Machine vision based leather surface defect detection system

Abstract

This invention relates to the field of leather production and processing technology, and discloses a leather surface defect detection system based on machine vision, comprising: a data acquisition terminal, the data acquisition terminal including a housing with an open lower end, the housing being used to form a closed darkroom environment with the leather surface during detection; a light source and a camera are provided at the top of the housing; the image acquired by the camera defines mutually perpendicular X-axis and Y-axis, wherein the X-axis is parallel to the traction direction of the leather production line; and an image analysis module, including: an image preprocessing unit, used to enhance the acquired image, making the outline of each closed particle formed by the texture clear and free of rough edges. This invention transforms blurred visual deformation into calculable and comparable mathematical indicators by extracting the density parameters and direction intersection ratio parameters of the texture particles in the image.

Description

Technical Field

[0001] This invention relates to the field of leather production and processing technology, and more specifically to a leather surface defect detection system based on machine vision. Background Technology

[0002] During leather embossing, factors such as uneven traction tension, different roller speeds, or abnormal substrate extensibility often cause tensile deformation of embossed patterns like lychee grain on the leather surface. This manifests as reduced particle density per unit area and elongation of the texture along the traction direction. Traditional manual visual inspection methods are inefficient and highly subjective, making real-time full inspection on continuous production lines difficult. Existing machine vision inspection solutions mostly rely on static image comparison or simple texture feature extraction, which can only identify significant defects such as scratches and holes. For structural deformation defects like "elongated texture," which require quantitative characterization, there is a lack of effective feature extraction and discrimination methods.

[0003] Furthermore, existing detection systems typically only make isolated judgments on currently acquired images, failing to utilize continuous detection data over time to monitor the drift trend of process conditions. When process parameters such as tension gradually deviate from the optimal range, the texture density may exhibit a slow, monotonically decreasing trend or increased fluctuations. Traditional static thresholding methods struggle to capture such early abnormal signals in a timely manner, often only identifying them after defects have exceeded specification limits and generated a large number of scraps. Summary of the Invention

[0004] The purpose of this invention is to provide a machine vision-based leather surface defect detection system to solve the above-mentioned technical problems:

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A machine vision-based leather surface defect detection system includes:

[0007] The acquisition terminal includes a housing with an opening at the bottom, which forms a closed darkroom environment with the leather surface during inspection; a light source and a camera are provided at the top inside the housing; the image acquired by the camera defines mutually perpendicular X-axis and Y-axis, wherein the X-axis is parallel to the traction direction of the leather production line;

[0008] Image analysis module, including:

[0009] The image preprocessing unit is used to enhance the acquired image, making the outline of each closed particle formed by texture clear and free of burrs.

[0010] The feature quantization unit is used to calculate the first parameter and the second parameter of the preprocessed image; the first parameter represents the density of particles per unit area; the second parameter represents the deformation ratio of particles in the image in the X-axis and Y-axis directions.

[0011] The analysis unit has a pre-stored reference parameter range obtained from the statistics of multiple qualified sample images. The analysis unit compares the first parameter of the current image with the reference range. If it exceeds the preset deviation, it is determined to be a defective area. For the image determined to be a defective area, the second parameter is compared with a preset deformation ratio threshold range to identify the deformation direction of the leather.

[0012] The execution module is used to output corresponding control signals and warning signals based on the judgment results of the analysis unit.

[0013] As a further technical solution, the housing and the leather surface are non-contact; the lower edge of the housing is provided with a flexible light-blocking skirt, which dynamically contacts the leather surface during detection.

[0014] As a further technical solution, the image preprocessing unit includes the following sub-units:

[0015] The grayscale subunit is used to convert the acquired color image into a grayscale image;

[0016] The filtering and denoising subunit is used to smooth grayscale images using Gaussian filtering or median filtering.

[0017] A contrast enhancement subunit is used to enhance the local contrast of an image using an adaptive histogram equalization algorithm.

[0018] Binarization subunits are used to convert enhanced images into black-and-white binary images using Otsu's method or an adaptive thresholding algorithm.

[0019] The morphological closing operation subunit is used to process binary images using a dilation-erosion morphological closing operation to fill pores inside particles and connect breaks on the contour.

[0020] As a further technical solution, the process by which the feature quantization unit obtains the first parameter includes:

[0021] Connectivity analysis is performed on the preprocessed binary image to identify and count the number of all independent closed particles in the image, denoted as . ;

[0022] Calculate the effective area of ​​the image, denoted as . ,in It equals the product of the image's width and height in pixels; calculate the first parameter. : .

