A consolidated abrasive tool surface quality detection method based on gray level extreme value analysis

Abstract

The application discloses a kind of based on gray extremum analysis consolidated abrasive tool surface quality detection method, method includes image acquisition and pre-processing, candidate area positioning and feature point extraction, target area boundary adaptive extraction, region screening and gray difference calculation, particle gray and spatial distribution feature quantization five main steps.Through line scanning camera, surface brightness chart and depth chart are collected, after gray enhancement and filtering pretreatment, local extreme point is extracted as candidate area by line scanning, and target boundary is adaptively generated in combination with cross neighborhood gray contrast and gradient chart dynamic threshold criterion;Effective area is screened by morphological operation and area constraint again, and gray difference is calculated to represent particle height;Finally, multidimensional quantitative evaluation is realized through gray difference histogram, protruding height statistics, nearest neighbor distance analysis and grid uniformity measurement.The application has high sensitivity, high robustness and strong adaptability, and can be embedded in industrial production line to realize automatic, non-contact and high-precision surface quality detection.

Description

Technical Field

[0001] This invention relates to the field of machine vision inspection technology, and in particular to a method for inspecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis. Background Technology

[0002] Bonded abrasive tools, as common high-performance cutting and grinding tools, are widely used in automotive manufacturing, aerospace, and optical device processing. The morphology and distribution of abrasive grains on their surface directly affect grinding efficiency, machining accuracy, and service life. With the development of intelligent manufacturing, the demand for online, non-contact inspection of the surface quality of bonded abrasive tools is increasing. Traditional manual inspection relies on visual inspection, which is inefficient and easily affected by subjective factors.

[0003] In recent years, machine vision-based surface quality inspection technology has been increasingly applied to the inspection of bonded abrasive tools. Prominence detection methods based on image segmentation rely on fixed thresholds, have limited ability to detect minute topographic features, and are sensitive to changes in illumination. Deep learning-based detection methods can improve the overall recognition rate, but their accuracy for small targets is low and they require large amounts of labeled data.

[0004] Existing two-dimensional vision methods struggle to capture minute particles on abrasive surfaces, often relying on fixed threshold segmentation. They are insensitive to subtle morphological variations with slight grayscale changes, leading to frequent missed detections. Traditional statistical methods are limited to the number and area of ​​abrasive particles, failing to quantitatively analyze depth differences and surface aggregation and uniformity, hindering a comprehensive assessment of abrasive tool performance. While simple morphological and thresholding methods are fast, they are extremely sensitive to lighting, background texture, and image trailing. High-precision deep learning-based methods require extensive training data and complex networks, making them unsuitable for real-time detection in high-resolution pipeline environments. Therefore, a detection method combining high precision, robustness, and real-time performance is needed. This method should automatically identify and extract minute particle regions on abrasive surfaces, while simultaneously quantifying their grayscale differences and spatial distribution in multiple dimensions, enabling a comprehensive and accurate assessment of the surface quality of bonded abrasive tools. Summary of the Invention

[0005] The purpose of this invention is to address the problems of insufficient identification of minute morphological features, poor robustness, and weak quantification capability in existing detection methods. It provides a surface quality detection method for bonded abrasive tools based on grayscale extreme value analysis. Through gradient analysis and local extreme value detection, it achieves accurate extraction and quantitative characterization of particle regions on the surface of bonded abrasive tools, improving detection accuracy, robustness, and real-time performance, and meeting the needs of industrial online detection.

[0006] The technical solution adopted in this invention is:

[0007] A method for surface quality inspection of bonded abrasive tools based on grayscale extreme value analysis includes the following steps:

[0008] Image acquisition and preprocessing: acquire the brightness map and depth map of the surface of the bonded abrasive tool, perform grayscale enhancement and filtering on the depth map to obtain the corresponding derivative map, and define the region of interest (ROI).

[0009] Candidate region feature extraction: Candidate points are extracted by scanning gray values ​​line by line and applying a local extremum detection algorithm. Candidate points are then selected based on the gray value comparison of the neighborhood.

