Titanium frame container weld quality detection method and system

Abstract

The application relates to the technical field of image processing, in particular to a titanium frame container weld quality detection method and system. The method comprises the following steps: acquiring a surface image of a weld to be detected of a titanium frame container, determining a global gray scale variance of the surface image, and determining a differential scale parameter based on the global gray scale variance; constructing a smooth structure tensor of a target pixel point in the surface image by using the differential scale parameter, and performing feature decomposition on the smooth structure tensor to determine a coherence index and a main direction angle of the target pixel point, so that the target pixel point is determined as a linear path pixel or a point path pixel; obtaining a pore intensity distribution graph composed of pore intensity values by using the linear path pixel and the point path pixel; and obtaining a weld quality detection result by using a crack enhancement graph and the pore intensity distribution graph. Through the above technical scheme, the weld quality of the titanium frame container can be detected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a titanium frame container weld quality detection method and system. BACKGROUND

[0002] Titanium frame containers, as the core carriers of special transport equipment, can be applied to deep-sea resource development material transportation, polar scientific research equipment transfer and high-purity chemical product cross-border transportation, etc. fields, due to the advantages of corrosion resistance, high strength and light weight, and low temperature toughness of titanium and titanium alloy materials. The main body of the titanium frame container is formed by processing titanium and titanium alloy materials, and the connection between the components is realized by welding process.

[0003] The weld of the titanium frame container not only needs to bear the overall load and deformation resistance of the frame, but also needs to ensure the sealing performance and weather resistance of the container. The quality of the weld is the core structural unit that determines the service life and transportation safety of the titanium frame container.

[0004] The weld of the titanium frame container is mainly distributed in the connection nodes of the frame columns and beams, the splicing edges of the container panels and the frame, and the load-bearing connection of the door shaft and the frame. According to the different stress characteristics, the weld forms include fillet weld, butt weld and plug weld. Among them, the T-shaped fillet weld of the column and the beam needs to bear the impact load and vibration stress in the transportation process, which is the key position of weld quality control.

[0005] Pores and cracks are typical defects in the weld. During the welding process of titanium material, pores and cracks may be generated on the surface of the weld. Pores will destroy the compactness of the weld surface, reducing the effective stress area of the titanium frame container. Surface cracks in the weld, as fatigue crack sources, may suddenly break down under the action of alternating loads during transportation, causing instability of the container structure of the titanium frame container. Therefore, it is necessary to detect the quality of the weld of the titanium frame container. SUMMARY

[0006] In order to detect the quality of the weld of the titanium frame container, the present application provides a titanium frame container weld quality detection method and system.

[0007] According to a first aspect of the embodiments of the present application, a titanium frame container weld quality detection method is provided, including: acquiring a surface image of a weld to be detected of a titanium frame container, determining a global gray variance of the surface image, and determining a differential scale parameter based on the global gray variance; constructing a smooth structure tensor of a target pixel in the surface image using the differential scale parameter and a preset integral scale parameter, and performing feature decomposition on the smooth structure tensor to determine a coherence index and a principal direction angle of the target pixel, so as to determine whether the target pixel is a linear path pixel or a point path pixel using the coherence index; for the linear path pixel, constructing an elliptical filter kernel whose long axis rotates with the principal direction angle and whose short axis dynamically changes with the coherence index, so as to perform directional filtering on the linear path pixel using the elliptical filter kernel to obtain a crack enhancement image; for the point path pixel, performing filtering on the point path pixel using a multi-scale Laplacian of Gaussian operator to obtain a pore intensity value, so as to obtain a pore intensity distribution image composed of the pore intensity values; performing binarization processing on the crack enhancement image and the pore intensity distribution image respectively, and fusing the binarization results to obtain a target image, so as to perform connected domain geometric analysis on the target image to obtain a weld quality detection result.

[0008] In this way, the differential scale is adaptively adjusted according to the global gray variance, the best gradient calculation scale can be automatically matched according to the roughness of the weld surface, so as to effectively suppress background noise, and the local geometric features of the pixel are quantified by feature decomposition of the structure tensor, so as to realize shunt processing of linear feature defects and point feature defects.

[0009] Optionally, the step of determining the differential scale parameter based on the global gray variance includes: acquiring a preset minimum scale parameter and a maximum scale parameter , determining the differential scale parameter , wherein, is the global gray variance of the surface image, is a natural exponential function.

[0010] In this way, a nonlinear mapping relationship between the overall complexity of the surface image and the differential scale is established using an exponential decay function, when the global gray variance of the surface image is larger, the differential scale can be reduced to avoid excessive blurring of details, and when the global gray variance of the surface image is smaller, the differential scale can be increased to suppress sensor noise.

