Special-shaped part press line intelligent monitoring method and system based on machine vision

Abstract

The invention relates to the field of image processing, in particular to a special-shaped part press line intelligent monitoring method and system based on machine vision, and the method comprises the steps: obtaining the information of edge pixel points in a to-be-detected image, and calculating the part feature intensity of the edge pixel points; constructing a first point set and a second point set according to the part feature intensity of the edge pixel points, and constructing a first descriptor and a second descriptor of the edge pixel points in the first point set; similarly, a context descriptor of the edge pixel points of the standard image of the defect-free part is constructed, and the feature matching degree of the image to be detected and the pixel points in the first point set in the standard image is calculated; and determining part defects according to the feature matching degree. The method has the effect of improving the part defect monitoring precision.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method and system for intelligent monitoring of a special-shaped parts stamping line based on machine vision. Background Art

[0002] With the production and advancement of processing technology, the complexity of workpieces has gradually increased, especially some automotive parts, which have irregular shapes. For some sheet metal parts, they are mostly formed by bending and stamping processes. Each process needs to match different process equipment, such as bending machines, stamping machines, etc.; each equipment is set up independently, so the production process often involves the transportation of parts. If the parts are transported in a chaotic manner during the transportation process, the phenomenon of missing processes will occur. In order to reduce the occurrence of the above situation, the general staff will observe whether the pre-process of the workpiece is missed before loading to reduce the occurrence of missed processes. However, with the increase of processes, the burden on the staff is gradually increasing, which is not conducive to improving production efficiency. In order to improve production efficiency, the related technology can monitor the features on the parts by extracting the edge of the image of the workpiece.

[0003] The Shape Context algorithm is an effective method for describing two-dimensional shape features. It mainly describes the global features of the shape by statistically analyzing the relative position distribution of pixels on the shape. It can be applied to the monitoring of workpiece feature shapes. However, the industrial environment is complex and has many interferences. Therefore, during the operation of the Shape Context algorithm, a large number of interference points unrelated to the parts will be collected, which will affect the recognition of part features and thus affect the monitoring accuracy of special-shaped part features. Summary of the invention

[0004] In order to solve the problem of poor monitoring accuracy of special-shaped workpieces, the present application provides a method and system for intelligent monitoring of a special-shaped parts stamping line based on machine vision.

[0005] In the first aspect, the present application provides an intelligent monitoring method for a special-shaped parts stamping line based on machine vision, which adopts the following technical solutions: The intelligent monitoring method of a special-shaped parts stamping line based on machine vision comprises the steps of: obtaining edge pixel point information in an image to be detected, and calculating the part feature intensity of the edge pixel point; The first point set and the second point set are constructed according to the part feature intensity of the edge pixel points; the first descriptor and the second descriptor of the edge pixel points in the first point set are constructed, and the feature matching degree of the pixel points in the first point set of the image to be detected and the standard image is calculated; the part defect is determined according to the feature matching degree; the calculation formula of the feature matching degree is: ; In the formula, is the first point in the image to be detected. The pixel point is the same as the first point in the standard image. The feature matching degree of each pixel, is the first point in the image to be detected. The first descriptor of pixels; is the first point in the image to be detected. The second descriptor of pixels; is the first point in the standard image. The first descriptor of pixels; is the first point in the standard image. The second descriptor of pixels; is the information entropy of the grayscale values ​​of the pixels in the first point set of the standard image; is the information entropy of the grayscale values ​​of the pixels in the second point set of the standard image; is the cosine similarity function.