[0010] Target region boundary determination: The gray-level discrimination region is obtained based on the gray-level value of the candidate point and the preset offset. The differential boundary region is set based on the differential value of the gradient map. The connected region of the target particle is generated by identifying the connected region in the neighborhood of the candidate point that satisfies the gray-level discrimination region and the differential boundary region.

[0011] Region filtering: Valid regions are filtered through morphological operations and area constraints, and the gray-scale difference is calculated as an estimated value of the exposure height;

[0012] Quantitative analysis: The grayscale difference statistics and spatial distribution uniformity evaluation of the selected effective areas are performed, and quality indicators are output.

[0013] Furthermore, the steps for image acquisition and preprocessing are as follows:

[0014] A line scan camera is used to acquire images of the surface of the bonded abrasive tool to obtain brightness and depth maps; specifically, the camera resolution is above 5 micrometers to ensure the capture of minute surface features.

[0015] Linear grayscale enhancement is applied to the depth map, using a transformation formula to enhance image contrast and depth levels. Specifically, as a feasible implementation, adjustments can be made based on the image brightness distribution.

[0016] The surface features of the region of interest (ROI) are made clear by manually drawing and defining the ROI, and then performing linear grayscale enhancement and Gaussian filtering on the ROI.

[0017] Furthermore, the region of interest (ROI) is eroded to reduce the edges, and then the complement of the eroded ROI is taken and dilated to obtain an enhanced image. In the enhanced image, the complement region is filled with a gray value of 10000 to simulate a uniform background. Finally, Gaussian filtering is applied to reduce the noise of the enhanced image and ensure that the surface features are clear.

[0018] Furthermore, the steps for candidate region feature extraction are as follows:

[0019] Gray values ​​of the depth map are scanned line by line: traverse each row of pixels in the image from top to bottom, extract the gray value sequence of the current row, and construct a one-dimensional gray function;

[0020] Gray-level smoothing and extremum extraction: The moving average smoothing is applied to the gray-level function to reduce the influence of noise; the local extremum detection algorithm is used to extract the local minimum and maximum values ​​within the row, generate a candidate point set, and record the sub-pixel coordinates;

[0021] Neighborhood grayscale extraction: Construct horizontal or vertical analysis windows centered on each extreme point.

[0022] Candidate point screening: Candidate points are screened based on grayscale difference to ensure that the maximum grayscale value on the left and right sides of the damaged abrasive particle is greater than the grayscale value of the center point plus the offset. The offset value is determined through experimental optimization to ensure the sensitivity of detecting tiny abrasive particles.

[0023] Specifically, for each local extremum point in the candidate point set, a neighborhood analysis window of a preset size is constructed with the extremum point as the center, and grayscale distribution information in the left-right or up-down directions is extracted respectively; by comparing the grayscale value of the extremum point with the maximum or minimum grayscale value in the corresponding direction in the neighborhood, the neighborhood grayscale difference is calculated; when the neighborhood grayscale difference is greater than a preset grayscale offset threshold, it is determined that the surface structure corresponding to the extremum point has obvious concave or convex features, and it is retained as a valid candidate point; wherein, the grayscale offset threshold is determined experimentally based on the resolution of the imaging system and the abrasive particle size.

[0024] Furthermore, the steps for determining the boundary of the target area are as follows:

[0025] The window area is constructed by expanding the window to the left and right sides with each candidate point as the center, thus creating a horizontal analysis area.

[0026] Gray-scale difference and differential criteria: In the depth map, for the region of broken abrasive particles, extract the region with gray value greater than the gray value of the candidate point plus the offset; for the region of normal abrasive particles, extract the region with gray value less than the gray value of the candidate point minus the offset, thus obtaining the gray-scale discrimination region.

[0027] In the derivative plot, for the region of damaged abrasive particles, extract the region with a left differential value less than -2 or a right differential value greater than 2; for the region of normal abrasive particles, extract the region with a left differential value greater than 1 or a right differential value less than -1 to obtain the differential boundary region.