[0023] As a further technical solution, the process by which the feature quantization unit obtains the second parameter includes:

[0024] Generate on the preprocessed binary image A first virtual scan line, evenly spaced and parallel to the X-axis, and A second virtual scan line, equally spaced and parallel to the Y-axis, wherein... It is a preset positive integer;

[0025] For each first virtual scan line, count the number of intersections between it and all particle contours in the image, and sum the number of intersections of all first virtual scan lines to obtain the total number of first intersections. ;in For the first The number of intersections of the first virtual scan lines;

[0026] For each second virtual scan line, count the number of intersections between it and all particle contours in the image, and sum the number of intersections of all second virtual scan lines to obtain the total number of second intersections. ;in For the first The number of intersections of the second virtual scan lines;

[0027] Calculate the second parameter: .

[0028] As a further technical solution, the range of reference parameters pre-stored in the analysis unit is obtained in the following way:

[0029] collection For each of the qualified sample images, calculate its first parameter. ,in , This is the preset sample size;

[0030] calculate Mean of a sample ; then calculate Standard deviation of each sample ;

[0031] by With the reference center, The baseline dispersion is used.

[0032] As a further technical solution, the analysis unit will use the first parameter of the current image. The process of comparing with the benchmark range is as follows:

[0033] Calculate the degree of deviation: ;

[0034] Preset warning coefficient and action coefficient ,in ;

[0035] like This is considered normal.

[0036] like This is considered a warning and marked as a candidate defect area;

[0037] like This area was determined to be a defective area.

[0038] As a further technical solution, the analysis unit uses a statistical process control model to dynamically monitor the sequence changes of the first parameter, specifically including:

[0039] Let the sequence number of the continuously acquired image frames be... The corresponding first parameter is ;

[0040] Calculate the moving range ,in ;

[0041] Based on the initial stage For a sample of qualified samples, calculate the mean of the moving range:

[0042] ;

[0043] Calculate individual control charts: centerline Upper control limit ;lower control limit ;in This is a preset constant;

[0044] Perform the following determination process:

[0045] like or This area was determined to be a defective area;

[0046] If Q out of a consecutive P points exceed [the limit] Issue an early warning. Preset early warning coefficient;

[0047] If all T consecutive points are greater than or all less than Issue a warning;

[0048] If U consecutive points show a monotonically increasing or monotonically decreasing trend, an early warning will be issued; where P, Q, T, and U are all preset values.

[0049] As a further technical solution, the specific method by which the analysis unit identifies the direction of leather deformation is as follows:

[0050] Preset deformation ratio threshold range ,in

[0051] Obtain the second parameter of the defect area image ;

[0052] like It is determined to be a lateral tensile deformation; if It is determined to be longitudinal tensile deformation; if It was determined that there was no significant deformation.

[0053] The execution module outputs prompt information related to the production process parameters based on the identified deformation direction: when it is determined to be transverse tensile deformation, it outputs a prompt of abnormal traction tension; when it is determined to be longitudinal tensile deformation, it outputs a prompt of uneven substrate elongation or difference in pressure roller speed.

[0054] The beneficial effects of this invention are:

[0055] (1) This invention extracts the density parameter and the direction intersection ratio parameter of the texture particles in the image, transforming the blurred visual deformation into a calculable and comparable mathematical index. The density parameter reflects the number of particles per unit area and is used to determine whether the texture is becoming sparse overall. The direction intersection ratio quantitatively describes the deformation ratio of the texture in the traction direction and its perpendicular direction by statistically analyzing the number of intersections with the particle outline in the horizontal and vertical directions. Thus, the system can accurately identify whether the texture has undergone stretching deformation and provide a quantified degree of deviation.

[0056] (2) This invention automatically determines whether the deformation is transverse or longitudinal tension by analyzing the numerical range of the intersection ratio. When the intersection ratio is less than a preset lower limit, it is determined to be transverse tensile deformation, corresponding to abnormal traction tension; when the intersection ratio is greater than a preset upper limit, it is determined to be longitudinal tensile deformation, corresponding to uneven substrate elongation or poor roller speed. The detection system outputs prompt information related to specific process parameters, enabling operators to quickly locate the root cause of the fault.

[0057] (3) This invention introduces a statistical process control model to dynamically monitor continuously acquired image frame sequences. By calculating the moving range and establishing a control chart, the system can identify various abnormal patterns such as single value exceeding the limit, multiple consecutive points exceeding the warning line, multiple consecutive points falling on the same side of the center line, and monotonically increasing or decreasing trends. When the process parameters gradually deviate from the optimal range, the system can issue an early warning signal before the defects exceed the specification limits, changing the traditional post-process quality inspection to in-process control, which can effectively reduce the generation of scrap. Attached Figure Description

[0058] The invention will now be further described with reference to the accompanying drawings.

