An image processing-based soft connection lamination side seam weld quality detection method

Abstract

The application belongs to the technical field of image processing, and particularly relates to a soft connection lamination side weld quality detection method based on image processing, which comprises the following steps: constructing a structure tensor according to a gradient component of a target pixel point and obtaining an extension direction through eigenvalue decomposition; performing a morphological closing operation on an original image based on the extension direction to obtain a fitting background value, and calculating a contrast residual in combination with a gray value; searching for a background reference point along a direction perpendicular to the extension direction, and calculating a lateral inhibition weight according to a contrast residual difference between the target pixel point and the background reference point; determining a trajectory cumulative path length according to the contrast residual, calculating a product of the contrast residual of each point on the cumulative path along the extension direction and the corresponding lateral inhibition weight, and obtaining a final defect response value; performing binarization on the defect response graph, and determining the quality of the soft connection lamination according to a connected domain area. The application improves the defect detection accuracy of the soft connection lamination.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a method for detecting the quality of soft connection lamination side welds based on image processing. BACKGROUND

[0002] Soft connection laminations are usually composed of dozens of metal foil layers stacked together, and the side edges are fused into one by polymer diffusion welding or ultrasonic welding process. In the production process, if the welding parameters are not properly controlled, interlayer defects of un-fusion will occur between the layers of laminations. Such defects will increase the contact resistance, leading to equipment overheating or even burning. Therefore, after industrial production, the compactness of the soft connection side welds needs to be detected for quality.

[0003] Due to the dual characteristics of high-reflective metal surface and complex curved geometry of soft connection laminations, general image processing techniques face serious challenges in detecting such defects. First, the reliable detection of interlayer un-fusion defects is affected by the bending deformation of laminations. In the bending or inclined area of soft connection laminations, the local layer stacking direction changes continuously with the position, while the fixed direction detection algorithm cannot adapt to this dynamically changing geometry. This geometric mismatch leads to a mismatch between the detection reference and the real defect morphology, reducing the detection reliability.

[0004] In addition, the unevenness of lighting conditions further exacerbates the detection difficulty. In the dark area, the gray difference between the background and the defect is small, and the real interlayer un-fusion defect is easily submerged by the background. While in the bright area, the strong structural noise may mask the weak defect signal. At the same time, surface scratches and real defects often present similar edge features on the image, making it difficult for traditional edge detection algorithms to effectively distinguish them, increasing the false positive rate and the missed detection rate. SUMMARY

[0005] To solve the technical problem that the soft connection lamination weak defect is difficult to accurately detect under the conditions of complex geometry and variable illumination, the present application provides a soft connection lamination side seam quality detection method based on image processing, which comprises: collecting a soft connection lamination side image, taking any pixel point in the image as a target pixel point, decomposing the structure tensor matrix of the target pixel point, and obtaining the extension direction according to the feature vector corresponding to the minimum eigenvalue obtained by decomposition; presetting a linear structure element, performing morphological operation on the linear structure element rotated in the extension direction of the target pixel point to obtain the fitting background value at the target pixel point, and calculating the contrast residual value of the target pixel point according to the gray value and the fitting background value of the target pixel point; searching for the background reference point of the target pixel point in the vertical direction of the extension direction to calculate the lateral suppression weight of the target pixel point; the background reference point is the pixel point with the maximum gray value in the clockwise and counterclockwise vertical direction of the extension direction within the preset maximum search radius; determining the length of the trajectory accumulation path based on the contrast residual value of the target pixel point; taking the target pixel point as the center, generating a trajectory accumulation path along the extension direction with the length of the accumulation path; obtaining the final defect response value of the target pixel point according to the contrast residual value and the lateral suppression weight of the pixel points on the trajectory accumulation path; combining the final defect response values of all pixel points into a final defect response graph and performing binaryzation processing, and determining the soft connection lamination side seam quality according to the area of the connected domain after binaryzation.