[0011] Optionally, the smooth structure tensor is constructed by the following method: a Gaussian first derivative operator with a standard deviation equal to a differential scale parameter is used to convolve the surface image to determine the horizontal gradient component and the vertical gradient component of each pixel point in the surface image, so as to construct an initial structure tensor matrix based on the horizontal gradient component and the vertical gradient component of the target pixel point; a Gaussian kernel with a standard deviation equal to an integral scale parameter is used to perform Gaussian convolution smoothing on the independent elements in the initial structure tensor matrix of the target pixel point, to obtain the smooth structure tensor of the target pixel point; the integral scale parameter is equal to an integer multiple of the differential scale parameter.

[0012] In this way, the gradient direction information in the neighborhood can be effectively integrated by performing smoothing in the tensor domain, and the introduction of the integral scale parameter enables the structure tensor to reflect not only the single-point gradient but also the local regional geometric trend.

[0013] Optionally, the coherence index of the pixel point is determined by the following method: eigenvalue decomposition is performed on the smooth structure tensor of the pixel point to obtain a first eigenvalue and a second eigenvalue; the first eigenvalue is greater than or equal to the second eigenvalue; the square of the difference between the first eigenvalue and the second eigenvalue is taken as a first intermediate variable, the square of the sum of the first eigenvalue and the second eigenvalue is taken as a second intermediate variable, and the ratio of the first intermediate variable to the second intermediate variable is taken as the coherence index of the pixel point.

[0014] In this way, the coherence index is constructed by the differential combination of the eigenvalues, which can normalize the representation of the degree of anisotropy of the local texture.

[0015] Optionally, the target pixel point is a linear path pixel or a point path pixel, which is determined by the following method: the coherence index is compared with a preset shunt threshold to determine whether the target pixel point is a linear path pixel or a point path pixel; the shunt threshold is determined in advance based on the average value and the standard deviation of the coherence index of the crack region pixel points in the linear crack sample image set.

[0016] In this way, the shunt threshold determined by the statistical method can realize accurate division of the pixel categories and ensure the adaptability to the crack features in a specific production environment.

[0017] Optionally, the elliptical filter kernel of the linear path pixel is constructed by the following method: a preset major axis radius and a short axis lower limit value are obtained, and the short axis radius of the elliptical filter kernel is determined; wherein, is the coherence index of the linear path pixel, is a rounding function; the elliptical filter kernel of the linear path pixel is constructed by using the major axis radius and the short axis radius; and the major axis direction of the elliptical filter kernel is parallel to the main direction angle of the linear path pixel.

[0018] Optionally, a crack enhancement map is obtained by performing directional filtering on linear path pixels using an elliptical filter kernel, including: for any neighboring pixel within the coverage area of ​​the elliptical filter kernel, the vertical distance from the neighboring pixel to the axis corresponding to the minor axis radius of the elliptical filter kernel is taken as a first distance, and the vertical distance from the neighboring pixel to the axis corresponding to the major axis radius of the elliptical filter kernel is taken as a second distance; a lateral attenuation term is determined using the first distance and the major axis radius, and a longitudinal attenuation term is determined using the second distance and the minor axis radius; the weighted response values ​​of the neighboring pixels of the linear path pixel are determined based on the lateral attenuation term and the longitudinal attenuation term; and a weighted filtering is performed on the linear path pixels using the weighted response values ​​of the neighboring pixels to obtain the crack enhancement map.

[0019] Optionally, the stomatal intensity value is obtained by filtering the point path pixels using a multi-scale Laplacian Gaussian operator, including: constructing a point path binary map that only marks the point path pixels; applying the Laplacian Gaussian operator to the marked positions in the point path binary map under a preset set of discrete scale spaces to obtain the absolute values ​​of the response intensity at multiple scales; and taking the maximum value among the absolute values ​​of the response intensity at multiple scales as the stomatal response intensity value of the target position for the marked target position in the point path binary map.

[0020] Optionally, the binarization result includes a first mask obtained by binarizing the crack enhancement map and a second mask obtained by binarizing the porosity intensity distribution map; fusing the binarization results to obtain the target image includes: performing a logical OR operation on the first mask and the second mask to obtain a defect distribution binary map, performing a morphological closing operation on the defect distribution binary map, and performing connected component analysis to obtain the target image.