[0059] Figure 1 This is a system framework diagram of the present invention;

[0060] Figure 2 This is a perspective view of the acquisition terminal in the system of the present invention;

[0061] Figure 3 This is a partial schematic diagram of the image acquired by the acquisition terminal in this invention;

[0062] Reference numerals: 1. Housing; 2. Light source; 3. Light-shielding skirt; 4. Camera. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 As shown, a machine vision-based leather surface defect detection system includes:

[0065] Data acquisition terminal, such as Figure 2 As shown, the device includes a housing 1 with an opening at the bottom. The housing 1 is cylindrical or other geometrically shaped, with a hollow interior and an open bottom. During inspection, the housing 1 is placed above the leather production line, with its lower opening facing the leather surface. The housing 1 and the leather surface together form a closed darkroom environment. Specifically, "closed" means that there is no transparent light path between the interior space of the housing 1 and the external environment; ambient light cannot directly enter the interior of the housing 1, thus providing stable illumination conditions for image acquisition. A light source 2 and a camera 4 are located at the top inside the housing 1. The light source 2 provides illumination to the leather surface, and the camera 4 captures images of the leather surface.

[0066] Camera 4 is fixed at the center of the top plate inside housing 1, with its lens pointing downwards towards the leather surface. Light source 2 is arranged around camera 4 or positioned beside it. In the image captured by camera 4, mutually perpendicular X-axis and Y-axis are defined. Specifically, the horizontal direction of the image is the X-axis, and the vertical direction is the Y-axis. The X-axis is set parallel to the traction direction of the leather production line. The traction direction refers to the direction in which the rollers or conveyor belts driving the leather movement on the production line move the leather. Aligning the X-axis of the image with the traction direction ensures that subsequent feature analysis can reflect the influence of traction motion on texture deformation.

[0067] The housing 1 is non-contact with the leather surface; the lower edge of the housing 1 is provided with a flexible light-blocking skirt 3, which dynamically contacts the leather surface during detection; the material of the light-blocking skirt 3 is selected from one or more of the following: black silicone bristles, black soft rubber sheet, black sponge strip, and black flocked bristle brush.

[0068] The image analysis module specifically includes the following units:

[0069] The image preprocessing unit enhances the acquired images. The purpose of this processing is to clarify the outlines of each closed grain formed by the texture in the image, eliminating burrs. Specifically, closed grains refer to the raised areas with closed geometric shapes surrounded by grooves within the lychee pattern or other floral designs formed on the leather surface after embossing. Eliminating burrs specifically means that the edges of each grain are continuous and smooth, without jagged breaks or fine burr-like protrusions.

[0070] The feature quantization unit is used to calculate the first parameter and the second parameter of the preprocessed image; the first parameter represents the density of particles per unit area; the second parameter represents the deformation ratio of particles in the image in the X-axis and Y-axis directions.

[0071] The analysis unit has a pre-stored reference parameter range obtained from the statistics of multiple qualified sample images. The analysis unit compares the first parameter of the current image with the reference range. If it exceeds the preset deviation, it is determined to be a defective area. For the image determined to be a defective area, the second parameter is compared with a preset deformation ratio threshold range to identify the deformation direction of the leather.

[0072] The execution module is used to output corresponding control signals or early warning signals based on the judgment results of the analysis unit. The execution module may include an audible and visual alarm device, a marking device, a display device, or a communication interface connected to the production control system. For example, when the judgment result is normal, the execution module does not output an alarm signal or outputs a "normal" indication signal; when the judgment result is a defective area, the execution module outputs an alarm signal and can output specific prompt information based on the direction of deformation.