[0028] Specifically, in the derivative map (Gaussian derivative along the x-direction, σ=1), for regions with broken abrasive particles, areas with a left derivative value less than -2 (indicating a rapid decrease in grayscale) or a right derivative value greater than 2 (indicating a rapid increase in grayscale) are extracted; for regions with normal abrasive particles, areas with a left derivative value greater than 1 (indicating an increase in grayscale) or a right derivative value less than -1 (indicating a decrease in grayscale) are extracted. The derivative map is generated during the depth map preprocessing stage and serves as an intermediate feature map for grayscale extremum analysis and target region boundary determination.

[0029] The intersection generation region is obtained by finding the intersection of the grayscale discrimination region and the differential boundary region to obtain a connected region that meets the conditions. Using connected region analysis, for the damaged abrasive particle region, the left endpoint of the leftmost right region and the right endpoint of the rightmost left region are selected and connected to form a complete target region boundary. For the normal abrasive particle region in the connected region, the left endpoint of the leftmost right region and the right endpoint of the rightmost left region are selected and connected to generate the boundary.

[0030] Specifically, the rightmost region on the left refers to the connected region with the smallest difference between its right boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained by intersecting the grayscale discrimination region and the differential boundary region within a horizontal analysis window extending to the left centered on the candidate point; the endpoint corresponding to the right boundary of this connected region is used as the boundary point of the target region in the left direction. The leftmost region on the right refers to the connected region with the smallest difference between its left boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained by intersecting the grayscale discrimination region and the differential boundary region within a horizontal analysis window extending to the right centered on the candidate point; the endpoint corresponding to the left boundary of this connected region is used as the boundary point of the target region in the right direction.

[0031] Furthermore, the steps for region filtering are as follows:

[0032] Region morphology processing: Perform morphological opening operations (circular structuring element, radius 1.5 pixels) on the target region within the target region boundary to smooth the edges and remove small pseudo-regions;

[0033] Area constraint: Filter the target area by area, retain areas within the specified range, and remove false detection areas that are too small or too large;

[0034] Gray-scale difference calculation: For each normal abrasive particle region, an expansion operation is performed, and the difference between the maximum gray-scale value of the original region and the minimum gray-scale value of the expanded region is calculated as the estimated exposure height; the region with gray-scale difference within the specified range is retained as the effective region;

[0035] Specifically, the region with a grayscale difference in the range of 10 to 500 is retained. This range was determined by grayscale analysis of various bonded abrasive tool samples and can effectively cover the height variations of micro-protrusions and micro-damage.

[0036] Region Connectivity and Feature Recording: Connect all valid regions to form the final candidate set, and record the gray-level difference of each target to generate a gray-level difference vector.

[0037] Furthermore, the steps of quantitative analysis are as follows:

[0038] Generate a grayscale difference histogram: This involves constructing a height vector from the estimated exposure heights of each abrasive particle. The intervals between the minimum value Δmin and the maximum value Δmax of} are divided into B bins of equal width, with each bin having an interval length of approximately Each estimated exposure height is assigned to its corresponding compartment, and the number of estimated exposure heights for each compartment is counted to obtain the compartment count. ;

[0039] Exposure height statistics: Calculate the minimum, maximum, average and standard deviation of the exposure height; the exposure height is the height of the abrasive particles relative to the binder surface, represented by the height vector.

[0040] Specifically, the overall exposure height level and dispersion are assessed to provide a basis for process optimization. A height vector is available. } Calculate the minimum value Δmin, the maximum value Δmax, and the mean. and standard deviation N represents the total number of abrasive particles that, after screening, participate in the exposure height statistics within the region of interest.

[0041] Nearest neighbor distance analysis: Calculate the minimum Euclidean distance between the geometric center coordinates of all pixels of each abrasive particle and the center coordinates of other abrasive particles, and calculate the mean, standard deviation, and coefficient of variation (CV = standard deviation / mean) of the minimum distance; a lower CV value indicates a more uniform particle distribution.

[0042] Grid uniformity measurement: Divide the depth map evenly into a grid of a specified size (e.g., 10×10 grid, which can be adjusted according to the image size), count the number of particles in each grid, remove empty grids, and calculate the mean and standard deviation of the number of particles. The uniformity index Grid_CV (standard deviation / mean) reflects the uniformity of particle arrangement.