[0006] The present application obtains the local extension direction of the pixel point by calculating the feature vector of the structure tensor, so that the feature extraction process conforms to the actual texture direction, and the direction mismatch phenomenon is improved; by rotating the linear structure element based on the extension direction to obtain the fitting background and calculating the contrast residual value containing the normalization factor, the interference of uneven illumination and metal reflection on the gray feature is reduced, and the detection reference of bright and dark areas is unified; by analyzing the gray distribution characteristics in the vertical direction of the extension direction, the lateral suppression weight is calculated using the background reference point, the defects with wave peak characteristics are distinguished from the processing scratches with step characteristics, and the false alarm caused by the pseudo edge is suppressed; on this basis, by generating the trajectory accumulation path along the extension direction and dynamically adjusting the path length and calculating the final defect response value according to the contrast residual, the connectivity of the weak intermittent signal is enhanced, and the small gap disturbed by noise can be identified; the present application can perform adaptive detection of defects in curved areas, and improves the weak signal extraction effect under the conditions of complex structure and variable illumination.

[0007] Preferably, the structure tensor matrix of the target pixel point is decomposed, and the extension direction is obtained according to the feature vector corresponding to the minimum eigenvalue obtained by decomposition, which comprises: the structure tensor matrix of the target pixel point is decomposed by using the second-order matrix eigenvalue decomposition method to obtain the eigenvalue and the corresponding feature vector, and the extension direction of the target pixel point is obtained according to the feature vector corresponding to the minimum eigenvalue of the target pixel point.

[0008] Preferably, the contrast residual value satisfies the expression: ; in the formula, is the contrast residual value of the target pixel point, is the original image gray value of the target pixel point, is the fitted background value of the target pixel point, is a constant to prevent the denominator from being zero.

[0009] The present application enhances the response ability of the contrast residual value to weak gray difference in the area with small fitted background value by normalizing the difference between the original image gray value and the fitted background value, improves the missed detection phenomenon, and suppresses the numerical fluctuation caused by high gray background in the area with large fitted background value, reduces the false alarm risk.

[0010] Preferably, the lateral suppression weight satisfies the expression: ; in the formula, is the lateral suppression weight of the target pixel point, is the contrast residual value of the target pixel point, is the contrast residual value of the background reference point in the clockwise vertical direction of the target pixel point along the extension direction, is the contrast residual value of the background reference point in the counterclockwise vertical direction of the target pixel point along the extension direction, is the standard deviation of the Gaussian function.

[0011] The present application constructs a lateral suppression weight by using the characteristics that the interlayer unfused defects present high contrast residual value of the middle pixel point and low contrast residual value of the two side pixel points in the horizontal distribution, and present the characteristics difference that the non-defect edge presents high contrast residual value on one side and low contrast residual value on the other side. In the subsequent defect enhancement stage, the lateral suppression weight can effectively suppress the false defect signal caused by the normal edge of the copper foil, and improve the purity of detection.

[0012] Preferably, the length of the trajectory accumulation path satisfies the expression: ; in the formula, is the length of the trajectory accumulation path of the target pixel point, is a basic length constant, is a gain length coefficient, is the contrast residual value of the target pixel point, is a Sigmoid function that maps the input to the interval of 0 to 1, is a rounding function.

[0013] Preferably, the preset linear structure element, according to the linear structure element after the target pixel point is rotated in the extension direction, performs morphological operation to obtain the fitting background value at the target pixel point, comprising: defining a linear structure element with a rectangle ratio of 2:1 with the target pixel point as the center, defining the long side of the linear structure element as the long axis, rotating the long axis to be perpendicular to the extension direction, and performing morphological closing operation on the original image by using the rotated linear structure element to obtain the fitting background value at the target pixel point.

[0014] The application dynamically adjusts the direction of the linear structure element according to the local laminated extension direction of each pixel point. This way ensures that the morphological operation is always perpendicular to the interlayer unfused defects, and no matter how the laminates are curved and deformed, the defect features can be effectively extracted while the surface scratch interference is inhibited.