[0021] According to a second aspect of the present application, a weld quality inspection system for titanium frame containers is provided, comprising: a processor and a memory, wherein the memory stores determination machine program instructions, and the determination machine program instructions, when executed by the processor, implement the steps of the weld quality inspection method for titanium frame containers provided in the first aspect of the present application.

[0022] The technical solutions provided by the embodiments of this application may include the following beneficial effects: acquiring a surface image of the weld to be tested of a titanium frame container, determining the global gray-level variance of the surface image, and determining differential scale parameters based on the global gray-level variance; using the differential scale parameters and preset integral scale parameters, constructing a smooth structure tensor of the target pixel in the surface image, and performing feature decomposition on the smooth structure tensor to determine the coherence index and principal direction angle of the target pixel, so as to determine whether the target pixel is a linear path pixel or a point-like path pixel using the coherence index; being able to determine the target image using linear path pixels or point-like path pixels, so as to obtain the weld quality inspection result by performing connected component geometric analysis on the target image.

[0023] It should be understood that the above general description and the following detailed description are only exemplary and explanatory and are not restrictive of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flow chart of a titanium frame container weld quality detection method according to an exemplary embodiment;

[0025] Figure 2 is a schematic diagram of a surface image of a weld to be detected;

[0026] Figure 3 is a schematic diagram of a target image obtained after processing the surface image of the weld to be detected;

[0027] Figure 4 is a structural schematic diagram of a titanium frame container weld quality detection system according to an exemplary embodiment. DETAILED DESCRIPTION

[0028] To detect the quality of the weld of the titanium frame container, the present embodiment provides a titanium frame container weld quality detection method and system, Figure 1 is a flow chart of a titanium frame container weld quality detection method according to an exemplary embodiment, as Figure 1 shown, the method comprises the following steps.

[0029] In step S101, the surface image of the weld to be detected of the titanium frame container is obtained, the global gray variance of the surface image is determined, and the differential scale parameter is determined based on the global gray variance.

[0030] In one embodiment, the step of determining the differential scale parameter based on the global gray variance comprises: obtaining a preset minimum scale parameter and a maximum scale parameter , determining the differential scale parameter , wherein, is the global gray variance of the surface image, is a natural exponential function.

[0031] In the actual titanium frame container detection process, in order to ensure the adaptability of the algorithm, the overall features of the surface image are evaluated, the collected RGB image of the weld is first converted into a single-channel gray image, the image bit depth is usually 8 bits, and the gray level range is [0, 255].

[0032] The variance of all pixel gray values in the entire surface image is taken as the global gray variance, the preset minimum scale parameter may be 0.5, and the maximum scale parameter For example, the parameter can be 3, and the setting of the parameters can be determined based on statistical results of a large number of titanium alloy weld samples, which will not be described herein again.

[0033] By introducing an exponential decay function negatively correlated with the global gray variance, when the global gray variance increases, the term tends to 0, so that the differential scale parameter tends to In the region with complex texture, a smaller differential scale can be used to capture high-frequency detail changes, and the large-scale smoothing operation can prevent the crack signal from being submerged in the background texture.

[0034] On the contrary, when the global gray variance is smaller, the term increases, tends to A larger differential scale can be used to suppress the thermal noise of the sensor itself by using stronger smoothing capability, and to ensure that the gradient mainly reflects the real structure change in the weld under the smoothed background.

[0035] In step S102, a smoothed structure tensor of a target pixel point in the surface image is constructed by using the differential scale parameter and a preset integral scale parameter, and the smoothed structure tensor is feature-decomposed to determine a coherence index and a main direction angle of the target pixel point, so as to determine whether the target pixel point is a linear path pixel or a point path pixel by using the coherence index.

[0036] In one embodiment, the smoothed structure tensor is constructed by the following manner: a Gaussian first derivative operator with a standard deviation equal to the differential scale parameter is used to convolve the surface image to determine a horizontal gradient component and a vertical gradient component of each pixel point in the surface image, so as to construct an initial structure tensor matrix based on the horizontal gradient component and the vertical gradient component of the target pixel point; a Gaussian kernel with a standard deviation equal to the integral scale parameter is used to perform Gaussian convolution smoothing on independent elements in the initial structure tensor matrix of the target pixel point respectively, to obtain the smoothed structure tensor of the target pixel point; and the integral scale parameter is equal to an integer multiple of the differential scale parameter.

[0037] A Gaussian first derivative kernel and are generated by using the determined differential scale parameter, and the two kernels are respectively convolved with the original image to obtain a horizontal gradient component and a vertical gradient component .

[0038] For any one pixel point in the surface image, an initial structure tensor matrix is constructed, and an integral scale parameter For example .