[0073] Through the above technical solution, this embodiment provides a machine vision-based leather surface defect detection system. Specifically, on the production line, the housing 1 of the acquisition terminal is fixedly installed at the detection station behind the embossing roller. The distance between the lower opening of the housing 1 and the leather surface is set to 5 mm to 10 mm. A black light-blocking skirt 3 is installed on the bottom edge of the housing 1, which contacts the leather surface to form a darkroom environment. The ring-shaped LED light source 2 at the top inside the housing 1 illuminates the leather surface at a 45-degree angle, and the camera 4 acquires images at a frame rate of 30 to 60 frames per second. The acquired images first enter the image preprocessing unit. The preprocessing unit converts the color image into a grayscale image, uses Gaussian filtering to remove noise, uses adaptive histogram equalization to enhance contrast, uses Otsu's method for binarization segmentation, and finally uses morphological closing operation to close the particle contours. After processing, each lychee grain particle in the image appears as an independent white closed area with a smooth and continuous contour. The feature quantization unit performs calculations on the preprocessed image. Suppose that N=1250 closed particles are identified in the current image, and the image size is W=2448 pixels wide and H=2048 pixels high. Then the image area is S=2448×2048≈5,013,504 pixels squared, and the first parameter D=1250 / 5013504≈0.000249 (unit: particles / pixel). If the pixel area is converted to physical area (a single pixel corresponds to 0.03mm×0.03mm), then Sphysical≈4512 square millimeters, Dphysical≈0.277 particles / square millimeter. The feature quantization unit also calculates the second parameter. Suppose that M=10 equally spaced horizontal scan lines and 10 equally spaced vertical scan lines are generated, and the total number of intersections of the horizontal scan lines is A=320, and the total number of intersections of the vertical scan lines is B=305. Then the second parameter R=320 / 305≈1.05. The analysis unit compares the first parameter D=0.000249 with the benchmark range. Assuming the benchmark mean is 0.000250 and the standard deviation is 0.000008, the deviation Z=|0.000249-0.000250| / 0.000008=0.125. If the preset warning coefficient is 2 and the action coefficient is 3, then Z=0.125 is less than 2, and is judged as normal. If Z is greater than 2, a warning is triggered or the area is judged as defective. For areas judged as defective, the analysis unit further compares the second parameter. Assuming the preset deformation ratio threshold range is [0.9, 1.1], the current R=1.05 falls within this range, and is judged as having no obvious deformation direction. If R=0.85, it is judged as lateral tensile deformation; if R=1.25, it is judged as longitudinal tensile deformation. The execution module outputs a signal based on the judgment result. If the condition is determined to be normal, no alarm will be output; if the condition is determined to be lateral tensile deformation, the execution module will control the marking device to spray a mark at the corresponding position on the edge of the leather and display the "abnormal traction tension" prompt message on the operation panel.

[0074] It should be noted that the specific values ​​in the above embodiments (such as the gap of housing 1 5-10mm, camera frame rate 30-60fps, image size 2448×2048, number of scan lines M=10, deformation ratio threshold [0.9,1.1], etc.) are all exemplary values. In practical applications, those skilled in the art can reasonably select and adjust these parameters according to factors such as specific leather type, production line speed, and detection accuracy requirements. For example, for leather with finer texture, the number of scan lines can be increased to obtain more accurate statistical results; for production lines with high detection speed requirements, the number of scan lines can be reduced to speed up the calculation.

[0075] The technical solution provided in this embodiment differs from existing technologies in that existing technologies typically only perform simple texture analysis on images or compare them with standard templates, failing to quantitatively characterize the structural deformation of "texture stretching." In contrast, this invention defines mutually perpendicular coordinate axes and aligns them with the traction direction. It quantifies particle density by calculating a first parameter and deformance ratio by calculating a second parameter, thereby achieving quantitative detection of embossing stretching deformation. Furthermore, the detection results are used to identify the deformation direction and correlate it with process parameters.

[0076] The image preprocessing unit includes the following sub-units:

[0077] The grayscale subunit is used to convert the color image captured by camera 4 into a grayscale image. Color images typically store the color information of each pixel in three RGB channels, resulting in a large amount of data that is irrelevant to texture and shape analysis. Grayscale processing converts this data into single-channel brightness information to reduce subsequent computation.

[0078] The filtering and denoising subunit is used to smooth grayscale images to eliminate sensor noise and environmental interference introduced during image acquisition. It can be implemented using algorithms such as Gaussian filtering or median filtering.

[0079] The contrast enhancement subunit is used to enhance the local contrast of the image, making the brightness differences between grains and grooves in the leather texture more pronounced. This can be achieved using an adaptive histogram equalization algorithm.

[0080] The binarization subunit is used to convert the enhanced grayscale image into a black-and-white binary image. The binarization process compares the grayscale value of each pixel with a preset or adaptively calculated threshold. Pixels with a grayscale value greater than the threshold are assigned a first value (e.g., 1, representing white). Figure 3 In medium-grained regions, pixels with values ​​less than or equal to the threshold are assigned a second value (e.g., 0, representing black). Figure 3 In the mid-gully region, the optimal segmentation threshold can be automatically calculated using the Otsu method or an adaptive thresholding algorithm.

[0081] The morphological closing subunit is used to perform morphological processing on binary images by first dilating and then eroding, in order to fill the tiny pores inside the particles and connect the fine breaks on the particle outline, so that each particle forms a complete and closed region.

[0082] It should be noted that the image processing algorithms used in the above sub-units, including grayscale conversion, Gaussian filtering, median filtering, adaptive histogram equalization, Otsu's binarization, adaptive threshold binarization, morphological dilation, morphological erosion, and morphological closing operations, are all well-known and mature existing technologies in the field of digital image processing. The specific implementation steps, parameter settings, mathematical expressions, and programming methods of these algorithms have been described in detail in numerous textbooks, technical manuals, and academic literature in this field. Those skilled in the art can select and implement them according to specific application scenarios without creative effort; therefore, this specification will not elaborate on the specific algorithm details.

