A method and system for detecting steel castings

Abstract

The present invention relates to the field of image data processing, and in particular to a method and system for detecting steel castings. The method comprises: obtaining closed edges and unclosed edges of a steel casting image, dividing the closed edge into two unclosed edges, and obtaining all unclosed edges as target edges; respectively obtaining circumscribed graphics of the closed edge and the unclosed edge; detecting each texture edge in each circumscribed graphic, generating an enclosing line to divide each circumscribed graphic into an inner area and an outer area; calculating the possibility of the target edge accordingly, and when the possibility of the target edge is greater than a threshold, the target edge is a crack; the possibility is positively correlated with a first indicator and a second indicator, and negatively correlated with a bending degree and a pixel difference of the target edge; the bending degree characterizes the tortuosity of the target edge; and the second indicator characterizes the deformation degree of each texture edge. The present invention solves the problem of inaccurate identification of crack defects in steel castings in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of image data processing and more specifically to a method and system for detecting steel castings. Background Art

[0002] During the production process of steel castings, due to various factors such as the design of the pouring system, the quality of the raw materials, and the production process, defects such as cracks, shrinkage cavities, and shrinkage may occur on the surface and inside of steel castings. These defects not only affect the appearance quality of steel castings, but may also have a serious impact on their intrinsic quality and quality of use. In particular, cracks on the surface of steel castings are a serious defect, and compared with other defects, cracks have a greater impact on steel castings. The presence of cracks will significantly reduce the strength and toughness of steel castings, making them prone to breakage during use, thereby affecting their safety and reliability. In addition, cracks may also cause the sealing of steel castings to decrease, affecting their performance in some specific application scenarios. Therefore, defect detection of steel castings, especially crack detection, is of great significance to ensure product quality and safety.

[0003] At present, the patent application document with the publication number "CN202310221714.4" and the name "A method for visual inspection of part surfaces" discloses a method of filtering and denoising the collected image, extracting a preliminary defect image from the preprocessed image through dynamic threshold segmentation, eliminating the interference area, and accurately extracting the edge position in the candidate defect image to obtain an accurate defect area image. By classifying the defect features and according to the roundness calculation formula, it can be measured whether the defects in the accurate defect area image are scratches or depressions.

[0004] The above method only considers the relationship between the original texture of the steel casting and the distortion and deformation of the defective part based on the brightness, and may identify the light shadow as a defect, resulting in inaccurate defect identification. Summary of the invention

[0005] In order to solve the problem of inaccurate identification of crack defects in steel castings in the prior art, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for detecting a steel casting, comprising:

[0007] Obtain the closed edge and the unclosed edge of the steel casting image, divide the closed edge into two unclosed edges, and obtain all the unclosed edges as the target edges; obtain the circumscribed graphics of the closed edge and the unclosed edge respectively, the pixel points of the area enclosed by the closed edge in the circumscribed graphics are taken as 0, and the values ​​of the other remaining pixel points are taken as LBP values; the pixel points on the unclosed edge in the circumscribed graphics are taken as 0, and the values ​​of the other remaining pixel points are taken as LBP values; obtain the point with the largest curvature on each texture edge in each circumscribed graphics, connect all points, obtain the enclosing line, and the enclosing line divides each circumscribed graphics into an inner area and an outer area; obtain the first and second histograms of the inner area and the outer area, and arrange the frequencies in the first and second histograms from large to small to obtain the third and fourth histograms; calculate the first index of each target edge , , and Respectively represent the number of peaks in the first and second histograms, and Represent the entropy of the first and second histograms respectively, and Respectively represent the third and fourth histograms The ratio of the peak height of a peak group to the width of the peak group; when the possibility of the target edge is greater than the threshold, the target edge is a crack; the possibility is positively correlated with the first indicator and the second indicator, and negatively correlated with the bending degree and the pixel difference of the target edge; the bending degree represents the tortuosity of the target edge; the second indicator represents the deformation degree of each texture edge.

[0008] Through the quantitative index of bending degree, the tortuosity and complexity of the edge are included in the analysis. The characteristic information of LBP image is used to construct the LBP histogram of the inner and outer regions. By calculating the multi-dimensional features such as the number of peaks, peak group ratio, entropy, etc., the texture difference between the crack area and the background area is fully reflected, thereby improving the resolution of edge cracks. The second index characterizes the slope change of each edge in the LBP image and the distance change of adjacent edges, which helps to identify the irregular edge distribution characteristics unique to cracks and make up for the cracks that may be missed by a single texture feature. With the help of the closed edge circumscribed graphics and enclosing line generation method, combined with geometric characteristic restrictions, the robustness of edge feature extraction is enhanced, laying a reliable foundation for the accurate detection of target cracks.