[0015] Preferably, the final defect response value satisfies the expression: ; in the formula, is the final defect response value of the target pixel point, is the contrast residual value of the target pixel point, k is the index of the track accumulation path length, is the track accumulation path length of the target pixel point, is the contrast residual value of the kth pixel point on the track accumulation path, is the lateral inhibition weight of the kth pixel point on the track accumulation path.

[0016] Preferably, the background reference point is obtained by: starting from the target pixel point, traversing the pixels in the clockwise vertical direction of the extension direction and the counterclockwise vertical direction of the extension direction respectively, selecting the pixel point with the maximum gray value in the clockwise vertical direction of the extension direction as the background reference point in the clockwise vertical direction of the extension direction, and selecting the pixel point with the maximum gray value in the counterclockwise vertical direction of the extension direction as the background reference point in the counterclockwise vertical direction of the extension direction within the maximum search radius range.

[0017] Preferably, the final defect response values of all pixel points are combined into a final defect response graph and binarized, comprising: traversing the final defect response values of all pixel points to obtain a final defect response graph, and binarizing the final defect response graph by using the Otsu algorithm to obtain a binary image.

[0018] Preferably, the generating a trajectory accumulation path with an accumulated path length along the extension direction comprises: iteratively extending in unit pixel steps along the extension direction and the opposite direction of the extension direction of the target pixel point; during the extending, each time extending one step, calculating the extending position of the next step along the extension direction of the current reached pixel position, and rounding off the non-integer coordinates of the position to the nearest integer coordinates until the trajectory accumulation path length is reached, to generate a trajectory accumulation path.

[0019] The present application solves the problems of missing detection and discontinuous detection in the curved area by constructing a trajectory accumulation path and locking the trajectory accumulation path on the curved trajectory of the laminated structure feature, and ensures that the weak defect signal can still be collected completely in the curved or inclined area of the soft connection lamination.

[0020] The present application has the beneficial effects that: the present application obtains the local extension direction of the pixel point by calculating the eigenvalue and eigenvector of the structure tensor of the pixel point in the soft connection lamination image, and overcomes the defect that the defect signal is weakened in the curved area due to direction mismatch in traditional image processing; the linear structure element is rotated according to the extension direction to obtain the fitting background value, and the contrast residual value considering the local background brightness is calculated, which realizes the suppression of uneven illumination and background noise and provides a data basis for the evaluation of defect features; by analyzing the gray level change characteristics of the pixel point in the vertical direction of the extension direction, a lateral suppression weight based on the difference of the background reference point is constructed, which can identify the interlayer unfused defects with the distribution characteristics of high in the middle and low on both sides, so as to retain the true defect signal while suppressing the machining scratch and other pseudo-edge interference; further, by constructing a trajectory accumulation path along the extension direction, the path length is dynamically adjusted by using the Sigmoid function and the contrast residual value, and the final defect response value is calculated in combination with the lateral suppression weight, so that the discontinuous weak defect signal can be connected and enhanced along the actual texture direction, avoiding the missing detection phenomenon; the connected domain area statistics of the final defect response graph provides a measurement basis for the quality judgment of the soft connection lamination, and improves the detection precision of the soft connection lamination. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flow chart schematically showing a soft connection lamination side seam quality detection method based on image processing in the present application;

[0022] Figure 2 is a schematic diagram showing the rotation state of a linear structure element;

[0023] Figure 3 is a gray scale image of a soft connection lamination;

[0024] Figure 4 is a binary image of a defect area of a soft connection lamination. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by a person skilled in the art without creative work belong to the protection scope of the present application.

[0026] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0027] The embodiments of the present application disclose a soft connection lamination side edge weld quality detection method based on image processing, referring to Figure 1 , comprising steps S1 to S5:

[0028] S1, collecting images of the soft connection lamination and calculating the extension direction of each pixel point in the image.