[0083] The process of the feature quantization unit obtaining the first parameter includes:

[0084] Connectivity analysis is performed on the preprocessed binary image to identify and count the number of all independent closed particles in the image, denoted as . The specific process of connected component analysis includes: First, scanning the binary image line by line, when encountering the first pixel with a value of 1 that has not yet been labeled, taking it as the starting point of a new connected region; Second, starting from this starting point, using a seed-fill algorithm or a scan-line-fill algorithm, traversing all connected pixels with a value of 1, and assigning these pixels the same region label; Third, repeating steps one and two until all pixels with a value of 1 in the image are labeled; After completing the above steps, each independent closed particle in the image corresponds to a unique region label. The total number of particles is obtained by counting the number of different region labels. It should be noted that particles located at the image edge may be cut off by the image boundary, forming incomplete particles. For such particles, they can be selectively included or excluded from the statistics. This implementation adopts an exclusion strategy, that is, only particles that are completely located within the image are counted to eliminate the influence of boundary effects on density calculation. The implementation method is as follows: during connected component analysis, if a connected region contains pixels located on the image boundary, it is marked as an invalid region and not included in the count. .

[0085] Calculate the effective area of ​​the image, denoted as . ,in It equals the product of the image's width and height in pixels; for example, if the image captured by the camera has a resolution of 2448 pixels (width) × 2048 pixels (height), then S = 2448 × 2048 = 5,013,504 (pixels squared). It should be noted that the unit of calculation for S is consistent with that for N. N is a dimensionless count value, while S is in pixels squared; therefore, the unit of the first parameter D is "pixels / pixel squared". If it is necessary to convert D to a physically meaningful density value (such as "pixels / square millimeter") later, it is necessary to perform the conversion in conjunction with the camera's calibration parameters. S can refer to pixel area or physical area.

[0086] Calculate the first parameter : This ratio reflects the number of closed particles per unit area. For the same type of leather processed by the same embossing roller, under normal process conditions, the D value should remain stable within a relatively narrow range. When the leather is subjected to abnormal stretching during traction, the particles are elongated in the traction direction, resulting in a decrease in the number of particles per unit area, and the D value decreases accordingly. Conversely, if the leather is compressed perpendicular to the traction direction (which is relatively rare in actual production), the D value may increase.

[0087] The process by which the feature quantization unit obtains the second parameter includes:

[0088] Generate on the preprocessed binary image A first virtual scan line, evenly spaced and parallel to the X-axis, and A second virtual scan line, equally spaced and parallel to the Y-axis, wherein... The value of M is a preset positive integer. The value of M determines the sampling density: the larger the value of M, the denser the sampling and the higher the statistical accuracy of the second parameter, but the computational load also increases accordingly; the smaller the value of M, the faster the computation speed, but detailed information may be lost. Those skilled in the art can choose an appropriate value of M according to the image size and detection speed requirements.

[0089] For each first virtual scan line, count the number of intersections between it and all particle contours in the image, and sum the number of intersections of all first virtual scan lines to obtain the total number of first intersections. ;in For the first The number of intersections of the first virtual scan lines; the A value reflects the particle boundary density of the image in the horizontal direction. When the texture is stretched in the X-axis direction (traction direction), the frequency of boundary changes in the horizontal direction decreases, and the A value decreases accordingly.

[0090] For each second virtual scan line, count the number of intersections between it and all particle contours in the image, and sum the number of intersections of all second virtual scan lines to obtain the total number of second intersections. ;in For the first The number of intersections of the second virtual scan lines reflects the particle boundary density of the image in the vertical direction. When the texture is stretched in the X-axis direction, the frequency of boundary changes in the vertical direction increases, and the B value increases accordingly.

[0091] It should be noted that, as Figure 3 As shown, the groove region is black, and the grain region is white. An intersection point is counted only when the scan line enters the grain region from the groove region or leaves the grain region to the boundary of the groove region. When the scan line coincides with the contour line (collinear), no new crossing occurs, and therefore it is not counted as an intersection point. In other words, the specific method for counting intersection points is as follows: for each pixel on the scan line, if the pixel belongs to the contour boundary of the grain region, and two adjacent pixels along the direction perpendicular to the scan line belong to the grain region and the groove region respectively, then the pixel is counted as an intersection point; if two adjacent pixels along the direction perpendicular to the scan line belong to either the grain region or the groove region, then they are not counted as an intersection point.

[0092] Calculate the second parameter: R represents the ratio of particle deformation in the X-axis to the Y-axis in an image. This ratio eliminates the influence of the overall texture density of the image on the absolute number of intersections, making the R value depend only on the degree of anisotropy of the texture.