[0006] In the present application, during the production process, if the process of the special-shaped workpiece is missing, the shape of the special-shaped part will change, which will affect the change of the information of the edge pixel points, such as the change of the grayscale or curvature of the edge pixel points. Using this feature, the part feature intensity of the image to be detected and the standard image is calculated. Pixels with strong part feature intensity can reflect the key feature area of ​​the part, and filter out pixels with high part feature intensity, thereby reducing the interference of background and noise, and providing a more accurate data basis for subsequent analysis. The edge pixels in the image to be detected are divided into two groups of point sets with different part feature intensities, namely the first point set and the second point set; the first descriptor and the second descriptor are constructed according to the first point set and the second point set, and the distribution of edge pixel points with higher part feature intensity and lower part feature intensity in the part is obtained to distinguish different features of the special-shaped parts. The feature matching degree is then calculated, and whether the part is defective is determined based on the feature matching degree. In the formula, when the information entropy of the first point set increases, the weight of the first descriptor increases when calculating the feature matching degree. Conversely, when the information entropy of the second point set is large, the weight of the second descriptor increases. The weight is automatically adjusted according to the point set characteristics to optimize the matching process, improve the matching accuracy, and thus improve the accuracy of special-shaped workpiece monitoring.

[0007] Optionally, the steps of constructing the first descriptor and the second descriptor in the first point set include: Divide the sampling area; Calculate the number of pixels belonging to the first point set in each sampling area, and perform normalization processing to obtain the first descriptor of the pixel; The number of pixel points belonging to the second point set in each sampling area is calculated, and normalized to obtain the second descriptor of the pixel point.

[0008] According to the pixels in different point sets, a descriptor that can reflect the key features and secondary features in the key image to be detected is constructed to obtain the distribution state of pixels with different part feature intensities in the part feature area.

[0009] Optionally, the method for dividing the sampling area includes: constructing a plurality of concentric circles with different radii with the edge pixel point as the center; The concentric circles are evenly divided into multiple parts along the circumferential scale to form multiple sampling areas.

[0010] Dividing the sampling area by concentric circles is simple and effective on the one hand; on the other hand, it can adapt to the complex and diverse shapes of special-shaped parts, such as protrusions, depressions and other different shapes on parts. By reasonably setting the radius and number of concentric circles, the sampling area can better fit the boundaries of these areas and effectively cover the characteristic areas of the parts.

[0011] Optionally, the method for constructing the first point set and the second point set includes: sorting the part feature intensities of multiple edge pixels from large to small, taking some edge pixels with the largest part feature intensities according to percentage to form the first point set, and the remaining edge pixels constitute the second point set.

[0012] The edge pixels are divided into two parts, the edge pixels with stronger part feature intensity are concentrated in the first point set, and the edge pixels with weaker part feature intensity are concentrated in the second point set.

[0013] Optionally, the calculation formula for the edge pixel part feature intensity is: ; In the formula is the part feature intensity of any edge pixel in the image, is the gray value of the edge pixel, is the curvature of the edge pixel, To calculate the abnormal score of the edge pixel using the isolation forest algorithm, is a linear normalization function; is the pixel gradient.

[0014] The parts are formed by stamping or bending metal sheets, which require a larger grayscale value in the image. The larger it is, the more likely the pixel is on the part, so the stronger the part feature strength is. Pixels that do not belong to parts have smaller grayscale values, so the part feature strength is also smaller, making it easier to remove background and interference. At the same time, key feature areas of parts are mostly corners, that is, the edge pixels of the part have a larger curvature. The larger the curvature of the edge pixels, the stronger the part feature strength is. The anomaly score reflects the discreteness of the pixel. When the anomaly score of a pixel is large, it means that it is more isolated in spatial distribution, that is, the probability that the pixel does not belong to a part is higher. Therefore The larger the value, the stronger the part feature intensity will be, making it easier to distinguish the key edge pixels of the part feature area.

[0015] Optionally, the step of determining part defects according to feature matching includes: Obtain a matching point corresponding to an edge pixel point in a first point set in a standard image; obtain a feature matching degree corresponding to the matching point; obtain a feature matching degree difference according to the feature matching degree of the matching point; set a matching threshold; and compare the feature matching degree difference with the matching threshold to determine a part defect.