[0029] It should be noted that the surface layer structure feature trend of the soft connection lamination in the bending or inclined area will continuously change with the spatial position. The fixed direction operator cannot adapt to the continuously changing structure trend, resulting in geometric mismatch between the operator direction and the actual texture direction. This geometric mismatch further weakens the weak defect signal between the bending area and the interlayer non-fusion, resulting in that the defect is often segmented into discontinuous fragments and cannot be completely identified, thereby increasing the missed detection rate. In order to enable the subsequent processing to follow the actual texture direction of the lamination, the present application calculates the local extension direction of each pixel point in the image to provide accurate direction reference for subsequent feature extraction.

[0030] Specifically, the soft connection lamination side edge image is collected by an industrial camera, and a gray scale conversion process is performed to obtain an original image. Taking any pixel point in the original image as a target pixel point, the gradient components of the target pixel point in the horizontal direction and the vertical direction are calculated by using the Sobel operator. The structure tensor matrix of the target pixel point is constructed by using the gradient components of the target pixel point. The structure tensor matrix of the target pixel point is decomposed by using the second-order matrix eigenvalue decomposition method to obtain the eigenvalues and the corresponding eigenvectors. The extension direction of the target pixel point is obtained according to the eigenvector corresponding to the minimum eigenvalue of the target pixel point.

[0031] It should be noted that the minimum eigenvalue of the structure tensor corresponds to the direction with the most gentle gray scale change, and in the soft connection lamination, the extension direction of the layer structure is the direction with the most gentle gray scale change, so the direction corresponding to the minimum eigenvalue of the structure tensor is the extension direction of the layer structure.

[0032] S2, constructing a self-adaptive rotating linear structure element based on the extension direction, and calculating the contrast residual value of the pixel point.

[0033] It should be noted that the metal reflection of the soft connection lamination side edge will produce uneven light distribution in the imaging process, and the uneven light makes the gray scale features of the defects no longer globally comparable, which is specifically manifested as: in the dark area, the real interlayer unfused defects are submerged by the background due to the small difference in gray scale, resulting in missed detection; and in the bright area, the normal layer structure features are easily misjudged as defects due to the strong reflection and the gray scale fluctuation amplitude even exceeding that of the defects in the dark area. In order to construct a seamless background reference capable of offsetting the light change and eliminating the scratch interference, the present application rotates the linear structure element of each pixel point according to the local extension direction to perform morphological processing on the image, and measures the defect features through the relative gray scale difference.

[0034] Specifically, a linear structure element is defined with the target pixel point as the center, the width of the linear structure element is set as W pixels, the length of the linear structure element is set as twice the number of width pixels, the long side of the linear structure element is set as the major axis, and in the initial state, the linear structure element of the target pixel point is a rectangle with the major axis along the horizontal direction. The linear structure element of the target pixel point is rotated along the major axis direction, so that the major axis direction of the linear structure element is perpendicular to the extension direction of the target pixel point. The target pixel point position of the original image is subjected to morphological closing operation by using the rotated linear structure element, to obtain a fitting background value of the target pixel point position. The fitting background values of all pixel points in the original image are traversed to obtain a fitting background image; and a contrast residual value of the target pixel point is calculated according to the gray scale value and the fitting background value of the target pixel point. In this embodiment, The width of the linear structure element is defined, which is usually an odd number to ensure symmetry, and the experience value range is pixels. In this embodiment, the parameter is set as 5. In other embodiments, the implementation personnel can adjust the preset value according to the actual resolution, for example, when the resolution of the imaging system is high, resulting in an increase in the pixel width of the target texture in the image, the parameter should be increased accordingly to ensure that the structure element size matches the actual physical texture.

[0035] Specifically, the contrast residual value satisfies the relationship formula:

[0036] ;

[0037] In the formula, is the contrast residual value of the target pixel point, is the original image gray scale value of the target pixel point, is the fitting background value of the target pixel point, is a constant for preventing the denominator from being zero. In this embodiment, is , the implementer can adjust the actual resolution value.