[0093] The range of pre-stored baseline parameters in the analysis unit is obtained in the following way:

[0094] collection Images of qualified samples, for example, are collected using the same set of acquisition terminals in a laboratory environment to acquire images of standard samples, and the first parameter is calculated for each image. ,in , This is the preset sample size;

[0095] calculate Mean of a sample ; then calculate Standard deviation of each sample ;

[0096] by With the reference center, The baseline dispersion is used.

[0097] The analysis unit will use the first parameter of the current image The process of comparing with the benchmark range is as follows:

[0098] Calculate the degree of deviation: ;

[0099] Preset warning coefficient and action coefficient ,in ;

[0100] When the deviation Z is less than or equal to the warning coefficient At that time, it indicates the first parameter of the current image. Compared with the benchmark mean The differences are within the normal fluctuation range. Even and They are not completely equal, and the difference between them can be attributed to random fluctuations under normal production conditions, rather than process abnormalities or product defects.

[0101] When the deviation Z is greater than the warning coefficient But less than or equal to the action coefficient At that time, it indicates the first parameter of the current image. The image has deviated from the normal fluctuation range, but has not yet reached the level of being directly judged as a defect. This state usually indicates that the production process parameters are drifting (such as a slow increase in traction tension leading to a gradual decrease in density), or that there is a slight quality abnormality in the area. In this case, the analysis unit outputs a "warning" judgment result and marks the leather area corresponding to the current image as a candidate defect area. The execution module issues a warning signal (such as a flashing yellow light), but does not trigger a shutdown or marking. The information of the candidate defect area is recorded in the inspection log for subsequent review by operators or as a basis for process adjustments;

[0102] When the deviation Z is greater than the action coefficient At that time, it indicates the first parameter of the current image. The deviation from the baseline average is significant, exceeding the range possible under normal production conditions. This state typically indicates a clear defect in the area, such as severely elongated texture, insufficient embossing depth, or missing particles. In this case, the analysis unit outputs a "defective area" determination. The execution module triggers a level one alarm signal (e.g., flashing red light and audible alarm), controlling the marking device to mark the corresponding physical location on the leather and recording the information of this area in the detection database for subsequent cutting and layout considerations or as a basis for rejecting defective products.

[0103] The analysis unit uses a statistical process control model to dynamically monitor the sequence changes of the first parameter, specifically including:

[0104] Let the sequence number of the continuously acquired image frames be... The corresponding first parameter is Each frame corresponds to a physical location on the leather, and the direction of the frame number increment is consistent with the direction of the leather's traction movement.

[0105] Calculate the moving range ,in The moving range refers to the absolute value of the difference between the first parameters of two adjacent frames, and is used to measure the fluctuation range between frames. The smaller the value, the smaller the density change between two adjacent frames, and the more stable the production process. A larger value indicates a sudden change in density, which may indicate local defects or process disturbances.

[0106] During system initialization and periodic calibration, baseline parameters for control charts need to be established based on qualified samples. This is based on the initial phase... For a sample of qualified samples, calculate the mean of the moving range:

[0107] ;

[0108] Calculate a single-value control chart, which is used to monitor the first parameter. Whether the value is under statistical control. This control chart contains three horizontal lines: the center line... , representing the target value under normal conditions; upper control limit The upper limit indicates the normal fluctuation range; exceeding this limit indicates that the process is out of control; the lower control limit... This represents the lower limit of normal fluctuations; values ​​below this limit indicate that the process is out of control. It is a constant related to the subgroup size. In moving range control, since each calculation of the moving range uses two adjacent points (subgroup size n=2),... The estimate used to convert the moving range into the standard deviation, specifically, when the subgroup size n=2, the estimate of the standard deviation σ is... ,in =1.128, and .in, It is a constant used in statistical process control (SPC) to estimate the population standard deviation.

[0109] Perform the following determination process:

[0110] like or This indicates that the density value of the current frame has deviated significantly from the normal range. This deviation is usually sudden and may be caused by local defects (such as severe elongation of the texture or missing embossing in the area) or instantaneous process disturbances (such as sudden fluctuations in tension), and is identified as a defective area;

[0111] If there are Q points out of a consecutive P points Exceeding Issue an early warning. This rule is used to identify deviation events that occur frequently within a short period of time, based on a preset warning coefficient.

[0112] If all T consecutive points are greater than or all less than Issue an early warning, for example, if 7 consecutive The values ​​are all greater than This indicates a persistently high density, which may indicate excessive embossing pressure or insufficient traction tension; if this continues for 7 consecutive days... The values ​​are all less than This indicates that the density remains low, and there may be a trend of gradually increasing traction tension.

[0113] If U consecutive points show a monotonically increasing or monotonically decreasing trend, an early warning is issued. For example, if the traction tension gradually increases over time, It will show a monotonically decreasing trend; if the temperature of the embossing roller gradually increases, It may show a monotonically increasing trend.