[0043] The present invention adopts the above technical solution, and compared with the prior art, the present invention has the following beneficial technical effects:

[0044] (1) By extracting local extreme points through an inline sliding window and combining it with cross-shaped gray-scale neighborhood comparison, the detection rate of weak feature structures is significantly improved;

[0045] (2) Based on the gradient map, the present invention combines a dynamic cropping strategy for candidate points, which avoids the sensitivity of fixed thresholds to factors such as imaging tailing and reflection artifacts. It maintains high stability under different lighting and background conditions, and is suitable for online detection in complex industrial environments.

[0046] (3) This invention not only extracts the particle region, but also statistically analyzes its gray-scale difference histogram, the mean and variance of the exposure height, and calculates the nearest neighbor distance, standard deviation, coefficient of variation and Gini coefficient of the grid distribution of the particle spatial distribution based on the centroid position of the region. It can comprehensively quantify the uniformity and distribution state of the particles, and is suitable for grade evaluation and process optimization of the quality of bonded abrasive tools. Attached Figure Description

[0047] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;

[0048] Figure 1 This is a schematic flowchart of a method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to the present invention.

[0049] Figure 2 This is a schematic diagram of the brightness of the bonded abrasive tool of the present invention;

[0050] Figure 3 This is a schematic diagram of the depth of the bonded abrasive tool of the present invention;

[0051] Figure 4 This is a schematic diagram of the region of interest (ROI) on the surface of the bonded abrasive tool of the present invention.

[0052] Figure 5 This is a schematic diagram showing the ROI optimization results of the image after background removal according to the present invention;

[0053] Figure 6 This is a schematic diagram of the preprocessed image of the present invention after linear grayscale transformation and Gaussian smoothing;

[0054] Figure 7 The horizontal derivative image is used to highlight the defect edges in this invention;

[0055] Figure 8 This is a schematic diagram of local extreme points (blue crosses) representing candidate convex points detected by the present invention;

[0056] Figure 9 This is a schematic diagram of the candidate protrusions (green cross) selected by the present invention;

[0057] Figure 10 This is a schematic diagram of the raised area (yellow) that is finally identified by superimposing the original image in this invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0059] To address the shortcomings of existing detection methods, such as insufficient ability to identify minute morphological features, weak ability to quantify particle distribution, and poor robustness in complex backgrounds, this invention constructs an algorithm based on image gradient and local extremum analysis to accurately extract particle regions from the surface of bonded abrasive tools. Furthermore, it improves the accuracy and completeness of target region segmentation by combining a cross-shaped grayscale matrix and gradient-guided boundary clipping methods. Building upon this foundation, the invention further introduces a multi-dimensional quality quantification mechanism, combining regional grayscale difference statistics, mean and variance calculation of exposure height, nearest neighbor distance analysis, and coefficient of variation evaluation to achieve a comprehensive quantitative characterization of the particle surface state. This method possesses high sensitivity, high robustness, and high adaptability, and can be embedded in industrial production lines to achieve automated, non-contact, and high-precision quality inspection and evaluation of bonded abrasive tool surfaces.

[0060] like Figures 1 to 10 As shown in the figure, this invention discloses a method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis, comprising the following steps:

[0061] Image acquisition and preprocessing: acquire the brightness map and depth map of the surface of the bonded abrasive tool, perform grayscale enhancement and filtering on the depth map to obtain the corresponding derivative map, and define the region of interest (ROI).

[0062] Candidate region feature extraction: Candidate points are extracted by scanning gray values ​​line by line and applying a local extremum detection algorithm. Candidate points are then selected based on the gray value comparison of the neighborhood.

[0063] Target region boundary determination: The gray-level discrimination region is obtained based on the gray-level value of the candidate point and the preset offset. The differential boundary region is set based on the differential value of the gradient map. The connected region of the target particle is generated by identifying the connected region in the neighborhood of the candidate point that satisfies the gray-level discrimination region and the differential boundary region.

[0064] Specifically, boundary conditions are set based on the differential values ​​of the gradient map to distinguish between broken abrasive particles and normal abrasive particles.