[0038] wherein, represents the absolute gray scale difference between the target pixel point and the surrounding background, The greater the value, the more significant the difference between the target pixel point and its neighborhood background; represent the brightness level of the local background, used for normalizing the gray scale difference.

[0039] The greater the value, the greater the difference between the target pixel point and the background, indicating that the target pixel point is more likely to be an interlayer unfused defect region; The smaller the value, the smaller the difference between the target pixel point and the background, indicating that the target pixel point is more likely to be in a flat background region.

[0040] Exemplarily, Figure 2 is a linear structural element rotation state diagram, in which the curved curve represents the local extension direction of the soft connection sheet in the curved area, and the solid point on the curve represents the target pixel point. In the processing process, the linear structural element adjusts the angle according to the extension direction at the target pixel point, rotates the long axis to the vertical state of the extension direction, so that the morphological closing operation can act along the cross-sectional direction of the laminated structure, while suppressing the interference of bending deformation, improving the reliability of extracting the fitting background value at the target pixel point, and improving the detection deviation caused by geometric mismatch of the fixed direction operator.

[0041] S3, based on the lateral distribution characteristics of the pixel points in the vertical direction of the extension direction, calculate the lateral inhibition weight.

[0042] It should be noted that the surface of the soft connection sheet is often distributed with processing scratches and interlayer unfused defects. Both of them appear as edge features with gray scale changes on the original image. Conventional edge detection algorithms cannot distinguish between high-brightness scratches and dark-colored defects, resulting in a high false alarm rate. The interlayer unfused defect shows a wave peak feature in the lateral distribution, with high contrast residual value of the middle pixel point and low contrast residual value of the two side pixel points, while the non-defect edge usually only has obvious changes on one side, showing a step shape. The present application uses the difference in this distribution feature to distinguish between true defects and false edges through lateral inhibition.

[0043] Specifically, a maximum search radius K is set. Starting from the target pixel point, the pixel points in the clockwise perpendicular direction of the extension direction and the counterclockwise perpendicular direction of the extension direction are traversed respectively. Within the maximum search radius range, the pixel point with the maximum gray value in the clockwise perpendicular direction of the extension direction is selected as the background reference point in the clockwise perpendicular direction of the extension direction, and the pixel point with the maximum gray value in the counterclockwise perpendicular direction of the extension direction is selected as the background reference point in the counterclockwise perpendicular direction of the extension direction. The contrast residual values of the two background reference points are obtained. The lateral inhibition weight is calculated according to the difference between the contrast residual values of the target pixel point and the two background reference points.

[0044] In the present embodiment, the preset parameter is used to define the spatial range for finding the background reference point in the lateral inhibition process, and the empirical value range is pixels. In the present embodiment, the parameter is set to 5 pixels. In other embodiments, the implementer can adjust the preset value according to the expected physical width of the actual defect to be detected. For example, when a larger width interlayer unfused defect needs to be detected, the parameter should be increased accordingly to ensure that the real background reference point can be found across the defect area, avoiding detection failure due to insufficient search distance.

[0045] It should be noted that the surface of the soft connection lamination is bright metal, and the defect is a dark gap. Therefore, the maximum gray value on the search path corresponds to the real metal background. Selecting this maximum value as the background reference point can cross the dark defect area and lock the normal metal surface for comparison.

[0046] Specifically, the lateral inhibition weight satisfies the relationship:

[0047] ;

[0048] In the formula, is the lateral inhibition weight of the target pixel point, is the contrast residual value of the target pixel point, is the contrast residual value of the background reference point in the clockwise perpendicular direction of the extension direction of the target pixel point, is the contrast residual value of the background reference point in the counterclockwise perpendicular direction of the extension direction of the target pixel point, is the standard deviation of the Gaussian function. In the present embodiment, the preset parameter is used to adjust the sensitivity of the lateral inhibition weight to the contrast residual difference, and the empirical value range is . In the present embodiment, the parameter is set to 0.5. In other embodiments, the implementer can adjust the preset For example, when there are more false alarm interferences caused by metal reflection in the collected image, the parameter should be increased to increase the suppression threshold of non-defect signals, thereby improving the purity of the detection result.