[0016] Optionally, the step of obtaining matching points corresponding to edge pixel points in the first point set in the standard image includes: comparing the feature matching degree between each edge pixel point in the first point set in the standard image and each pixel point in the first point set in the image to be detected; Each edge pixel point in the first point set of the standard image corresponds to multiple feature matching degrees, and the edge pixel point in the image to be detected with the maximum feature matching degree in the standard image is taken as the matching point.

[0017] Optionally, the step of comparing the feature matching degree difference with the matching threshold to determine the part defect includes: in response to the absolute value of the feature matching degree difference among multiple matching points being greater than the matching threshold, determining that the part has a defect.

[0018] Optionally, the step of obtaining the feature matching degree difference includes: calculating the mean of the feature matching degrees corresponding to all matching points in the image to be detected; and calculating the difference between each matching point and the mean to obtain the feature matching degree difference.

[0019] In the second aspect, the present application provides an intelligent monitoring system for a special-shaped parts stamping line based on machine vision, which adopts the following technical solutions: A method and system for intelligent monitoring of a stamping line for special-shaped parts based on machine vision include a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the method and system for intelligent monitoring of a stamping line for special-shaped parts based on machine vision are implemented.

[0020] The above-mentioned machine vision-based intelligent monitoring method for special-shaped parts stamping line is generated into a computer program and stored in a memory so as to be loaded and executed by a processor. Thus, a system is made based on the memory and the processor for easy use.

[0021] This application has the following technical effects: The material of the part and the characteristics of the key area affect the pixels in the image to be detected. Based on this feature, the edge pixels of the part in the image to be detected are extracted, and the part feature intensity of the edge pixels is calculated. The edge pixels are divided into two different point sets according to the part feature intensity. The two point sets are: a first point set containing edge pixels with stronger part feature intensity and a second point set containing edge pixels with weaker part feature intensity. Construct the first descriptor and the second descriptor in the first point set to obtain the spatial distribution characteristics of the edge pixels with higher and lower part feature intensity in the key feature area of ​​the part, and determine the defects of the part by comparing with the first descriptor and the second descriptor in the standard image. In this application, the extraction of part feature intensity can reduce the influence of image background and noise, and combine the first descriptor and the second descriptor to achieve accurate monitoring of part features. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a method flow chart of the intelligent monitoring method of the special-shaped parts stamping line based on machine vision in this application.

[0023] Figure 2 This is a method flow chart of step S1 of the method for intelligent monitoring of special-shaped parts stamping lines based on machine vision in this application.

[0024] Figure 3 This is a method flow chart of step S3 of the method for intelligent monitoring of special-shaped parts stamping lines based on machine vision in this application.

[0025] Figure 4 This is an example diagram of the sampling area division of the intelligent monitoring method for special-shaped parts stamping line based on machine vision in this application. DETAILED DESCRIPTION

[0026] The embodiment of the present application discloses a method and system for intelligent monitoring of a special-shaped part stamping line based on machine vision. The information of edge pixels in the image to be detected is used to obtain the part feature strength of the edge pixels related to the workpiece feature; the edge pixels related to the part structure are screened out according to the strength of the physical sign, and the background and noise interference in the background are eliminated. At the same time, multiple context descriptors of edge pixels are constructed to describe the features in layers, thereby improving the robustness and accuracy of system monitoring and the accuracy of part process monitoring.

[0027] Reference Figure 1 , comprising steps S1 to S4: S1: Obtain edge pixel information in the image to be detected and calculate the part feature intensity of the edge pixel.

[0028] The shape context algorithm is usually collected through the contour sampling points of the object in the image. The edge monitoring algorithm will collect the edges in the image background, which may cause the edges in the image background to be identified as sampling points, causing certain interference to the production monitoring of special-shaped parts. In order to reduce the impact of this phenomenon on the production monitoring of special-shaped parts, this application calculates the part feature intensity of the edge pixel points through the grayscale value, curvature and anomaly score of the edge pixel points in the image.