[0065] Region filtering: Valid regions are filtered through morphological operations and area constraints, and the gray-scale difference is calculated as an estimated value of the exposure height;

[0066] Quantitative analysis: The grayscale difference statistics and spatial distribution uniformity evaluation of the selected effective areas are performed, and quality indicators are output.

[0067] Furthermore, the steps for image acquisition and preprocessing are as follows:

[0068] A line scan camera is used to acquire images of the surface of the bonded abrasive tool to obtain brightness and depth maps; specifically, the camera resolution is above 5 micrometers to ensure the capture of minute surface features.

[0069] Linear grayscale enhancement is applied to the depth map, using a transformation formula to enhance image contrast and depth levels. Specifically, as a feasible implementation, adjustments can be made based on the image brightness distribution.

[0070] The surface features of the region of interest (ROI) are made clear by manually drawing and defining the ROI, and then performing linear grayscale enhancement and Gaussian filtering on the ROI.

[0071] Furthermore, the region of interest (ROI) is eroded to reduce the edges, and then the complement of the eroded ROI is taken and dilated to obtain an enhanced image. In the enhanced image, the complement region is filled with a gray value of 10000 to simulate a uniform background. Finally, Gaussian filtering is applied to reduce the noise of the enhanced image and ensure that the surface features are clear.

[0072] Furthermore, the steps for candidate region feature extraction are as follows:

[0073] Gray values ​​of the depth map are scanned line by line: traverse each row of pixels in the image from top to bottom, extract the gray value sequence of the current row, and construct a one-dimensional gray function;

[0074] Gray-level smoothing and extremum extraction: The moving average smoothing is applied to the gray-level function to reduce the influence of noise; the local extremum detection algorithm is used to extract the local minimum and maximum values ​​within the row, generate a candidate point set, and record the sub-pixel coordinates;

[0075] Neighborhood grayscale extraction: Construct horizontal or vertical analysis windows centered on each extreme point.

[0076] Candidate point screening: Candidate points are screened based on grayscale difference to ensure that the maximum grayscale value on the left and right sides of the damaged abrasive particle is greater than the grayscale value of the center point plus the offset. The offset value is determined through experimental optimization to ensure the sensitivity of detecting tiny abrasive particles.

[0077] Specifically, for each local extremum point in the candidate point set, a neighborhood analysis window of a preset size is constructed with the extremum point as the center, and grayscale distribution information in the left-right or up-down directions is extracted respectively; by comparing the grayscale value of the extremum point with the maximum or minimum grayscale value in the corresponding direction in the neighborhood, the neighborhood grayscale difference is calculated; when the neighborhood grayscale difference is greater than a preset grayscale offset threshold, it is determined that the surface structure corresponding to the extremum point has obvious concave or convex features, and it is retained as a valid candidate point; wherein, the grayscale offset threshold is determined experimentally based on the resolution of the imaging system and the abrasive particle size.

[0078] Furthermore, the steps for determining the boundary of the target area are as follows:

[0079] The window area is constructed by expanding the window to the left and right sides with each candidate point as the center, thus creating a horizontal analysis area.

[0080] Gray-scale difference and differential criteria: In the depth map, for the region of broken abrasive particles, extract the region with gray value greater than the gray value of the candidate point plus the offset; for the region of normal abrasive particles, extract the region with gray value less than the gray value of the candidate point minus the offset, thus obtaining the gray-scale discrimination region.

[0081] In the derivative plot, for the region of damaged abrasive particles, extract the region with a left differential value less than -2 or a right differential value greater than 2; for the region of normal abrasive particles, extract the region with a left differential value greater than 1 or a right differential value less than -1 to obtain the differential boundary region.

[0082] Specifically, in the derivative plot (Gaussian derivative along the x-direction, σ=1), for the region of damaged abrasive particles, the region with a left derivative value less than -2 (indicating a rapid decrease in grayscale) or a right derivative value greater than 2 (indicating a rapid increase in grayscale) is extracted; for the region of normal abrasive particles, the region with a left derivative value greater than 1 (indicating an increase in grayscale) or a right derivative value less than -1 (indicating a decrease in grayscale) is extracted.