[0049] wherein, represents the difference degree of the target pixel point and the two side background reference points in the residual intensity. The greater the value, the more significant the intensity difference of the current target pixel point relative to the two side background reference points, and the more likely to present the distribution characteristics of high middle value and low both sides. At this time, tends to 1, indicating that the target pixel point has more wave peak characteristics of interlayer unfused defects, and the target pixel point signal should be retained; The smaller the value, the more similar the current target pixel point and the single side or two side background reference point, and the more likely to present the step edge or flat characteristics. At this time, tends to 0, indicating that the target pixel point is more likely to belong to the normal interlayer edge of the laminated sheet or the flat background area, and does not have defect characteristics, and the target pixel point signal should be suppressed.

[0050] S4, generating a trajectory accumulation path based on the extension direction, and calculating a final defect response value.

[0051] It should be noted that the interlayer unfused defect is a gap sandwiched between the laminated sheets, and the direction changes with the bending of the soft connection laminated sheet. The fixed direction detection window cannot adapt to the direction of the bending laminated sheet, resulting in poor defect detection effect in the curved area. At the same time, the weak defect signal is often disconnected, and it is difficult to form a complete and continuous recognition result. In order to accurately detect the defects in the curved area, the present application needs to make the detection direction follow the actual direction of the laminated sheet in real time, connect the intermittent weak defect signal into a complete line, and filter out the interference of non-defect structures in the signal collection process.

[0052] Specifically, the length of the trajectory accumulation path is determined according to the contrast residual value of the target pixel point; the target pixel point is taken as the center, and the unit pixel step is iterated in both directions along the extension direction and the opposite direction of the extension direction; in the extension process, every step of extension, the next extension position is calculated along the extension direction of the current reached pixel position, and the non-integer coordinates of the position are rounded to the nearest integer coordinates, until the length of the trajectory accumulation path is reached, and a trajectory accumulation path is generated.

[0053] Specifically, the length of the trajectory accumulation path satisfies the relationship:

[0054] ;

[0055] In the formula, is the length of the trajectory accumulation path of the target pixel point, is a basic length constant, is a gain length coefficient, is a contrast residual value of the target pixel point, is a Sigmoid function mapping the input to the interval [0, 1], is a rounding function.

[0056] In the embodiment, the preset parameter is set to 3. In other embodiments, the implementer can adjust the preset according to the length distribution characteristics of the actual defects to be detected. For example, when the physical size of the defects expected to be captured is generally longer, the parameter should be increased accordingly to match the size characteristics of the defects, so as to ensure effective coverage or extraction of long defect targets.

[0057] In the embodiment, the preset parameter is set to 6. In other embodiments, the implementer can adjust the preset according to the aspect ratio or morphological extension characteristics of the actual defects. For example, when defects with significant elongated characteristics need to be screened, the parameter should be increased accordingly to improve the suppression ability of short noise or non-striated interference, thereby improving the detection specificity.

[0058] wherein, The greater the value, the higher the probability of the presence of a real interlayer unfused defect at the pixel point, and the higher the signal confidence. At this time The greater the value, the longer the trajectory accumulation path is extended to capture and connect potential breakpoints distributed along the defect direction, ensuring that weak but continuous defect signals can be completely extracted; The smaller the value, the higher the probability that the target pixel point is background noise or non-defect characteristics. At this time The smaller the value, the shorter the trajectory accumulation path is shortened to avoid incorporating irrelevant background information into the calculation, preventing noise accumulation and false detection.