[0029] Reference Figure 2 , step S1 includes steps S10 to S13; S10: Arrange an image acquisition device, and obtain an image to be detected of the special-shaped workpiece to be monitored through the image acquisition device.

[0030] S11: Obtain edge pixel points of the object in the image to be detected; The image to be detected is grayed out to obtain the gray value of each pixel in the image to be detected. All edge features in the image to be detected are obtained through the edge detection algorithm, and the pixels on the edge are regarded as edge pixels.

[0031] S12: Using the coordinates of edge pixels as features, the isolation forest algorithm is used to calculate the anomaly score of each edge pixel.

[0032] S13: Calculate the part feature strength according to the feature information of the edge pixel points; the calculation formula of the part feature strength is: ; in, is the part feature intensity of any edge pixel in the image, is the gray value of the edge pixel, is the curvature of the edge pixel, is the abnormal score of the pixel, is a linear normalization function.

[0033] Special-shaped parts are formed by metal stamping or bending. Metal products have higher grayscale values ​​in the image. Therefore, the larger the grayscale value of the edge pixel, the more likely it is that the edge pixel is on the part. In order to highlight the pixel with prominent features on the part, the part feature intensity of the pixel should be greater. The smaller the grayscale value, the less likely the edge pixel is on the part. In order to reduce the impact of the background part in the image on the part production monitoring, the part feature intensity of the pixel should be smaller.

[0034] Special-shaped parts often have more unique structural features than regular parts. If this feature is missing in the image, it can be judged that the stamping step of the part is missing. Therefore, the unique structural features of special-shaped parts can be used to calculate the feature strength of the parts to better monitor the production of parts. The unique features of special-shaped parts are shown in the image as the curvature of the edge, the curvature of the edge pixel point The larger the value is, the more likely the edge pixel is at the complex edge of the image, the more likely the pixel is at the unique structure of the part, and the greater the part feature intensity of the pixel. The smaller it is, the less likely the edge pixel is at a complex edge in the image, and the smaller the part feature intensity of the pixel is.

[0035] Anomaly score of edge pixels It reflects the discreteness of the pixels. The larger the value, the more isolated the pixel is in spatial distribution. The more isolated the pixel is, the more likely it is that the edge pixel is formed by image background or noise. In order to reduce the impact of the background part in the image on part production monitoring, the part feature intensity of the pixel should be smaller. The smaller it is, the more concentrated the pixel point is in spatial distribution. Since the pixels on the part are concentrated, the edge pixel point is more likely to be on the part. In order to highlight the pixels with unique features on the part, the part feature intensity of the edge pixel point should be greater.

[0036] therefore The larger the value is, the more likely the pixel is on the edge of the structural protrusion on the special-shaped part, and the greater the part feature intensity of the pixel is. The smaller the value, the less likely the pixel point is on the edge of the structural protrusion on the special-shaped part, and the smaller the part feature intensity of the pixel point. In order to facilitate subsequent calculations, a linear normalization function is used here to Perform normalization.

[0037] In another embodiment, the calculation formula of the characteristic strength of a part can also be expressed as: ; is the part feature intensity of any edge pixel in the image, is the gray value of the edge pixel, is the curvature of the edge pixel, is the abnormal score of the pixel, is a linear normalization function; is the gradient value of the pixel.

[0038] The gradient value of a pixel in this formula reflects the rate of change of the pixel in the image. The larger the gradient value, the more the pixel corresponds to an area in the image where the brightness changes dramatically, that is, the edge of the object in the image. Noise may also cause edges in the image background, but since the workpiece is usually photographed on a single background and the reflection of light is relatively consistent, the grayscale value of the background area does not change much, so the gradient value of this edge is usually small, so the gradient value The larger the value, the greater the possibility that the pixel point is on the special edge of the special-shaped part, and the greater the part feature intensity of the edge pixel point; the gradient value The smaller the pixel, the less likely it is to be on the special edge of the special-shaped part, and the more likely it is to be on the noise edge. Introducing the gradient value of the pixel can suppress the noise interference of these non-part areas to a certain extent, so as to better focus on the key edges of the special-shaped parts.