[0039] Each element in the initial structure tensor matrix is respectively convoluted and smoothed using a Gaussian kernel with a standard deviation equal to the integral scale parameter to obtain the final smoothed structure tensor.

[0040] The surface of the metal may have light reflection and fine scratches, so that the gradient direction of a single pixel point in the surface image is often chaotic. If the direction is directly analyzed on the original gradient of the pixel point, a large number of false random directions may be obtained.

[0041] The gradient information is weighted and averaged in the neighborhood of the pixel point, so that the dominant trend in a local area can be found instead of focusing on the mutation of a single pixel point. Through the smoothing processing of the tensor, the effective aggregation of the direction information of the pixel point is realized.

[0042] For example, on an edge line belonging to a crack, a single pixel point may deviate from the normal of the crack due to noise. If the gradients of most pixel points in the neighborhood covered by the integral scale point in the same direction, the smoothed structure tensor can correctly reflect the normal direction of the crack.

[0043] The integral scale parameter is set to an integer multiple of the differential scale, which can ensure the smoothing of the gradient while retaining sufficient spatial resolution, avoiding the aliasing of adjacent crack signals.

[0044] In one embodiment, the coherence index of the pixel point is determined by: performing eigenvalue decomposition on the smoothed structure tensor of the pixel point to obtain a first eigenvalue and a second eigenvalue; the first eigenvalue is greater than or equal to the second eigenvalue; the square of the difference between the first eigenvalue and the second eigenvalue is taken as a first intermediate variable, and the square of the sum of the first eigenvalue and the second eigenvalue is taken as a second intermediate variable; and the ratio of the first intermediate variable to the second intermediate variable is taken as the coherence index of the pixel point.

[0045] The obtained smoothed structure tensor is a real symmetric matrix, and the real symmetric matrix necessarily has two eigenvalues that are non-negative and real. Eigenvalue decomposition is performed on the smoothed structure tensor to obtain two eigenvalues of the smoothed structure tensor, which can be denoted as and , and .

[0046] The first intermediate variable , the second intermediate variable , the coherence index , and if the denominator is 0, it can be set to , or a positive number can be set in the denominator to avoid the denominator being 0, and embodiments of the present application will not be described hereinafter.

[0047] The value range of the coherence index is limited to between 0 and 1, so that the coherence index has a normalization characteristic; when the coherence index tends to 1, it indicates Less than , that is, the local structure has stronger heterotropism, and the pixel point meets the typical characteristics of a crack; when the coherence index tends to 0, it indicates Close to , that is, the local structure has stronger homotropism, and the pixel point meets the characteristics of a circular point or a flat area.

[0048] In one embodiment, the target pixel point is a linear path pixel or a point path pixel, which is determined by comparing the coherence index with a preset shunt threshold value, and determining that the target pixel point is a linear path pixel or a point path pixel; the shunt threshold value is determined in advance based on the average value and the standard deviation of the coherence index of the pixel points in the crack region in the linear crack sample image set.

[0049] A titanium weld sample image set containing typical linear cracks is collected in advance, the crack regions in the titanium weld sample image can be labeled in advance, and the coherence index of all pixel points in these crack regions is determined to obtain the average value and the standard deviation of the coherence index of different pixel points; the shunt threshold value can be equal to the average value of the coherence index minus the standard deviation multiplied by a predetermined multiple, for example, it can be equal to the average value of the coherence index minus 1.5 times the standard deviation; the sample statistical method is used to determine the shunt threshold value, which can adapt to the statistical law of cracks under different production processes.

[0050] Pixel points with higher coherence are highlighted to enhance linear continuity; pixel points with low coherence are highlighted to find circular mutations, avoiding shape stretching caused by using a linear filter to process pores, and avoiding cracking caused by using a spot filter to process cracks, thereby improving the classification and recognition ability of the detection system.

[0051] In step S103, for the linear path pixel, an elliptical filter kernel whose major axis rotates with the main direction angle and whose minor axis dynamically changes with the coherence index is constructed to perform directional filtering on the linear path pixel using the elliptical filter kernel to obtain a crack enhancement image.

[0052] In one embodiment, the elliptical filter kernel of the linear path pixel is constructed by: obtaining a preset major axis radius and a minor axis lower limit value , determining the minor axis radius of the elliptical filter kernel; wherein is the coherence index of the linear path pixel, is an integral function; an elliptical filter kernel of linear path pixels is constructed using the major axis radius and the minor axis radius; the major axis direction of the elliptical filter kernel is parallel to the main direction angle of the linear path pixels.