[0114] P, Q, T, and U are all preset values ​​that can be manually set by the operator according to actual production conditions and testing requirements. Specifically, the operator inputs the values ​​of P, Q, T, and U through the human-machine interface of the testing system. The system stores these values ​​in a configuration file and executes the judgment rules according to the set values ​​during subsequent testing. The typical range of P is 3 to 10, Q is 2 to P, T is 5 to 9, and U is 5 to 8. The operator can adjust these values ​​according to the stringency of product quality requirements: for high-end products, smaller P, Q, T, and U values ​​can be set to improve detection sensitivity; for ordinary products, larger P, Q, T, and U values ​​can be set to reduce the false alarm rate. The system also provides a "restore default values" function so that the operator can quickly restore the system's recommended parameter configuration.

[0115] The specific method by which the analysis unit identifies the direction of leather deformation is as follows:

[0116] Preset deformation ratio threshold range ,in . and The value of R is determined based on the specific leather type, embossing pattern, and production requirements. For circular or near-circular pebbled patterns, the R value should be close to 1 under normal conditions; therefore, the threshold range can be set in a symmetrical form about 1, for example... =0.9, =1.1. For textures that are anisotropic (such as brushed texture or bark texture), the R value under normal conditions may deviate from 1. In this case, the threshold interval should be set according to the statistical distribution of normal samples. For example, the interval boundary can be determined by using the mean of the R value of normal samples as the center and the standard deviation of the preset multiple.

[0117] Obtain the second parameter of the defect area image ;

[0118] like This indicates that A (the total number of the first intersection points, reflecting the boundary density in the horizontal direction) is too small relative to B (the total number of the second intersection points, reflecting the boundary density in the vertical direction). That is, the particle boundary density in the horizontal direction is lower than that in the vertical direction, which is determined to be longitudinal tensile deformation.

[0119] like This indicates that A is too large relative to B, meaning the particle boundary density in the horizontal direction is higher than in the vertical direction. The particles are elongated in the Y-axis direction (perpendicular to the traction direction), or more commonly, compressed in the X-axis direction. Longitudinal tensile deformation is typically caused by uneven stretching of the substrate in the width direction, unbalanced pressure at both ends of the embossing roller, or malfunction of the flattening roller, leading to wrinkles or stretching of the leather in the width direction. If... If the first parameter D is judged to be abnormal, the defect is not due to tensile deformation, but may be due to other factors, such as insufficient texture depth caused by wear of the embossing roller, unclear texture caused by batch difference of raw materials, or local contamination.

[0120] The execution module outputs prompts related to production process parameters based on the identified deformation direction: when transverse tensile deformation is detected, an abnormal traction tension prompt is output; when longitudinal tensile deformation is detected, an uneven substrate elongation or differential roller speed prompt is output. The specific output methods for these prompts are as follows:

[0121] When the system determines that the deformation is lateral tensile, the execution module may output one or more of the following prompts: Abnormal traction tension (the front tension may be too high; it is recommended to reduce the pneumatic pressure of the traction roller); Abnormal traction speed (it is recommended to check whether the traction motor speed setting matches the actual speed); Lateral tensile deformation (low density, the texture is stretched; please check the traction system).

[0122] When the system determines that the deformation is longitudinal tensile, the execution module may output one or more of the following prompts: Uneven stretching of the substrate (it is recommended to check the thickness uniformity of the leather in the width direction); Uneven pressure at both ends of the pressure roller (it is recommended to check the pressure settings on the left and right sides of the embossing roller); Abnormal operation of the flattening roller (it is recommended to check whether the flattening roller rotates flexibly and whether there is any wear on the surface); Longitudinal tensile deformation (the texture is stretched in the vertical direction, please check the balance of the pressure roller).

[0123] When the system determines that there is no significant deformation, the execution module may output one or more of the following prompts: Abnormal density but no deformation (it is recommended to check the temperature or pressure of the embossing roller); Abnormal density but no deformation (it is recommended to check the batch consistency of raw materials); Abnormal density but no deformation (please manually check whether there is contamination or insufficient embossing depth in this area).

[0124] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A machine vision based leather surface defect detection system characterized in that, include: The acquisition terminal includes a housing (1) with an opening at the bottom, which is used to form a closed darkroom environment with the leather surface during detection; a light source (2) and a camera (4) are provided at the top inside the housing (1); the image acquired by the camera (4) has mutually perpendicular X-axis and Y-axis defined, wherein the X-axis is parallel to the traction direction of the leather production line; Image analysis module, including: The image preprocessing unit is used to enhance the acquired image, making the outline of each closed particle formed by texture clear and free of burrs. The feature quantization unit is used to calculate the first parameter and the second parameter of the preprocessed image; the first parameter represents the density of particles per unit area; the second parameter represents the deformation ratio of particles in the image in the X-axis and Y-axis directions. The analysis unit has a pre-stored reference parameter range obtained from the statistics of multiple qualified sample images. The analysis unit compares the first parameter of the current image with the reference range. If it exceeds the preset deviation, it is determined to be a defective area. For the image determined to be a defective area, the second parameter is compared with a preset deformation ratio threshold range to identify the deformation direction of the leather. The execution module is used to output corresponding control signals and warning signals based on the judgment results of the analysis unit.