[0059] Further, the final defect response value satisfies the expression:

[0060] ;

[0061] In the formula, is the final defect response value of the target pixel point, is a contrast residual value of the target pixel point, and k is an index of the trajectory accumulation path length, is the trajectory accumulation path length of the target pixel point, is the contrast residual value at the kth pixel point on the trajectory accumulation path, is the lateral suppression weight at the kth pixel point on the trajectory accumulation path.

[0062] wherein, representing the cumulative intensity of the effective defect signal along the path in the extension direction. The greater the value, the more continuous and significantly different signals exist after the lateral inhibition weight screening along the trajectory cumulative path direction of the target pixel point, that is, the higher the probability that the target pixel point belongs to the interlayer unfused defect. The greater the final defect response value , thereby achieving cascade enhancement of weak intermittent defects; The smaller the value, the weaker the defect signal of the target pixel point along the trajectory cumulative path direction, and the greater the probability that the target pixel point is isolated noise or a non-defect edge. The smaller the final defect response value , thereby suppressing the interference of background noise and non-defect edges.

[0063] S5, based on the final defect response map, the area of the defect region is counted to determine the quality of the soft connection laminates.

[0064] It should be noted that the soft connection laminates often have curved or inclined regions, and direct measurement of the size is easily disturbed by shape changes. In order to obtain reliable soft connection laminate quality determination basis, the present application evaluates the severity of the defect by counting the overall area of the defect region, which can more accurately reflect the size of the defect on the soft connection laminates and provide a robust quality determination standard.

[0065] Specifically, the final defect response values of all pixel points are traversed to obtain a final defect response map, and the final defect response map is binarized using the Otsu algorithm to obtain a binary image. The binary image is analyzed for connected domains, and the total number of pixel points in all connected domains is counted to obtain a total area A. When the total area A is greater than a preset area threshold A1, it is determined that the soft connection laminates are unqualified; otherwise, it is determined that the soft connection laminates are qualified. In this embodiment, A1 is set to 100. In other embodiments, the implementer can adjust the preset A1 according to industry standards, for example, when the industry standards change the accuracy or range requirements of related technical indicators, the value should be adjusted accordingly to ensure consistency with the standard specification.

[0066] Exemplarily, Figure 3 is a soft connection laminate grayscale image, Figure 4 is a soft connection laminate defect binary image, from Figure 4 It can be seen from the above that the high-light connected domain represents the interlayer unfused defect feature of the soft connection side weld, and the pixel coordinates thereof are in a spatial mapping relationship with the physical defect position in the original image; the dark area represents the normal laminate structure background and non-defect interference features that are effectively suppressed after threshold segmentation. Binarization segmentation realizes the separation of defect targets and background interference, converts the weak and fuzzy gray difference in the original image into a determined geometric area, thereby providing data support for subsequent quality determination based on connected domain area statistics.

Claims

1. An image processing-based soft connection lamination side seam weld quality detection method, characterized by, The method comprises the following steps: Collecting a soft connection lamination side edge image, taking any pixel point in the image as a target pixel point, decomposing the structure tensor matrix of the target pixel point, and obtaining the extension direction according to the eigenvector corresponding to the minimum eigenvalue obtained by decomposition; A preset linear structure element is used to perform morphological operation on the linear structure element rotated in the extension direction of the target pixel point to obtain a fitting background value at the target pixel point, and the contrast residual value of the target pixel point is calculated according to the gray value and the fitting background value of the target pixel point; A background reference point of the target pixel point is searched in the vertical direction of the extension direction to calculate the lateral inhibition weight of the target pixel point; The background reference point is the pixel point with the maximum gray value in the clockwise and counterclockwise vertical directions of the extension direction within a preset maximum search radius; the length of the trajectory accumulation path is determined based on the contrast residual value of the target pixel point; the trajectory accumulation path is generated along the extension direction with the length of the accumulation path and taking the target pixel point as the center; and the final defect response value of the target pixel point is obtained according to the contrast residual value and the lateral inhibition weight of the pixel points on the trajectory accumulation path. The final defect response values of all pixel points are combined into a final defect response map and subjected to binarization processing, and the quality of the soft connection lamination side edge weld is determined according to the area of the connected domain after binarization.