[0053] The major axis radius can be preset and the minor axis lower limit value can be preset For example, the preset major axis radius can be equal to 15 pixels, and the preset minor axis lower limit value can be equal to 5 pixels. The minor axis lower limit value affects the filtering width of the elliptical filter kernel in the narrowest case, and the major axis radius affects the range of smoothing along the crack direction of the elliptical filter kernel.

[0054] For each pixel determined as a linear path pixel, the coherence index and the main direction angle of the linear path pixel can be determined. The main direction angle can be determined by the structural tensor eigenvector direction. The main direction of the pixel point is along the direction of the crack trend.

[0055] The real crack often presents an intermittent feature in the image. If the minor axis radius is too small, although the sharpness of the edge after filtering can be maintained, the small crack segments cannot be connected. If the minor axis radius is too large, the crack edge will be blurred after filtering, resulting in a decrease in the contrast of the crack edge.

[0056] When the coherence is close to 1, tends to The elliptical filter kernel appears as a slender needle-shaped ellipse. The elliptical filter kernel can perform long-distance smoothing along the crack trend, suppresses the noise inside the crack, and has lower blur in the vertical direction, thereby sharpening the crack to the greatest extent.

[0057] When the coherence is lower, for example, the pixel point is at the end of the crack or in the blur area, becomes larger, The elliptical filter kernel becomes more rounded, increases the smoothing range of the elliptical filter kernel in the transverse direction, can introduce surrounding information to repair the crack signal in the area with lower contrast of crack features, and maintains the sharpness in the area with more obvious crack features, so that the shape of the elliptical filter kernel is adaptively matched with the local features of the image.

[0058] In one embodiment, the crack enhancement map is obtained by directional filtering of the linear path pixel using an elliptical filter kernel, including: for any neighborhood pixel point within the elliptical filter kernel coverage area, taking the vertical distance of the neighborhood pixel point to the short axis radius corresponding axis of the elliptical filter kernel as a first distance, and taking the vertical distance of the neighborhood pixel point to the long axis radius corresponding axis of the elliptical filter kernel as a second distance; determining a horizontal attenuation term using the first distance and the long axis radius, determining a vertical attenuation term using the second distance and the short axis radius, and determining a weight response value of the neighborhood pixel point of the linear path pixel based on the horizontal attenuation term and the vertical attenuation term; and performing weighted filtering on the linear path pixel using the weight response value of the neighborhood pixel point to obtain the crack enhancement map.

[0059] The target pixel point can be any one pixel point in the surface image of the weld to be measured, and a local coordinate system can be established with the target pixel point as the origin, wherein the major direction of the target pixel point is the long axis direction, and the direction perpendicular to the major direction of the target pixel point is the short axis direction.

[0060] For any neighborhood pixel point within the elliptical filter kernel of the target pixel point, the coordinates of the neighborhood pixel point in the local coordinate system are calculated by coordinate rotation , the first distance is the vertical distance of the neighborhood pixel point to the short axis radius corresponding axis of the elliptical filter kernel, the second distance is the vertical distance of the neighborhood pixel point to the long axis radius corresponding axis of the elliptical filter kernel, and the local coordinate system of the target pixel point takes the target pixel point as the origin, so that is equal to the first distance, is equal to the second distance.

[0061] A weight function in the form of a Gaussian function can be constructed for the neighborhood pixel point located within the elliptical filter kernel of the target pixel point , wherein the first term multiplied in the weight function is a horizontal attenuation term, the second term multiplied in the weight function is a vertical attenuation term, and exp is a natural exponential function.

[0062] After normalizing the weight response values of all neighborhood pixel points within the elliptical filter kernel of the target pixel point, the pixel values of all neighborhood pixel points within the elliptical filter kernel of the target pixel point can be weighted and summed using the weight response values of the neighborhood pixel points, and the weighted summation result is taken as the pixel value of the target pixel point in the crack enhancement map.

[0063] The Gaussian attenuation weight is adopted, so that the pixels closer to the center and closer to the crack axis contribute more. The weight response value determined in the embodiment of the present application can selectively retain the low-frequency components along the axis direction, and retain the low-frequency components along the The high-frequency component in the axial direction makes the weak crack signal that can be submerged by noise be continuously highlighted and connected, and the surrounding clutter texture can be smoothly suppressed due to the mismatch in direction, thereby improving the signal-to-noise ratio of the crack in the obtained crack enhancement image.

[0064] In step S104, the point path pixel is filtered by using the multi-scale Laplace Gaussian operator to obtain the pore intensity value, so as to obtain a pore intensity distribution composed of the pore intensity value.

