging module into channels, remove noise and enhance the image.

[0090] The image recognition module uses an improved edge gradient extraction method to extract the edge gradient of the enhanced image generated by the image processing module, then obtains a minimum value point set through an extreme value method, fits the minimum value point set through a least square method, generates two edge center lines, and calculates the distance of the two edge center lines.

[0091] The judgment module compares the distance information output by the image recognition module with the pre-specified qualified value range, and outputs the quality judgment result of the appearance of the black rubber module according to the comparison result.​

[0092] The light source of the imaging module of the detection system of the embodiment can adopt a ring light source or a coaxial light source. For a planar object with a liquid crystal screen panel and a relatively serious reflection degree, the coaxial light source provides more uniform illumination than the ring light source, can highlight the unevenness of the surface of the object, overcome the interference caused by the surface reflection, enhance the contrast of the black adhesive and the screen edge and the protective film and the screen edge, and can improve the accuracy and reproducibility of machine vision.

[0093] The above only describes the preferred embodiments of the present application, and it should be noted that for those skilled in the art, without departing from the inventive concept, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A machine vision-based black rubber module appearance detection method, characterized in that, The method comprises the following steps: S1: acquiring an image containing a gap between the black rubber and the protective film using an imaging device; S2: performing channel separation on the image to obtain a plurality of single-channel images, detecting the contrast of all single-channel images, and selecting a single-channel image with the highest contrast as a to-be-tested image; S3: performing Gaussian filter denoising on the to-be-tested image, performing image enhancement on the denoised to-be-tested image in a histogram equalization manner to obtain an enhanced image; S4: performing edge detection on the enhanced image by using an improved edge gradient extraction method to detect the edges of the black rubber and the screen and the edges of the protective film and the screen in the enhanced image, and obtaining an edge image; The improved edge gradient extraction method sets a convolution kernel with a size of 3*3 according to the edge features of the enhanced image, calculates the average gray gradient of each pixel in the enhanced image using the convolution kernel, and replaces the original gray gradient of each pixel with the average gray gradient; An average gray gradient maximum value is selected, the average gray gradient of each pixel is combined with a pre-specified original high threshold value, and a new high threshold value is recalculated; The new high threshold value is combined with a pre-specified original low threshold value, the enhanced image is traversed pixel by pixel, and the pixel points with a gray gradient between the new high threshold value and the original low threshold value are extracted, so that the set of extracted pixel points are the edges of the black rubber and the screen and the edges of the protective film and the screen, and the edge image is obtained; S5: according to the texture features of the edge image and the gray features of the enhanced image, two extreme points are selected according to the gray gradient of each row or each column of pixels on the edge image by using an extreme value method, a double extreme point set is generated by traversing the image, and then two edge center lines are fitted by using a least square method, the two edge center lines are detected as a first edge center line and a second edge center line, a double edge line image is obtained, and the distance between the first edge center line and the second edge center line in the double edge line image is calculated; S6: comparing the distance calculated in step S5 with a pre-specified qualified value range, and outputting a quality judgment result of the black rubber module appearance according to the comparison result.

2. The machine vision-based black module appearance detection method according to claim 1, wherein, The formula for calculating the contrast in step S2 is: wherein is the contrast, is the maximum luminance value in each single channel image, is the minimum luminance value in each single channel image.

3. The machine vision-based black module appearance detection method according to claim 1, wherein, The center point of the convolution kernel is used as the origin to establish a plane rectangular coordinate system for calculating four direction gradients: x direction gradient, 45° direction gradient, y direction gradient and 135° direction gradient.

4. The machine vision-based black module appearance detection method according to claim 3, characterized in that, The calculation formulas of the x direction gradient, 45° direction gradient, y direction gradient and 135° direction gradient are respectively: wherein, is an x-direction gradient, is a 45° direction gradient, is a y-direction gradient, is a 135° direction gradient, is a gray value of a corresponding pixel, is a row number of a single pixel in a 3x3 pixel kernel, is a column number of a single pixel in a 3x3 pixel kernel; The calculation formula of the average gray gradient is: wherein is the row column pixel average gray level gradient, is the row column pixel x-direction gradient, is the row column pixel 45°-direction gradient, is the row column pixel y-direction gradient, is the row column pixel 135°-direction gradient.

5. The machine vision-based black module appearance detection method according to claim 1, wherein, The new high threshold value calculation method is: wherein is a new high threshold, is a set original high threshold, is the maximum gray level gradient value of each enhanced image, is the average gray level gradient of the row column pixels.

6. The machine vision-based black module appearance detection method according to claim 1, wherein, The least square method fits the set of bipolar points into two straight lines, and a straight line is set in a plane rectangular coordinate system wherein is a slope, is an intercept; the least square method calculates the slope and the intercept according to the principle that the vertical error of each point in the set of bipolar points to the straight line is minimum; and the best estimation value of the slope and the intercept is calculated by: wherein is the best estimate of the intercept, is the best estimate of the slope, is the number of coordinate points to be fitted to a straight line, is the coordinate value of point is the coordinate value of point is the coordinate value of point is the 7. A machine vision-based black rubber module appearance detection system, characterized in that, It comprises: An imaging module, an image processing module, an image recognition module and a judgment module, and the detection system is configured to perform the machine vision-based black rubber module appearance detection method in any one of claims 1-6; The imaging module is used to acquire an image containing a gap between the black rubber and the protective film; The image processing module is used to perform channel separation, denoising and image enhancement on the image acquired by the imaging module; The image recognition module uses an improved edge gradient extraction method to extract the edge gradient of the enhanced image generated by the image processing module, then obtains a set of double extreme points through an extreme value method, and generates two edge center lines by fitting the set of double extreme points through a least square method, and calculates the distance of the two edge center lines; The determination module compares the distance information output by the image recognition module with a pre-specified qualified value range, and outputs a quality determination result of the appearance of the black vinyl module according to the comparison result.

8. The machine vision-based black module appearance detection system of claim 7, wherein, The light source of the imaging module adopts a coaxial light source.

Images