2. The machine vision based leather surface defect detection system as claimed in claim 1, wherein, The housing (1) is non-contact with the leather surface; the lower edge of the housing (1) is provided with a flexible light-blocking skirt (3), which dynamically contacts the leather surface during detection.

3. The machine vision based leather surface defect detection system as claimed in claim 2, wherein, The image preprocessing unit includes the following sub-units: The grayscale subunit is used to convert the acquired color image into a grayscale image; The filtering and denoising subunit is used to smooth grayscale images using Gaussian filtering or median filtering. A contrast enhancement subunit is used to enhance the local contrast of an image using an adaptive histogram equalization algorithm. Binarization subunits are used to convert enhanced images into black-and-white binary images using Otsu's method or an adaptive thresholding algorithm. The morphological closing operation subunit is used to process binary images using a dilation-erosion morphological closing operation to fill pores inside particles and connect breaks on the contour.

4. The machine vision based leather surface defect detection system as claimed in claim 3, wherein, The process by which the feature quantization unit obtains the first parameter includes: The pre-processed binary image is subjected to connected domain analysis to identify and count the number of all independent closed particles in the image, denoted as ; the effective area of the image is calculated, denoted as wherein is equal to the product of the width pixel value and the height pixel value of the image; a first parameter : is calculated.

5. The machine vision based leather surface defect detection system as claimed in claim 4, wherein, The process by which the feature quantization unit obtains the second parameter includes: generating on the pre-processed binary image a first plurality of equally spaced, virtual scan lines parallel to the X axis, and a second plurality of equally spaced, virtual scan lines parallel to the Y axis, wherein is a predetermined positive integer; For each first virtual scanning line, count the number of intersections with all particle contours in the image, sum the number of intersections of all first virtual scanning lines to obtain a first total number of intersections ; wherein is the number of intersections of the first virtual scanning line; and the first virtual scanning line is the first. For each second virtual scan line, count the number of intersections between it and all particle contours in the image, and sum the number of intersections of all second virtual scan lines to obtain the total number of second intersections. ;in For the first The number of intersections of the second virtual scan lines; Calculate the second parameter: .

6. The machine vision-based leather surface defect detection system according to claim 5, characterized in that, The range of reference parameters pre-stored in the analysis unit is obtained in the following way: collection For each of the qualified sample images, calculate its first parameter. ,in , This is the preset sample size; calculate Mean of a sample ; then calculate Standard deviation of each sample ; by With the reference center, The baseline dispersion is used.

7. The machine vision-based leather surface defect detection system according to claim 6, characterized in that, The analysis unit will use the first parameter of the current image. The process of comparing with the benchmark range is as follows: Calculate the degree of deviation: ; Preset warning coefficient and action coefficient ,in ; like This is considered normal. like This is considered a warning and marked as a candidate defect area; like This area was determined to be a defective area.

8. The machine vision-based leather surface defect detection system according to claim 7, characterized in that, The analysis unit uses a statistical process control model to dynamically monitor the sequence changes of the first parameter, specifically including: Let the sequence number of the continuously acquired image frames be... The corresponding first parameter is ; Calculate the moving range ,in ; Based on the initial stage For a sample of qualified samples, calculate the mean of the moving range: ； Calculate individual control charts: centerline Upper control limit ;lower control limit ;in This is a preset constant; Perform the following determination process: like or This area was determined to be a defective area; If Q out of a consecutive P points exceed [the limit] Issue an early warning. Preset early warning coefficient; If all T consecutive points are greater than or all less than Issue a warning; If U consecutive points show a monotonically increasing or monotonically decreasing trend, an early warning will be issued; where P, Q, T, and U are all preset values.

9. The machine vision-based leather surface defect detection system according to claim 8, characterized in that, The specific method by which the analysis unit identifies the direction of leather deformation is as follows: Preset deformation ratio threshold range ,in Obtain the second parameter of the defect area image ; like It is determined to be a lateral tensile deformation; if It is determined to be longitudinal tensile deformation; if It was determined that there was no significant deformation. The execution module outputs prompt information related to the production process parameters based on the identified deformation direction: when it is determined to be transverse tensile deformation, it outputs a prompt of abnormal traction tension; when it is determined to be longitudinal tensile deformation, it outputs a prompt of uneven substrate elongation or difference in pressure roller speed.

Images