[0065] In one embodiment, filtering the point path pixel by using the multi-scale Laplace Gaussian operator to obtain the pore intensity value comprises: constructing a point path binary image only marking the point path pixel, applying the Laplace Gaussian operator to the marked position in the point path binary image under a preset group of discrete scale spaces to obtain the response intensity absolute value under a plurality of scales respectively; and taking the maximum value in the response intensity absolute value under a plurality of scales as the pore response intensity value of the target position for the marked target position in the point path binary image.

[0066] A mask that only retains the point path pixel determined in the surface image can be generated, and the obtained mask is taken as the point path binary image.

[0067] A group of discrete scale parameters {1, 1.5, 2, 2.5} can be defined in advance, different scale parameters correspond to potential pores of different scale parameter diameters respectively, and the point path binary image is convolved under these different scale parameters by using the Laplace Gaussian operator.

[0068] For each pixel point in the point path binary image, the absolute value of the response value of the pixel point under different scales can be extracted, and the maximum value in the absolute value of the response value of the same pixel point under different scales is taken as the pore response intensity value of the pixel point.

[0069] The causes of the pore defects in the titanium frame container weld may be relatively complex. The pore defects may be small pinholes (for example, less than 0.2 mm in diameter) formed by hydrogen precipitation, or large holes (for example, greater than 2 mm in diameter) formed by insufficient protective gas, and a single scale Laplace Gaussian operator can only produce the maximum response to a spot of a specific size.

[0070] The second derivative of Laplace-Gaussian is sensitive to local extreme points of gray value. By constructing a scale space containing multiple scale parameters and taking the maximum response, the scale parameter that resonates with the local gray distribution can be found, which not only ensures that pores of different sizes can be highlighted, but also suppresses the uneven background light due to the zero-crossing point characteristic of the Laplace-Gaussian operator, so that the center of the pore presents a higher brightness value in the pore intensity distribution graph, while the background presents a black color close to 0.

[0071] In step S105, the crack enhancement graph and the pore intensity distribution graph are binarized respectively, and the binarization results are fused to obtain a target image. The target image is subjected to connected domain geometric analysis to obtain a welding seam quality detection result.

[0072] In one embodiment, the binarization results include a first mask obtained by binarizing the crack enhancement graph, and a second mask obtained by binarizing the pore intensity distribution graph; and the fusing of the binarization results to obtain the target image includes: performing a logical OR operation on the first mask and the second mask to obtain a defect distribution binary graph, performing a morphological closing operation on the defect distribution binary graph, and performing connected region analysis to obtain the target image.

[0073] The crack enhancement graph and the pore intensity distribution graph can be binarized by the maximum inter-class variance method or a fixed threshold, for example, the crack enhancement graph is binarized to obtain a first mask, and the pore intensity distribution graph is binarized to obtain a second mask.

[0074] The first mask and the second mask are subjected to a pixel-by-pixel logical OR operation to obtain a defect distribution binary graph, and the defect distribution binary graph is subjected to a morphological closing operation, which means that an expansion operation is followed by an erosion operation. The structural element in the morphological closing operation can be a 3x3 rectangular kernel.

[0075] Cracks and pores can be generated at the same time, or can be different manifestations of the same defect, for example, the end of a crack can be identified as a pore. Through the logical OR operation, all suspicious defect areas can be unified into a panoramic graph.

[0076] The morphological closing operation is used to repair the small cracks that may be generated in the binarization process, for example, a long and narrow crack may be disconnected into two sections due to the low contrast of some intermediate pixels. The closing operation can reconnect these adjacent breakpoints to restore the complete topological structure of the crack.

[0077] In addition to connecting possible fractures, the morphological closing operation can also fill possible holes inside the pores due to high light reflection, so that the finally obtained connected domain is a complete geometric figure, which can facilitate subsequent calculation of the area, perimeter and aspect ratio of the connected domain, and realize actual and complete morphological description of the connected domain.

[0078] Figure 2 is a schematic diagram of a surface image of a weld to be tested, as shown in Figure 2 , there are partial crack defects and pore defects in the surface image of the weld to be tested, and after processing Figure 2 , a target image as shown in Figure 3 can be obtained, as shown in Figure 3 , the extraction of crack defects and pore defects in Figure 2 is realized, which can facilitate subsequent analysis according to the characteristics of the connected domain in Figure 3 .