[0009] Preferably, dividing the closed edge into two unclosed edges comprises: using two points on the closed edge that are farthest apart as dividing points to divide the closed edge.

[0010] Preferably, the second indicator is specifically: ,in, represents the second indicator, is the total number of texture edges within the bounding graphics area, , Indicates The average slope value of the strip texture edge in the inner area and the outer area, , Indicates All pixels in the inner and outer regions on the edge of the texture strip to the The average distance between the edges of the stripes.

[0011] By constructing the second index and integrating the total number of texture edges, slope changes, and average distance characteristics of inner and outer areas, the present invention achieves a refined characterization of crack morphology and enhances the accuracy and robustness of crack detection.

[0012] Preferably, the possibilities include: ,in, For possibility, As the first indicator, As the second indicator, is the degree of bending, , and They are the average, minimum and median of the grayscale values ​​of all pixels on the edge of the target image of the steel casting.

[0013] Comprehensively considering multiple indicators from multiple dimensions can more comprehensively analyze the characteristics of the target edge and ensure the accuracy of detection.

[0014] Preferably, the bending degree includes: ,in, Indicates the degree of bending. Indicates the goodness of fit of the target edge straight line fit, Indicates the number of corner points on the edge of the target, is the total number of pixels on the edge of the target, Represents the average distance of all adjacent corner points on the target edge. Indicates the maximum distance between all adjacent corner points on the target edge. Represents the variance of the distances of all adjacent corner points on the target edge, Indicates the first The curvature of a pixel.

[0015] By measuring the degree of deviation between the target edge and the straight line fit, the overall tortuosity of the edge is reflected. The smaller the goodness of fit, the more tortuous the edge is, which helps to identify complex edge features. The number of corner points directly reflects the irregularity of the edge and the density of turning points, which is an important manifestation of crack characteristics. Combining the uniformity of corner point distribution and the difference in edge morphology helps to capture the geometric characteristics of cracks. Using the statistical value of pixel curvature, the local tortuosity analysis of the edge is further refined to effectively reflect the morphological complexity of the edge.

[0016] Preferably, the least square method is used to fit the edge of the target, a fitting straight line is obtained, and the goodness of fit is calculated; and the Harris corner point detection algorithm is used to detect the corner points on the edge of the target.

[0017] Preferably, the circumscribed figure is a circumscribed square.

[0018] The circumscribed square can cover the closed area to the maximum extent without covering too much other area, so the closed edge can be fully analyzed while reducing the amount of calculation as much as possible.

[0019] The beneficial effects of the present invention are as follows: through the quantitative index of bending degree, the tortuosity and complexity of the edge are included in the analysis, and the characteristic information of the LBP image is used to construct the LBP histogram of the inner and outer regions. By calculating the multi-dimensional features such as the number of peaks, peak group ratio, entropy, etc., the texture difference between the crack area and the background area is fully reflected, thereby improving the resolution of edge cracks. The second index helps to identify the irregular edge distribution characteristics unique to cracks by characterizing the slope change of each edge in the LBP image and the distance change of adjacent edges, and makes up for the cracks that may be missed by a single texture feature. With the help of the closed edge circumscribed graphics and the enclosing line generation method, combined with the geometric characteristic restrictions, the robustness of edge feature extraction is enhanced, laying a reliable foundation for the accurate detection of target cracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0021] Figure 1 is a flow chart of a steel casting detection method provided by an embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of an external graphic of a closed edge in a steel casting image provided by an embodiment of the present invention;

[0023] Figure 3is a schematic diagram of a steel casting detection system provided by an embodiment of the present invention;

[0024] Figure numerals: 1, first dividing point; 2, second dividing point; 3, closed edge; 4, enclosing line. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0026] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] Figure 1 is a flow chart of a steel casting detection method according to an embodiment of the present invention, comprising the following steps:

[0028] S101, obtaining closed edges and unclosed edges of a steel casting image, dividing the closed edge into two unclosed edges, and obtaining all unclosed edges as target edges.

[0029] In some embodiments, dividing the closed edge into two open edges includes: using two points on the closed edge that are farthest apart as dividing points to divide the closed edge.