2. The method for detecting the quality of the soft connection lamination side weld according to claim 1, characterized in that, The method for decomposing the structure tensor matrix of the target pixel point and obtaining the extension direction according to the eigenvector corresponding to the minimum eigenvalue obtained by decomposition comprises the following steps: the structure tensor matrix of the target pixel point is decomposed by using a second-order matrix eigenvalue decomposition method to obtain eigenvalues and eigenvectors corresponding to the eigenvalues, and the extension direction of the target pixel point is obtained according to the eigenvector corresponding to the minimum eigenvalue of the target pixel point.

3. The method for detecting the quality of the side weld of the soft connection sheet metal based on image processing according to claim 1, characterized in that, The contrast residual value satisfies the expression: ； In the formula, is the contrast residual value of the target pixel point, is the original image gray value of the target pixel point, is the fitting background value of the target pixel point, is a constant to prevent the denominator from being zero.

4. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The lateral inhibition weight satisfies the expression: ； In the formula, is a lateral inhibition weight of the target pixel point, is a contrast residual value of the target pixel point, is a contrast residual value of a background reference point in a clockwise vertical direction of the target pixel point along the extension direction, is a contrast residual value of a background reference point in an anticlockwise vertical direction of the target pixel point along the extension direction, is a standard deviation of a Gaussian function.

5. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The length of the trajectory accumulation path satisfies the expression: ； wherein, is a length of a path of a trajectory of a target pixel point, is a base length constant, is a gain length coefficient, is a contrast residual value of a target pixel point, is a Sigmoid function that maps an input to the interval 0 to 1, is a rounding function.

6. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The method for performing morphological operation on the linear structure element rotated in the extension direction of the target pixel point to obtain the fitting background value at the target pixel point comprises the following steps: a rectangular linear structure element with an aspect ratio of 2:1 is defined with the target pixel point as the center, the long side of the linear structure element is defined as the major axis, the major axis is rotated to be perpendicular to the extension direction, and morphological closing operation is performed on the original image by using the rotated linear structure element to obtain the fitting background value at the target pixel point.

7. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The final defect response value satisfies the expression: ； In the formula, is the final defect response value of the target pixel point, is the contrast residual value of the target pixel point, and k is an index of the track accumulation path length, is the track accumulation path length of the target pixel point, is the contrast residual value at the kth pixel point on the track accumulation path, is the lateral inhibition weight at the kth pixel point on the track accumulation path.

8. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The method for obtaining the background reference point comprises the following steps: starting from the target pixel point, the clockwise vertical direction of the extension direction and the counterclockwise vertical direction of the extension direction are traversed pixel by pixel, and within the maximum search radius, the pixel point with the maximum gray value in the clockwise vertical direction of the extension direction is selected as the background reference point in the clockwise vertical direction of the extension direction, and the pixel point with the maximum gray value in the counterclockwise vertical direction of the extension direction is selected as the background reference point in the counterclockwise vertical direction of the extension direction.

9. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The method for combining the final defect response values of all pixel points into a final defect response map and performing binarization processing comprises the following steps: the final defect response values of all pixel points are traversed to obtain a final defect response map, and the final defect response map is subjected to binarization processing by using the Otsu algorithm to obtain a binary image.

10. The method for detecting the quality of the side seam weld of the soft connection lamination based on image processing according to claim 1, characterized in that, The method for generating a trajectory accumulation path along an extension direction with an accumulation path length comprises: iteratively extending in a unit pixel step along the extension direction and the opposite direction of the extension direction of a target pixel point; during the extending, each time the extending is performed by one step, the extending position of the next step is calculated along the extension direction of the current reached pixel position, and the non-integer coordinates of the position are rounded to the nearest integer coordinates until the trajectory accumulation path length is reached, thereby generating a trajectory accumulation path.

Images