[0079] In one embodiment, the target image is subjected to connected domain geometric analysis to obtain a weld quality detection result, including: removing connected domains in the target image with a pixel area less than a preset noise threshold, determining the aspect ratio parameter and the circularity parameter of the circumscribed rectangle of the remaining connected domains after removal; if the aspect ratio parameter of the connected domain is greater than a first preset threshold, or the circularity parameter of the connected domain is less than a second preset threshold, the connected domain is determined to be a linear crack defect; if the aspect ratio parameter of the connected domain is less than or equal to the first preset threshold, and the circularity parameter of the connected domain is greater than the second preset threshold, the connected domain is determined to be a point-like pore defect; and the determined point-like pore defect and linear crack defect are used to output the weld quality detection result.

[0080] For example, all connected domains in the target image can be traversed first, and the pixel areas of different connected domains are calculated respectively. If the pixel area of a connected domain is less than a preset noise threshold, the connected domain can be removed as an isolated noise point. The preset noise threshold may, for example, be equal to the pixel area corresponding to 10 pixels in the image; wherein the preset noise threshold can be determined according to the pixel value of the pixel point pre-determined as noise in the target image corresponding to the weld sample image.

[0081] For the remaining connected domains after the removal operation, the aspect ratio and the circularity of the smallest circumscribed rectangle of the remaining connected domains can be calculated. The first preset threshold may, for example, be equal to 3, and the second preset threshold may, for example, be equal to 0.6.

[0082] If the aspect ratio is greater than the first preset threshold, or the circularity is less than the second preset threshold, the connected domain can be determined to be a linear crack; on the contrary, if the aspect ratio is less than or equal to the first preset threshold, and the circularity is greater than or equal to the second preset threshold, the connected domain can be determined to be a point-like pore.

[0083] The real cracks are characterized by being elongated, for example, having a higher aspect ratio and a lower roundness, while the real pores are generally more close to a circle, having a lower aspect ratio and a higher roundness; the aspect ratio describes the extension of the object, and the roundness describes the compactness of the object, and the combination of the aspect ratio and the roundness can effectively reduce the false positive rate.

[0084] In one embodiment, the determined point-like pore defects and linear crack defects are used to output the weld quality detection result, including: taking the sum of the areas of all connected domains determined as linear crack defects as the total crack area, and taking the sum of the areas of all connected domains determined as point-like pore defects as the total pore area; determining the weld quality grade according to the total crack area and the total pore area, and outputting the weld quality grade as the weld quality detection result.

[0085] The total number of pixels marked as linear cracks in the entire weld image can be counted to determine the total crack area, and the total number of pixels marked as point-like pores in the entire weld image can be counted to determine the total pore area.

[0086] The grading thresholds for the crack area and the pore area can be set according to industry standards or other standards, and different grades correspond to different total pore areas and different total crack areas, which will not be described in detail in this embodiment.

[0087] Different types or degrees of defects have different effects on structural strength, and by classifying and counting the areas, complex image information can be converted into simpler grading information to guide the sorting system on the production line to act.

[0088] The quantitative grading mechanism can not only realize the digitization of quality control, but also facilitate long-term statistics for feedback of the welding process; for example, if the abnormal proportion of the pore area in the recent preset time period is greater than the predetermined proportion, it indicates that the flow of the welding protective gas may be abnormal; if the abnormal proportion of the total crack area in the recent preset time period is greater than the predetermined proportion, it indicates that the cooling speed may be too fast.

[0089] Figure 4 is a structural schematic diagram of a titanium frame container weld quality detection system 1000 according to an example embodiment. Referring to Figure 4 , the titanium frame container weld quality detection system 1000 includes a processor 1100 and a memory 1200, the memory 1200 stores determination machine program instructions, and the determination machine program instructions are executed by the processor 1100 to realize all or part of the steps of the titanium frame container weld quality detection method in the present application.

[0090] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

[0091] It should be understood that the application is not limited to the precise construction and combinations of parts and steps described above and shown in the accompanying drawings.