[0030] It should be noted that, in some embodiments, generally, cracks in steel castings are linear regions with a certain width, and the edges of the linear regions can be detected using the Canny edge detection algorithm. Therefore, cracks detected using the Canny edge detection algorithm are generally closed edges, and closed edges in the image may also be the edges of other defects, such as pores, sand inclusions, or scars. Therefore, in some embodiments, it is necessary to first screen out potential crack edges based on the characteristics of the closed edges. Specifically, for each pixel point on the closed edge, its gradient direction is calculated, and a straight line where the gradient direction is located is generated at the pixel point position, and the pixel point at the intersection of the straight line and the closed edge is marked, which is called a matching pixel point. The distance between the two pixels is calculated, thereby obtaining the distance between each pixel point on the closed edge and its matching pixel point. For the closed edge whose average distance of all pixel pairs on the entire closed edge is less than 15 pixels (15 is a threshold reference value, which can be adjusted by the implementer according to the specific situation, such as the resolution of the image, the shooting distance, etc., which will affect the segmentation effect of the threshold on the edge), a circumscribed rectangle is drawn, and the aspect ratio of the circumscribed rectangle is calculated. When the aspect ratio is greater than or equal to 3 (3 is the aspect ratio threshold, which can be adjusted by the implementer according to the specific situation), the closed edge is called a potential crack edge. In this way, the screening of potential crack edges is achieved. Then the potential crack edge is segmented into two target edges.

[0031] S102, respectively obtain the circumscribed graphics of the closed edge and the unclosed edge, the pixel values ​​of the area enclosed by the closed edge in the circumscribed graphics are 0, and the values ​​of the other remaining pixel points are LBP values; the pixel values ​​on the unclosed edge in the circumscribed graphics are 0, and the values ​​of the other remaining pixel points are LBP values.

[0032] This step is mainly to obtain the LBP image of the closed edge and the unclosed edge circumscribed graphics, and process it, setting the values ​​of the pixels in the area surrounded by the closed edge and the pixels on the unclosed edge to 0, and the values ​​of the pixels in other places are normal LBP values. The circumscribed graphics can generally be a circumscribed square or a circumscribed circle.

[0033] The LBP image can be understood as the LBP value matrix of the entire image, and each element in the matrix corresponds to the LBP value of a pixel. The LBP value of each pixel is calculated as follows: calculate the gray value of each pixel, select a window of the target pixel, generally a 3×3 or larger window, and the target pixel is generally located at the center of the window. If the gray value of the neighborhood pixel is greater than or equal to the gray value of the center pixel, it is represented by 1, otherwise it is represented by 0, and the pixels represented by 0 and 1 are arranged in a clockwise direction to obtain a binary number, and the binary number is converted to a decimal number, which is the LBP value of the target pixel.

[0034] S103, obtaining the point with the largest curvature on each texture edge in each circumscribed figure, connecting all the points to obtain an enclosing line, the enclosing line divides each circumscribed figure into an inner area and an outer area, and calculating a first index of each target edge.

[0035] Specifically, the method for calculating the first indicator of each target edge is: obtain the first and second histograms of the inner area and the outer area, and arrange the frequencies in the first and second histograms from large to small to obtain the third and fourth histograms; the first and second histograms are LBP histograms of the inner area and the outer area, in which each group of data is sorted according to the size of the LBP value, and the third and fourth histograms are sorted according to the frequency size in each group of data. When the frequencies of several groups of data are the same, peak groups will be formed in the third and fourth histograms.

[0036] Calculate the first index of each target edge , , and Respectively represent the number of peaks in the first and second histograms, and Represent the entropy of the first and second histograms respectively, and Respectively represent the third and fourth histograms The ratio of the peak height of a peak group to the peak group width, for and The maximum value in .

[0037] The texture edge is a texture feature of the steel casting itself. Figure 2 As shown, the entire figure is the circumscribed figure of the closed edge 3, the horizontal line near the closed edge 3 is the texture edge, that is, the texture of the steel casting itself, the enclosing line 4 is obtained by connecting the points with the largest curvature on the texture edge, the area inside the enclosing line 4 is the inner area, and the entire figure except the inner area is the outer area.

