Brake pad surface damage detection method and system based on machine vision

Abstract

The invention belongs to the technical field of image data processing, and particularly relates to a brake pad surface damage detection method and system based on machine vision, and the method comprises the steps: obtaining an image of the surface of a brake pad, and calculating the gradient vector of each pixel point; constructing a gradient interaction field based on the intensity of the gradient magnitude and the gradient direction angle, and obtaining the structural significance of each pixel point; path searching is carried out in the local main direction, and linear propagation potential energy of each pixel point is obtained by accumulating structural significance and direction consistency weights of the pixel points on the path; and adaptive threshold segmentation and morphological processing are carried out based on the linear propagation potential energy diagram, and accurate detection of the surface damage area of the brake pad is realized. According to the invention, the technical problems of strong noise interference on the surface of the brake pad and difficulty in detection of discontinuous micro crack characteristics are solved, and the robustness and accuracy of damage detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing. More particularly, the present application relates to a brake pad surface damage detection method and system based on machine vision. BACKGROUND

[0002] In the field of image data processing (DIP), machine vision (MV) based industrial detection technology has been widely applied in the quality control process of precision parts. Brake pads, as the core components of the automobile braking system, their surface quality directly affects driving safety. In the existing brake pad production line, surface damage detection mainly faces complex environmental interference. Because the brake pad surface is pressed by a variety of composite materials, it contains a large number of metal friction material particles with high reflection characteristics, which will produce a large number of randomly distributed highlight spots and noise in the gray-scale image collected by the industrial camera.

[0003] In the traditional image processing method, edge detection or simple gradient threshold segmentation algorithm is usually used to extract surface damage. However, the background texture of the brake pad surface is very chaotic, and the gradient amplitude produced by the metal particles is often similar to that of the real micro-cracks, making it difficult for amplitude-based detection algorithms to distinguish between isolated noise points and real structural features. In addition, real surface damage such as micro-cracks, due to the influence of light angle and surface undulation, its characteristics in the image often show intermittent distribution state, lack of spatial connectivity, this weak connectivity of micro-damage features leads to the traditional algorithm in the aspect of stitching crack fragments, suppressing background interference, the detection accuracy and the robustness of the system are difficult to meet the high-precision industrial quality inspection demand. Therefore, how to accurately extract the damage features with topological continuity from the strong noise background and realize the effective enhancement of the intermittent cracks has become the research focus in the field of surface damage detection of parts. SUMMARY

[0004] To solve the technical problems of strong noise interference and intermittent detection of micro-crack features on the surface of the brake pad, the present application provides solutions in the following aspects.

[0005] In a first aspect, the present application provides a brake pad surface damage detection method based on machine vision, comprising: acquiring an image of the brake pad surface and preprocessing; calculating the gradient amplitude and gradient direction angle of each pixel point of the preprocessed image; for any target pixel point, setting an interaction window, based on the gradient amplitude and gradient direction angle of the target pixel point and the neighborhood pixel points in the interaction window, obtaining the structural saliency of the target pixel point; obtaining the local principal direction of the target pixel point, performing path search on the structural saliency graph along the positive direction and the reverse direction of the local principal direction, and based on the structural saliency and the angle consistency of the local principal direction of the pixel points on the search path, obtaining the linear propagation potential of the target pixel point; based on the maximum inter-class variance method, thresholding the linear propagation potential graph to obtain a binary image, and performing morphological closing operation and connected component filtering on the binary image to obtain the brake pad surface damage area.

[0006] The present application calculates the gradient amplitude and gradient direction angle of each pixel point, and further obtains the structural saliency and the linear propagation potential, uses the gradient cooperative relationship between pixel points to suppress the background texture interference caused by the metal particles on the brake pad surface, and uses the cumulative characteristics of the potential along the local principal direction to connect the intermittent micro-crack features, thereby realizing the detection of the brake pad surface damage area in a complex background and improving the robustness of the detection.

[0007] Preferably, the acquiring an image of the brake pad surface and preprocessing comprises: performing smoothing processing on the image by using a Gaussian filter with a Gaussian kernel size of and a standard deviation of .