[0083] The intersection generation region is obtained by finding the intersection of the grayscale discrimination region and the differential boundary region to obtain a connected region that meets the conditions. Using connected region analysis, for the damaged abrasive particle region, the left endpoint of the leftmost right region and the right endpoint of the rightmost left region are selected and connected to form a complete target region boundary. For the normal abrasive particle region in the connected region, the left endpoint of the leftmost right region and the right endpoint of the rightmost left region are selected and connected to generate the boundary.

[0084] Specifically, the rightmost region on the left refers to the connected region with the smallest difference between its right boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained by intersecting the grayscale discrimination region and the differential boundary region within a horizontal analysis window extending to the left centered on the candidate point; the endpoint corresponding to the right boundary of this connected region is used as the boundary point of the target region in the left direction. The leftmost region on the right refers to the connected region with the smallest difference between its left boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained by intersecting the grayscale discrimination region and the differential boundary region within a horizontal analysis window extending to the right centered on the candidate point; the endpoint corresponding to the left boundary of this connected region is used as the boundary point of the target region in the right direction.

[0085] Furthermore, the steps for region filtering are as follows:

[0086] Region morphology processing: Perform morphological opening operations (circular structuring element, radius 1.5 pixels) on the target region within the target region boundary to smooth the edges and remove small pseudo-regions;

[0087] Area constraint: Filter the target area by area, retain areas within the specified range, and remove false detection areas that are too small or too large;

[0088] Gray-scale difference calculation: For each normal abrasive particle region, an expansion operation is performed, and the difference between the maximum gray-scale value of the original region and the minimum gray-scale value of the expanded region is calculated as the estimated exposure height; the region with gray-scale difference within the specified range is retained as the effective region;

[0089] Specifically, the region with a grayscale difference in the range of 10 to 500 is retained. This range was determined by grayscale analysis of various bonded abrasive tool samples and can effectively cover the height variations of micro-protrusions and micro-damage.

[0090] Region Connectivity and Feature Recording: Connect all valid regions to form the final candidate set, and record the gray-level difference of each target to generate a gray-level difference vector.

[0091] Furthermore, the steps of quantitative analysis are as follows:

[0092] Generate a grayscale difference histogram: This involves constructing a height vector from the estimated exposure heights of each abrasive particle. The intervals between the minimum value Δmin and the maximum value Δmax of} are divided into B bins of equal width, with each bin having an interval length of approximately Each estimated exposure height is assigned to its corresponding compartment, and the number of estimated exposure heights for each compartment is counted to obtain the compartment count. ;

[0093] Exposure height statistics: Calculate the minimum, maximum, average and standard deviation of the exposure height; the exposure height is the height of the abrasive particles relative to the binder surface, represented by the height vector.

[0094] Specifically, the overall exposure height level and dispersion are assessed to provide a basis for process optimization. A height vector is available. } Calculate the minimum value Δmin, the maximum value Δmax, and the mean. and standard deviation N represents the total number of abrasive particles that, after screening, participate in the exposure height statistics within the region of interest.

[0095] Nearest neighbor distance analysis: Calculate the minimum Euclidean distance between the geometric center coordinates of all pixels of each abrasive particle and the center coordinates of other abrasive particles, and calculate the mean, standard deviation, and coefficient of variation (CV = standard deviation / mean) of the minimum distance; a lower CV value indicates a more uniform particle distribution.

[0096] Grid uniformity measurement: Divide the depth map evenly into a grid of a specified size (e.g., 10×10 grid, which can be adjusted according to the image size), count the number of particles in each grid, remove empty grids, and calculate the mean and standard deviation of the number of particles. The uniformity index Grid_CV (standard deviation / mean) reflects the uniformity of particle arrangement.