[0039] S2: Construct the first point set and the second point set according to the part feature intensity of the edge pixel points; In one embodiment, the first point set and the second point set are constructed by the following steps: It is necessary to sort the part feature strengths of multiple edge pixels from large to small, and select the edge pixels with the largest part feature strength according to the percentage to form the first point set; the remaining edge pixels constitute the second point set. For example, there are 100 edge pixels, and select (This percentage can be adjusted according to actual conditions), then the top 20 parts with the highest characteristic strength are obtained. edge pixel points, and these 20 edge pixel points constitute the first point set, and the other 80 edge pixel points constitute the second point set.

[0040] In another embodiment, the first point set and the second point set are constructed by the following steps: The edge pixels are divided using a k-means clustering algorithm. In this embodiment, the edge pixels are divided into two categories, the category with a higher average feature intensity is the first point set, and the other category is the second point set.

[0041] The edge pixels are divided into two categories by clustering to form point sets with different levels of feature intensity. This method is more targeted and improves the effectiveness and accuracy of part feature extraction.

[0042] S3: constructing a first descriptor and a second descriptor of edge pixels in the first point set, and calculating a feature matching degree between the pixel points in the first point set of the image to be detected and the standard image.

[0043] Reference Figure 3 , step S3 includes steps S30 to S32.

[0044] S30: Divide the sampling area; construct with edge pixels as the center concentric circles; using the radial direction of the concentric circle as the dividing line, the concentric circle is divided into Finally, we get Sampling area. Figure 4 In this embodiment, , .

[0045] The radii of the concentric circles are equally divided within a certain area. For example, if four concentric circles with a minimum radius of 10 and a maximum radius of 100 need to be constructed, the radii of the four concentric circles are 10, 40, 70, and 100 respectively.

[0046] S31: construct descriptor; Calculate the number of edge pixel points belonging to the first point set in each sampling area; Calculate the number of edge pixel points belonging to the second point set in each sampling area; The number of edge pixels belonging to the first point set in each region is normalized to obtain the value of the position of the edge pixel corresponding to the sampling region in the first descriptor.

[0047] The number of edge pixels belonging to the second point set in each region is normalized to obtain the value of the position of the edge pixel corresponding to the sampling region in the second descriptor.

[0048] The normalization method is to divide the number of pixel points belonging to the first point set or the second point set in the sampling area by the number of pixel points belonging to the first point set or the second point set in all sampling areas corresponding to the edge pixel point.

[0049] Combination Figure 4 , take an edge pixel in the first point set as an example, its surrounding area is divided into 48 sampling areas, and the concentric circles are numbered from 1 to 12 along the circumference; the areas corresponding to the concentric circles along the radial direction are numbered A, B, C, and D. Calculate the number of pixels in each sampling area that belong to the first point set and sum them up. Assume that the sum is 100, and the pixel is in There are 5 pixels in the sampling area that belong to the first point set, then the corresponding pixels in the first descriptor are The location of the sampling area is =0.05.

[0050] S32: Calculate feature matching degree; The calculation of the first descriptor and the second descriptor of the edge pixel points in the first point set in the standard image is the same as the calculation of the first descriptor and the second descriptor in the image to be detected, which will not be repeated here.