[0008] The present application uses a Gaussian filter with a specific size and standard deviation to perform smoothing processing on the image, and by performing weighted average on the image gray data, the noise caused by the micro-rough particles on the brake pad surface is smoothed while the main damage edge features are retained, thereby providing a better data basis for the subsequent accurate calculation of the gradient amplitude and the gradient direction angle.

[0009] Preferably, the structural saliency satisfies the expression: ; wherein, denotes the structural saliency of the target pixel point with coordinates of ; denotes the total number of neighborhood pixel points in the interaction window except the target pixel point ; denotes the interaction window; denotes the neighborhood pixel points in the interaction window; denotes the gradient amplitude of the target pixel point; denotes the gradient amplitude of the neighborhood pixel point . Indicates the gradient direction angle of the target pixel; Represents neighboring pixels The gradient direction angle; It is a constant used to prevent the denominator from being zero.

[0010] This invention comprehensively considers the intensity affinity of gradient magnitude and the synergy of gradient direction angle when obtaining structural saliency. It uses the difference in gradient characteristics between neighboring pixels and target pixels to evaluate their structural importance, thereby enhancing the real damage features with spatial regularity and suppressing the random background texture, thus highlighting the potential damage features on the brake pad surface.

[0011] Preferably, the constant for preventing the denominator from being zero is set to... The size of the interaction window is set to... .

[0012] Preferably, the local principal direction of the target pixel is the direction perpendicular to the gradient direction angle of the target pixel.

[0013] Preferably, the linear propagation potential energy satisfies the expression: ;in, Indicates coordinates as The linear propagation potential energy of the target pixel; Indicates the structural saliency of the target pixel; Indicates the length of a one-sided search path; Indicates the first position on the search path along the positive direction of the local principal direction. 1 pixel; Indicates the first position on the search path in the opposite direction of the local principal direction. 1 pixel; Represents pixels Structural salience; Represents pixels Structural salience; The angle representing the local principal direction of the target pixel; Represents pixels The angle of the local principal direction; Represents pixels The angle of the local principal direction; The base of the natural logarithm; Represents the absolute value symbol.

[0014] This invention searches for paths in both the positive and negative directions along the local principal direction and obtains linear propagation potential energy. It uses angle consistency weights to accumulate the structural saliency on the search path, so that discontinuous crack features with spatial extension trends can obtain higher potential energy values, thereby reducing the missed detection caused by weak or discontinuous damage features on the brake pad surface.

[0015] Preferably, the length of the one-sided search path is set to Pixel.

[0016] Preferably, the step of thresholding the linear propagation potential map to obtain a binarized image based on the maximum inter-class variance method includes: calculating the optimal segmentation threshold of the linear propagation potential map using the maximum inter-class variance method; marking pixels in the linear propagation potential map that are greater than the optimal segmentation threshold as candidate damage points, and marking pixels that are less than or equal to the optimal segmentation threshold as background, thereby obtaining a binarized image.

[0017] Preferably, the morphological closing operation and connected component filtering of the binarized image includes: selecting... Perform morphological closing operations on the rectangular kernel; filter out areas smaller than Pixels or aspect ratio smaller than Connected components.

[0018] This invention performs morphological closing operations and connected component filtering on binarized images. It uses rectangular kernels to fill the tiny holes inside the damaged area and connect adjacent breakpoints. At the same time, it filters out pseudo-damage interference that does not conform to the crack morphology based on the area and aspect ratio features of the connected components, ensuring that the damaged area on the surface of the brake pad in the final output is more consistent with the physical characteristics of real damage in terms of shape and size.

[0019] Secondly, the present invention provides a machine vision-based brake pad surface damage detection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned machine vision-based brake pad surface damage detection method is implemented.

[0020] By adopting the above technical solution, the above-mentioned machine vision-based brake pad surface damage detection method is generated into a computer program and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0021] The beneficial effects of this invention are as follows: This invention utilizes the principle of gradient interaction fields to obtain structural saliency. By evaluating the synergistic relationship between gradient magnitude and direction between pixels, it reduces noise interference caused by the reflection of metal friction material particles on the brake pad surface, enabling the detection algorithm to focus more on real damage with specific structural features.