[0038] Depend on Figure 2 It can be seen that in the outer region, due to the uniform stress distribution and complete structure, the texture in the outer region is relatively regular, which is reflected in the LBP histogram, which may have only a few obvious peaks, that is, smaller, the peak may be higher and narrower, i.e. Larger, regular textures result in lower entropy values, i.e. Small; in the inner area, if the target edge is a crack, the texture in the inner area will be distorted by the crack, the complexity of the texture will increase, and the number of peaks may increase, that is, is larger, the peak may become lower and wider, i.e. Smaller, more complex and irregular textures result in larger entropy values, i.e. This series of changes ultimately led to If the target edge is not a crack, the texture in the inner area will not be affected and distorted, and the complexity of the texture is similar to that in the outer area, so the various indicators in the histogram are also similar, which ultimately leads to Smaller.

[0039] S104. When the probability of the target edge is greater than a threshold, the target edge is a crack; the probability is positively correlated with the first indicator and the second indicator, and negatively correlated with the bending degree and the pixel difference of the target edge; the bending degree represents the tortuosity of the target edge; the second indicator represents the deformation degree of each texture edge.

[0040] The bending degree represents the bending degree of the target edge. ,in, Indicates the goodness of fit of the target edge straight line fit, Indicates the number of corner points on the edge of the target, is the total number of pixels on the edge of the target, Represents the average distance of all adjacent corner points on the target edge. Indicates the maximum distance between all adjacent corner points on the target edge. Represents the variance of the distances of all adjacent corner points on the target edge, Indicates the first The curvature of a pixel.

[0041] Goodness of fit The value range is between 0 and 1. The larger the value is, the closer it is to 1, indicating that the fitting effect of the fitting line on the target edge is better, that is, the straighter the target edge is, the closer the shape is to a straight line, and the degree of curvature of the target edge is The smaller, The smaller the value, the less straight the target edge is and the more curved the target edge is. The bigger. It reflects the uniformity of the distribution of corner points on the target edge. The more uniform the distribution of corner points is, the more it means that all parts of the entire edge have the characteristics of curvature, rather than only one section having the characteristics of curvature. At this time, the smaller the variance of the distances between all adjacent corner points is, the better the result is. The greater the increase of The larger the value, the higher the and It reflects the overall bending degree of the target edge. It reflects the bending degree at each pixel on the edge of the target, that is, the local bending degree. Combining the two can more accurately reflect the bending degree of the target edge.

[0042] In some embodiments, the least square method may be used to fit the edge of the target, obtain a fitting straight line, and calculate the goodness of fit; and the Harris corner point detection algorithm may be used to detect corner points on the edge of the target.

[0043] In addition, from Figure 2 It can be clearly seen that the slope of the texture of the steel casting itself in the inner and outer regions and the distance between adjacent textures have changed. Therefore, in some embodiments, the second index is used to characterize the degree of deformation of the edge of each texture.

[0044] The second indicator is specifically: ,in, represents the second indicator, is the total number of texture edges within the bounding graphics area, , Indicates The average slope value of the strip texture edge in the inner area and the outer area, , Indicates All pixels in the inner and outer regions on the edge of the texture strip to the The average distance between the edges of the stripes.

[0045] Get the first indicator , Second indicator , bending degree Then, the probability that the target edge is a crack can be calculated by combining the average, minimum and median grayscale values ​​of all pixels on the target edge of the steel casting image. for: , As the first indicator, As the second indicator, is the degree of bending, , and They are the average, minimum and median of the grayscale values ​​of all pixels on the edge of the target image of the steel casting.

[0046] Since the light reflection and scattering characteristics of the crack are different from those of the surrounding normal casting surface, its gray value is usually darker than the surrounding normal part. The larger the value, the less likely it is that the edge of the target has dark grayscale values. In this case, the edge of the target is likely to be a crack. Smaller; if the average gray value of the target edge The smaller the value, the darker the gray value is at the edge of the target, and the possibility that the target edge is a crack is high. At the same time, when the bending degree of the target edge The larger the value, the more likely the target edge is a crack. The larger the target edge is, the more curved it is. The smaller the target edge is, the more likely it is that there is a crack. The smaller the value, the smaller the first and second indicators of the texture around the edge of the target. and The larger the value, the more likely the target edge is a crack. The bigger.