[0051] The calculation formula for calculating the feature matching degree between the edge pixel points in the first point set of the image to be detected and the edge pixel points in the first point set of the standard image is: ; is the first point in the image to be detected. The pixel point is the same as the first point in the standard image. The feature matching degree of each pixel, is the first point in the image to be detected. The first descriptor of pixels; is the first point in the image to be detected. The second descriptor of pixels; is the first point in the standard image. The first descriptor of pixels; is the first point in the standard image. The second descriptor of pixels; is the information entropy of the grayscale values ​​of the pixels in the first point set of the standard image; is the information entropy of the grayscale values ​​of the pixels in the second point set of the standard image; is the cosine similarity function; for and The cosine similarity between ; for and The cosine similarity between .

[0052] It represents the similarity between the edge pixels in the first point set of the image to be detected and the edge pixels in the first point set of the standard image on the first descriptor. The larger the value, the more similar the first descriptor is in the image to be detected. The pixel is the same as the first pixel in the standard image The more similar the pixel points are in the features of high part feature intensity, the greater the feature matching degree between the two points. Similarly, the smaller the value, the greater the matching degree between the first pixel points in the image to be detected. The pixel is the same as the first pixel in the standard image The greater the difference between the two pixel points in the feature of high part feature intensity, the smaller the feature matching degree between the two points. For the convenience of calculation, we add 1 and divide by 2 to calculate Normalized.

[0053] Used to determine the weights of different context descriptors in feature matching calculation; The larger it is, the more complex the feature distribution of the first point set is, the richer the information is, and the more effective information it contains. Therefore, the first descriptor generated by the first point set has a greater impact on the correctness of the matching. The greater the weight of The smaller it is, the more uniform the feature distribution of the first point set is, the less information it contains, and the less effective information it contains. In this way, the first descriptor generated by the first point set has less influence on the correctness of matching. The smaller the weight.

[0054] and The feature matching calculation and and The calculation of is similar and will not be repeated here.

[0055] In another embodiment, the calculation formula of the feature matching degree is: ; is the first point in the image to be detected. The pixel point is the same as the first point in the standard image. The feature matching degree of each pixel point is and are respectively the first point set in the image to be detected. The first descriptor and the second descriptor of the pixel; and They are respectively the first point set in the standard image The first descriptor and the second descriptor of the pixel; and are the information entropy of the grayscale values ​​of the pixels in the first point set and the information entropy of the grayscale values ​​of the pixels in the second point set of the standard image respectively; for and The cosine similarity function between ; For the bottom The exponential function of .

[0056] This formula introduces , adjust the feature matching degree according to the difference in pixel feature intensity. If and Close, then Close to 1, it has little effect on feature matching; similarly, if and When the difference is large, The value becomes smaller, reducing the feature matching degree, so that the feature matching degree can accurately reflect the feature differences of parts and improve the detection accuracy of special-shaped parts.

[0057] S4: Determine part defects based on feature matching; Get the matching points corresponding to the edge pixels in the first point set in the standard image according to the feature matching degree: The feature matching degree between each edge pixel point in the first point set in the standard image and each pixel point in the first point set in the image to be detected is compared; each edge pixel point in the first point set of the standard image corresponds to a plurality of feature matching degrees.

[0058] The edge pixel point in the image to be detected that has the greatest feature matching degree with the standard image is taken as the matching point.

[0059] Obtain the feature matching degree corresponding to the matching point, and obtain the feature matching degree difference according to the feature matching degree of the matching point; Calculate the mean of the feature matching degrees corresponding to all matching points in the image to be detected, denoted as ; Calculate the absolute value difference between each matching point and the mean value to obtain the feature matching degree difference corresponding to the matching point.

[0060] Set the matching threshold and compare the feature matching difference with the matching threshold to determine part defects.

[0061] When the absolute value of the feature matching degree difference among multiple matching points is greater than the matching threshold, it is determined that the part is defective. In this embodiment, the matching threshold is 0.1C. During the part monitoring process, when the feature matching degree difference corresponding to multiple matching points is greater than 0.1C, the system alarms and the staff promptly reviews.

[0062] An embodiment of the present application also discloses a machine vision-based intelligent monitoring system for a special-shaped parts stamping line, comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a machine vision-based intelligent monitoring method for a special-shaped parts stamping line is implemented according to the present application.