Claims

1. A method for inspecting the weld quality of a titanium frame container, characterized in that, include: Surface images of the weld seams to be tested in a titanium frame container are acquired, the global grayscale variance of the surface images is determined, and the differential scale parameters are determined based on the global grayscale variance. Using differential scale parameters and preset integral scale parameters, a smooth structure tensor for target pixels in a surface image is constructed, including: convolving the surface image with a Gaussian first derivative operator whose standard deviation equals the differential scale parameter to determine the horizontal and vertical gradient components of each pixel in the surface image, and constructing an initial structure tensor matrix based on the horizontal and vertical gradient components of the target pixels; smoothing the independent elements in the initial structure tensor matrix of the target pixels by using a Gaussian kernel with a standard deviation equal to the integral scale parameter, and obtaining the smooth structure tensor of the target pixels; the integral scale parameter is an integer multiple of the differential scale parameter; The smooth structure tensor is decomposed to determine the coherence index and principal direction angle of the target pixel, so as to determine whether the target pixel is a linear path pixel or a point path pixel. For linear path pixels, an elliptical filter kernel is constructed with the major axis rotating with the principal direction angle and the minor axis dynamically changing with the coherence index. The elliptical filter kernel is used to perform directional filtering on linear path pixels to obtain crack enhancement maps. For point-like path pixels, the multi-scale Laplacian Gaussian operator is used to filter the point-like path pixels to obtain stomatal intensity values, thereby obtaining a stomatal intensity distribution map composed of stomatal intensity values. The crack enhancement map and the porosity intensity distribution map are binarized separately, and the binarization results are fused to obtain the target image. Connected component geometric analysis is then performed on the target image to obtain the weld quality inspection result.

2. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The steps for determining the differential scale parameter based on the global gray-level variance include: Obtain the preset minimum scale parameter and maximum scale parameter Determine the differential scale parameters ,in, The global grayscale variance of the surface image. It is a natural exponential function.

3. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The coherence index of a pixel is determined in the following way: The smooth structure tensor of a pixel is decomposed to obtain a first eigenvalue and a second eigenvalue; the first eigenvalue is greater than or equal to the second eigenvalue. The square of the difference between the first feature value and the second feature value is used as the first intermediate variable, the square of the sum of the first feature value and the second feature value is used as the second intermediate variable, and the ratio of the first intermediate variable to the second intermediate variable is used as the coherence index of the pixel.

4. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The target pixel is determined as either a linear path pixel or a dotted path pixel in the following way: The coherence index is compared with the preset diversion threshold to determine whether the target pixel is a linear path pixel or a dotted path image. The shunting threshold is predetermined based on the average value and standard deviation of the coherence index of pixels in the crack region in the linear crack sample image set.

5. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The elliptic filter kernel for linear path pixels is constructed in the following way: Get the preset major axis radius and the lower limit of the minor axis Determine the minor axis radius of the elliptic filter kernel. ;in, This is a coherence index for pixels along a linear path. It is a rounding function; An elliptic filter kernel for linear path pixels is constructed using the major axis radius and the minor axis radius; the major axis direction of the elliptic filter kernel is parallel to the principal direction angle of the linear path pixels.

6. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, Crack enhancement maps are obtained by directional filtering of linear path pixels using an elliptic filter kernel, including: For any neighboring pixel within the coverage area of ​​the elliptical filter kernel, the vertical distance from the neighboring pixel to the axis corresponding to the minor axis radius of the elliptical filter kernel is taken as the first distance, and the vertical distance from the neighboring pixel to the axis corresponding to the major axis radius of the elliptical filter kernel is taken as the second distance. The lateral attenuation term is determined using the first distance and the major axis radius, and the longitudinal attenuation term is determined using the second distance and the minor axis radius. The weighted response values ​​of the neighboring pixels of the linear path pixel are determined based on the lateral and longitudinal attenuation terms. The crack enhancement map is obtained by weighted filtering of the linear path pixels using the weighted response values ​​of the neighboring pixels.

7. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The porosity values ​​are obtained by filtering point-like path pixels using a multi-scale Laplacian Gaussian operator, including: Construct a binary map of point paths that only marks the pixels of point paths. In a set of preset discrete scale spaces, apply the Laplacian Gaussian operator to the marked positions in the binary map of point paths to obtain the absolute values ​​of response intensity at multiple scales. For the target location marked in the binary map of the point path, the maximum value among the absolute values ​​of the response intensity at multiple scales is taken as the pore response intensity value of the target location.

8. The method for inspecting the weld quality of titanium frame containers according to claim 1, characterized in that, The binarization results include a first mask obtained by binarizing the crack enhancement map and a second mask obtained by binarizing the porosity intensity distribution map. The process of fusing the binarization results to obtain the target image includes: A binary map of defect distribution is obtained by performing a logical OR operation on the first mask and the second mask. A morphological closing operation is then performed on the binary map of defect distribution, and connected component analysis is performed to obtain the target image.

9. A weld quality inspection system for titanium frame containers, characterized in that, include: A processor and a memory, the memory storing deterministic program instructions, which, when executed by the processor, implement the method for inspecting weld quality of titanium frame containers according to any one of claims 1-8.

Images