[0047] For Figure 2 For example, the first segmentation point 1 and the second segmentation point 2 divide the closed edge 3 into two unclosed edges as the first target edge and the second target edge. Taking the first target edge as an example, the bending degree of the first target edge is calculated, and the encircling line 4 of the closed edge 3 is made. The encircling line 4 in the figure is the inner area, and the remaining other places are the outer area. The pixel points inside the closed edge 3 are taken as 0, and the pixel points outside the closed edge 3 are taken as LBP values, and the first index of the first target edge is calculated based on this. Then, the second index of the first target edge is calculated according to the slope of each texture edge in the inner area and the outer area in the figure, and the distance between adjacent texture edges. The bending degree of the first target edge, the first index, the second index and the average, minimum and median of the grayscale values ​​of all pixels on the first target edge in the cast steel image are combined to calculate the possibility that the first target edge is a crack.

[0048] The above-mentioned steel casting detection method provided by an embodiment of the present invention incorporates the tortuosity and complexity of the edge into the analysis through the quantitative index of bending degree, and constructs the LBP histogram of the inner and outer regions by using the characteristic information of the LBP image. By calculating the multi-dimensional features such as the number of peaks, peak group ratio, and entropy, it comprehensively reflects the texture difference between the crack area and the background area, thereby improving the resolution of edge cracks. The second indicator helps to identify the irregular edge distribution characteristics unique to cracks by characterizing the slope change of each edge in the LBP image and the distance change of adjacent edges, and makes up for the cracks that may be missed by a single texture feature. With the help of the closed edge circumscribed graphics and enclosing line generation method, combined with geometric characteristic restrictions, the robustness of edge feature extraction is enhanced, laying a reliable foundation for the detection of target cracks. Finally, the problem of inaccurate defect identification in the prior art is solved.

[0049] The present invention also provides a steel casting detection system. Figure 3 As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a steel casting detection method of the present invention is implemented.

[0050] The system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface, whose configuration and functions are known in the art and thus will not be described in detail here.

[0051] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0052] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0053] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for detecting steel castings, characterized in that: include: Acquire the closed edge and the unclosed edge of the steel casting image, divide the closed edge into two unclosed edges, and obtain all the unclosed edges as target edges; Get the circumscribed graphics of closed edges and unclosed edges respectively, the pixel values ​​of the area enclosed by the closed edges in the circumscribed graphics are 0, and the values ​​of the other remaining pixels are LBP values; the pixel values ​​on the unclosed edges in the circumscribed graphics are 0, and the values ​​of the other remaining pixels are LBP values; Obtain the point with the largest curvature on each texture edge in each circumscribed figure, connect all the points to obtain an enclosing line, and the enclosing line divides each circumscribed figure into an inner area and an outer area; obtain the first and second histograms of the inner area and the outer area, and arrange the frequencies in the first and second histograms from large to small to obtain the third and fourth histograms; Calculate the first index of each target edge , , and Respectively represent the number of peaks in the first and second histograms, and Represent the entropy of the first and second histograms respectively, and Respectively represent the third and fourth histograms The ratio of the peak height of a peak group to the width of the peak group; When the probability of a target edge is greater than a threshold, the target edge is a crack; possibility ,include: , For possibility, As the second indicator, is the degree of bending, , and are the average, minimum and median of the grayscale values ​​of all pixels on the edge of the target image of the steel casting; the bending degree represents the tortuosity of the target edge; the second index represents the deformation degree of each texture edge; The second indicator is: , is the total number of texture edges within the bounding graphics area, , Indicates The average slope value of the strip texture edge in the inner area and the outer area, , Indicates All pixels in the inner and outer regions on the edge of the texture strip to the The average distance from the edge of the strip; Degree of bend, including: , Indicates the goodness of fit of the target edge straight line fit, Indicates the number of corner points on the edge of the target, is the total number of pixels on the edge of the target, Represents the average distance of all adjacent corner points on the target edge. Indicates the maximum distance between all adjacent corner points on the target edge. Represents the variance of the distances of all adjacent corner points on the target edge, Indicates the first The curvature of a pixel.

2. The steel casting detection method according to claim 1, characterized in that: The step of dividing the closed edge into two unclosed edges includes: using two points on the closed edge that are farthest apart as dividing points to divide the closed edge.

3. The steel casting detection method according to claim 1, characterized in that: include: The least square method is used to fit the target edge, the fitting line is obtained, and the goodness of fit is calculated; the Harris corner detection algorithm is used to detect the corner points on the target edge.

4. The steel casting detection method according to claim 1, characterized in that: The circumscribed figure is a circumscribed square.

5. A steel casting inspection system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steel casting detection method according to any one of claims 1 to 4.

Images