[0063] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0064] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. An intelligent monitoring method for a special-shaped parts stamping line based on machine vision, characterized in that: The method comprises the steps of: obtaining edge pixel information in the image to be detected, and calculating the part feature intensity of the edge pixel; Constructing a first point set and a second point set according to the part feature strength of edge pixels; constructing a first descriptor and a second descriptor of edge pixels in the first point set, and calculating the feature matching degree of the pixels in the first point set between the image to be detected and the standard image; according to Feature matching determines part defects; The calculation formula of feature matching is: ; In the formula, is the first point in the image to be detected. The pixel point is the same as the first point in the standard image. The feature matching degree of each pixel, is the first point in the image to be detected. The first descriptor of pixels; is the first point in the image to be detected. The second descriptor of pixels; is the first point in the standard image. The first descriptor of pixels; is the first point in the standard image. The second descriptor of pixels; is the information entropy of the grayscale values ​​of the pixels in the first point set of the standard image; is the information entropy of the grayscale values ​​of the pixels in the second point set of the standard image; is the cosine similarity function.

2. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 1 is characterized in that: The steps of constructing the first descriptor and the second descriptor in the first point set include: Divide the sampling area; construct descriptors: calculate the number of pixels in each sampling area that belong to the first point set, and perform normalization to obtain the first descriptor of the pixel; calculate the number of pixels in each sampling area that belong to the second point set, and perform normalization to obtain the second descriptor of the pixel.

3. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 2 is characterized in that: The method for dividing the sampling area includes: constructing multiple concentric circles with different radii with edge pixels as the center; The concentric circles are evenly divided into multiple parts along the circumferential scale to form multiple sampling areas.

4. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 2 is characterized in that: The method for constructing the first point set and the second point set includes: sorting the part feature strengths of multiple edge pixels from large to small, taking some edge pixels with the largest part feature strengths according to percentage to form the first point set, and the remaining edge pixels constitute the second point set.

5. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 1 is characterized in that: The calculation formula for the edge pixel part feature intensity is: ； In the formula is the part feature intensity of any edge pixel in the image, is the gray value of the edge pixel, is the curvature of the edge pixel, To calculate the abnormal score of the edge pixel using the isolation forest algorithm, is a linear normalization function; is the pixel gradient.

6. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 1 is characterized in that: The steps to determine part defects based on feature matching include: Obtain matching points corresponding to edge pixel points in the first point set in the standard image; obtain the feature matching degree corresponding to the matching point; obtain and calculate the feature matching degree difference according to the feature matching degree of the matching point; set a matching threshold; compare the feature matching degree difference with the matching threshold to determine part defects.

7. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 6 is characterized in that: The step of acquiring matching points corresponding to edge pixel points in the first point set in the standard image comprises: comparing the feature matching degree between each edge pixel point in the first point set in the standard image and each pixel point in the first point set in the image to be detected; Each edge pixel point in the first point set of the standard image corresponds to multiple feature matching degrees, and the edge pixel point in the image to be detected with the maximum feature matching degree in the standard image is taken as the matching point.

8. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 7 is characterized in that: The step of comparing the feature matching degree difference with the matching threshold to determine the part defect includes: in response to the absolute value of the feature matching degree difference among multiple matching points being greater than the matching threshold, determining that the part has a defect.

9. The intelligent monitoring method for special-shaped parts stamping line based on machine vision according to claim 8 is characterized in that: The step of obtaining the feature matching degree difference includes: calculating the mean value of the feature matching degrees corresponding to all matching points in the image to be detected; and calculating the difference between each matching point and the mean value to obtain the feature matching degree difference.

10. An intelligent monitoring system for special-shaped parts stamping line based on machine vision, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent monitoring method for a special-shaped part stamping line based on machine vision according to any one of claims 1 to 9 is implemented.

Images