[0022] This invention introduces a linear propagation potential energy model to process the structural saliency map, searches for and integrates the energy along the local principal direction of the texture, and numerically connects the originally discontinuously distributed micro-cracks or scratches, thereby improving the ability to detect weak and discontinuous damage on the brake pad surface.

[0023] This invention combines the Otsu's method with morphological processing techniques to analyze linear propagation potential energy maps, adaptively obtains segmentation thresholds, and performs geometric screening on the binarized connected components. This reduces the uncertainty caused by manually set parameters and ensures the stability of brake pad surface damage detection results under different production environments. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the machine vision-based brake pad surface damage detection method of the present invention; Figure 2 This is a schematic diagram showing the grayscale image after noise reduction preprocessing; Figure 3 This is a schematic diagram illustrating the distribution of structural saliency. Figure 4 This is a schematic diagram illustrating the damage identification results. Detailed Implementation

[0025] 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, not all, of the embodiments of the present invention. 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a machine vision-based method for detecting surface damage on brake pads, referring to... Figure 1 This includes steps S1-S4: S1: Acquire an image of the brake pad surface and preprocess it; calculate the gradient magnitude and gradient direction angle of each pixel in the preprocessed image.

[0028] It should be noted that in order to extract the subtle features of the brake pad surface from the image, this invention requires acquiring high-quality image data and converting it into gradient field data that can reflect the drastic changes in grayscale. Since the brake pad surface contains a large number of tiny friction material particles, directly processing the original image would introduce a significant amount of noise. Therefore, this invention first calculates gradient information as the basis for subsequent texture analysis.

[0029] Specifically, raw images of the brake pad surface are acquired using an industrial camera and a coaxial light source, and then converted into single-channel images. Gaussian smoothing is then applied to the images to obtain the smoothed images. The Gaussian kernel size for the smoothing process is set to... Standard deviation Set as .

[0030] Furthermore, for any pixel in the image, the Sobel operator is used to calculate its gray-level derivatives in the horizontal and vertical directions, respectively, to obtain the horizontal and vertical gradient components of that pixel. These components form the gradient vector of the pixel, and the gradient magnitude and gradient direction angle are calculated. The gradient vectors of all pixels constitute the gradient vector field of the entire image.

[0031] For example, Figure 2 This is a grayscale illustration after denoising preprocessing. The image shows the true state of the surface to be detected, revealing uneven background texture. Through enhancement processing, the crack details in the image become clearer, and the contrast between the crack area and the surrounding background is improved, providing a clear base image for accurate crack identification.

[0032] S2: For any target pixel, set an interaction window, and obtain the structural saliency of the target pixel based on the gradient magnitude and gradient direction angle of the target pixel and its neighboring pixels within the interaction window.

[0033] It should be noted that traditional gradient magnitude methods or vector summation methods are easily affected by noise, as some bright metal particles on the brake pad surface can also generate extremely large gradient amplitudes. To more fundamentally distinguish between disordered textures and ordered damage through physical properties, this invention introduces the concept of a gradient interaction field. This invention treats each pixel as a particle with a specific gradient amplitude and gradient direction angle, calculating the cooperative interaction force between the target pixel and all its neighboring pixels within its interaction window. This interaction force is significantly enhanced only when neighboring pixels are close to the target pixel in gradient amplitude and highly cooperative in gradient direction angle, exhibiting parallelism or antiparallelism, thereby effectively suppressing isolated bright noise and cluttered texture backgrounds.

[0034] Specifically, for any target pixel in the image, an interaction window centered on that target pixel is defined to obtain the structural saliency of that target pixel. The size of the interaction window is set to... .

[0035] The structural saliency satisfies the expression:

[0036] In the formula, Indicates coordinates as The structural saliency of the target pixel; Representing the interaction window Remove target pixels The total number of neighboring pixels outside of the specified range; Represents the interaction window; This represents the neighboring pixels within the interaction window; This represents the gradient magnitude of the target pixel. Represents neighboring pixels The gradient magnitude; Indicates the gradient direction angle of the target pixel; Represents neighboring pixels The gradient direction angle; It is a constant to prevent the denominator from being zero; an empirical value. .