[0097] The present invention adopts the above technical solution, and compared with the prior art, the present invention has the following beneficial technical effects:

[0098] (1) By extracting local extreme points through an inline sliding window and combining it with cross-shaped gray-scale neighborhood comparison, the detection rate of weak feature structures is significantly improved;

[0099] (2) Based on the gradient map, the present invention combines a dynamic cropping strategy for candidate points, which avoids the sensitivity of fixed thresholds to factors such as imaging tailing and reflection artifacts. It maintains high stability under different lighting and background conditions, and is suitable for online detection in complex industrial environments.

[0100] (3) This invention not only extracts the particle region, but also statistically analyzes its gray-scale difference histogram, the mean and variance of the exposure height, and calculates the nearest neighbor distance, standard deviation, coefficient of variation and Gini coefficient of the grid distribution of the particle spatial distribution based on the centroid position of the region. It can comprehensively quantify the uniformity and distribution state of the particles, and is suitable for grade evaluation and process optimization of the quality of bonded abrasive tools.

[0101] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. A method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis, characterized in that: Includes the following steps: Image acquisition and preprocessing: acquire the brightness map and depth map of the surface of the bonded abrasive tool, perform grayscale enhancement and filtering on the depth map to obtain the corresponding derivative map, and define the region of interest; Candidate region feature extraction: Candidate points are extracted by scanning gray values ​​line by line and applying a local extremum detection algorithm. Candidate points are then selected based on the gray value comparison of the neighborhood. Target region boundary determination: The gray-level discrimination region is obtained based on the gray-level value of the candidate point and the preset offset. The differential boundary region is set based on the differential value of the gradient map. The connected region of the target particle is generated by identifying the connected region in the neighborhood of the candidate point that satisfies the gray-level discrimination region and the differential boundary region. Region filtering: Valid regions are filtered through morphological operations and area constraints, and the gray-scale difference is calculated as an estimated value of the exposure height; Quantitative analysis: The grayscale difference statistics and spatial distribution uniformity evaluation of the selected effective areas are performed, and quality indicators are output.

2. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 1, characterized in that: The steps for image acquisition and preprocessing are as follows: A line scan camera was used to acquire images of the surface of the bonded abrasive tool to obtain brightness and depth maps. Linear grayscale enhancement processing is performed on the depth map, and a transformation formula is used to enhance image contrast and depth levels; The surface features of the region of interest are made clear by manually drawing and defining the region of interest, and then applying linear grayscale enhancement and Gaussian filtering to the region of interest.

3. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 2, characterized in that: The region of interest is eroded to reduce the edges. Then, the complement of the eroded region of interest is taken and dilated to obtain an enhanced image. In the enhanced image, the complement region is filled with a gray value of 10000 to simulate a uniform background. Finally, Gaussian filtering is applied to reduce the noise of the enhanced image and ensure that the surface features are clear.

4. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 1, characterized in that: The steps for candidate region feature extraction are as follows: Gray values ​​of the depth map are scanned line by line: traverse each row of pixels in the image from top to bottom, extract the gray value sequence of the current row, and construct a one-dimensional gray function; Gray-level smoothing and extreme value extraction: Applying moving average smoothing to the gray-level function to reduce the impact of noise; The local extremum detection algorithm is used to extract the local minimum and maximum values ​​within a row, generate a candidate point set, and record the sub-pixel coordinates. Neighborhood grayscale extraction: Construct horizontal or vertical analysis windows centered on each extreme point; Candidate point screening: Candidate points are screened based on grayscale difference to ensure that the maximum grayscale value on the left and right sides of the damaged abrasive particle is greater than the grayscale value of the center point plus the offset. The offset value is determined through experimental optimization to ensure the sensitivity of detecting tiny abrasive particles.

5. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 4, characterized in that: The method for candidate point selection is as follows: construct a neighborhood analysis window of a preset size with each local extreme point as the center, and extract the gray-level distribution information in the left-right or up-down directions of the neighborhood analysis window respectively; calculate the neighborhood gray-level difference by comparing the gray-level value of the local extreme point with the maximum or minimum gray-level value in the corresponding direction within the neighborhood analysis window. When the gray level difference in the neighborhood is greater than the preset gray level offset threshold, the surface structure corresponding to the local extreme point is determined to have obvious concave or convex features, and the local extreme point is retained as a valid candidate point; wherein, the gray level offset threshold is determined experimentally based on the resolution of the imaging system and the abrasive particle size.

6. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 1, characterized in that: The steps for determining the boundary of the target area are as follows: Window region construction: Expand the window to the left and right sides with each candidate point as the center to construct a horizontal analysis region; Gray-scale difference and differential criterion: In the depth map, for the area of ​​broken abrasive particles, extract the area with gray value greater than the gray value of the candidate point plus the offset; for the area of ​​normal abrasive particles, extract the area with gray value less than the gray value of the candidate point minus the offset, thus obtaining the gray-scale discrimination area. In the derivative plot, for the region of damaged abrasive particles, extract the region with a left differential value less than -2 or a right differential value greater than 2; for the region of normal abrasive particles, extract the region with a left differential value greater than 1 or a right differential value less than -1 to obtain the differential boundary region. Intersection generation region: The intersection of the gray-level discrimination region and the differential boundary region is used to obtain a connected region that meets the conditions; Using connected component analysis, for the region of damaged abrasive particles, select the left endpoint of the rightmost region on the left and the right endpoint of the leftmost region on the right, and connect them to form the complete boundary of the target region; for the region of normal abrasive particles in the connected region, select the left endpoint of the rightmost region on the left and the right endpoint of the leftmost region on the right, and connect them to generate the boundary.

7. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 6, characterized in that: The rightmost region on the left refers to the connected region with the smallest difference between the right boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained after intersecting the grayscale discrimination region and the differential boundary region within a horizontal analysis window centered on the candidate point and extending to the left; the endpoint corresponding to the right boundary of the connected region with the smallest difference is taken as the boundary point of the target region in the left direction. The leftmost region on the right refers to the connected region with the smallest difference between its left boundary position and the column coordinates of the candidate point, selected from multiple connected regions obtained by intersecting the gray-level discrimination region and the differential boundary region within a horizontal analysis window centered on the candidate point and extending to the right. The endpoint corresponding to the left boundary of the connected region with the smallest difference is taken as the boundary point of the target region in the right direction.

8. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 1, characterized in that: The steps for region filtering are as follows: Region morphology processing: Perform morphological opening operations on the target region within the target region boundary to smooth the edges and remove small pseudo-regions; Area constraint: Filter the target area by area, retain areas within the specified range, and remove false detection areas that are too small or too large; Gray-scale difference calculation: For each normal abrasive particle region, an expansion operation is performed, and the difference between the maximum gray-scale value of the original region and the minimum gray-scale value of the expanded region is calculated as the estimated exposure height; the region with gray-scale difference within the specified range is retained as the effective region; Region Connectivity and Feature Recording: Connect all valid regions to form the final candidate set, and record the gray-level difference of each target to generate a gray-level difference vector.

9. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 8, characterized in that: Regions with a grayscale difference in the range of 10 to 500 are retained as valid regions.

10. The method for detecting the surface quality of bonded abrasive tools based on grayscale extreme value analysis according to claim 1, characterized in that: The steps of quantitative analysis are as follows: Generate a grayscale difference histogram: This involves constructing a height vector from the estimated exposure heights of each abrasive particle. The intervals between the minimum value Δmin and the maximum value Δmax of} are divided into B bins of equal width, with each bin having an interval length of approximately Each estimated exposure height is assigned to its corresponding compartment, and the number of estimated exposure heights for each compartment is counted to obtain the compartment count. ; Exposure height statistics: Calculate the minimum, maximum, average and standard deviation of the exposure height; the exposure height is the height of the abrasive particles relative to the binder surface, represented by the height vector. Nearest neighbor distance analysis: Calculate the minimum Euclidean distance between the geometric center coordinates of all pixels of each abrasive particle and the center coordinates of other abrasive particles, and calculate the mean, standard deviation and coefficient of variation of the minimum distance; Grid uniformity measurement: The depth map is uniformly divided into grids of a specified size, the number of particles in each grid is counted, empty grids are removed, and the mean and standard deviation of the number of particles are calculated. The uniformity index Grid_CV reflects the degree of uniformity of particle arrangement.

Images