[0037] In the formula, It is a normalization coefficient used to eliminate the influence of window size on the magnitude of the calculation result; It is a normalized weight based on the harmonic average principle. When the gradient magnitude of the target pixel is closer to that of its neighboring pixels, this term approaches 1. When the gradient magnitudes of the two pixels are significantly different, this term approaches 0. This results in the suppression of isolated, highly reflective particle noise. The square of the cosine is used to evaluate the consistency of direction. The squaring operation eliminates the positive and negative differences in direction, so that gradients that are 180 degrees apart can also be regarded as highly coordinated, which leads to the enhancement of the features of the edges on both sides of the crack.

[0038] For example, Figure 3 This is a schematic diagram of the structural saliency distribution. The bright areas represent the crack skeleton with significant extension, while the dark areas represent irregular background noise that has been smoothed and suppressed. This invention effectively filters out the interference of fine surface textures and preliminarily locates the main structural morphology of the crack.

[0039] S3: Obtain the local principal direction of the target pixel, perform path search on the structural saliency map along the positive and negative directions of the local principal direction, and obtain the linear propagation potential of the target pixel based on the structural saliency of the pixels on the search path and the angular consistency of the local principal direction.

[0040] It should be noted that, after structural saliency calculations, most random textures in the image have been suppressed, but discontinuous damage features or tiny scratch fragments may still exist. Real cracks or scratches have significant topological continuity in space. To stitch together fracture features and further amplify real damage, this invention utilizes the concept of linearly propagating potential energy, performing energy integration along the potential extension direction of the texture.

[0041] Specifically, for any target pixel, the local principal direction of that target pixel is obtained. The local principal direction is the direction perpendicular to the gradient direction angle of that target pixel.

[0042] Furthermore, along the positive and negative directions of the local principal direction, path searches are performed on the structural saliency map, and the weighted cumulative value of the structural saliency of all pixels on the search path is calculated to obtain the linear propagation potential of the target pixel. The structural saliency map is an image composed of the structural saliency of all pixels.

[0043] The linear propagation potential energy satisfies the expression:

[0044] In the formula, Indicates coordinates as The linear propagation potential energy of the target pixel; Indicates the structural saliency of the target pixel; Indicates the length of a one-sided search path; Indicates the first position on the search path along the positive direction of the local principal direction. 1 pixel; Indicates the first position on the search path in the opposite direction of the local principal direction. 1 pixel; Represents pixels Structural salience; Represents pixels Structural salience; The angle representing the local principal direction of the target pixel; Represents pixels The angle of the local principal direction; Represents pixels The angle of the local principal direction; The base of the natural logarithm; Represents the absolute value symbol.

[0045] In the formula, if the pixels on the search path not only have high structural saliency, but also have an angle deviation between the local principal direction and the local principal direction of the target pixel, then... If the exponent is very small, then the exponent term Approaching 1 results in a linear propagation potential energy. Significant cumulative enhancement is achieved. This means that the linear propagation potential energy is conducted in space, allowing continuous real damage to obtain extremely high potential energy values, while isolated pseudo-features are suppressed because they cannot achieve path accumulation.

[0046] It should be further noted that the length of the unilateral search path in this invention... Set as Pixels. This length is set based on the discontinuous intervals of a typical crack. If the length is too short, the broken crack segments cannot be stitched together; if the length is too long, it will increase the computational load and may introduce incorrect connections.

[0047] S4: Threshold segmentation of the linear propagation potential energy map is performed based on the Otsu's method to obtain a binarized image. Morphological closing operation and connected component filtering are then performed on the binarized image to obtain the damage area on the brake pad surface.

[0048] It should be noted that, after potential energy accumulation, the actual damaged area appears as a bright connected region in the linear propagation potential energy map, while background texture and noise are significantly suppressed to low grayscale areas. At this point, the contrast between the damaged features and the background has been greatly increased, allowing for segmentation using a thresholding method.

[0049] Specifically, the optimal segmentation threshold of the linear propagation potential map is automatically calculated using the Otsu's method. Pixels in the linear propagation potential map with a value greater than the optimal segmentation threshold are marked as candidate damage points, while pixels with a value less than or equal to the optimal segmentation threshold are marked as background, thus generating a binarized image. The linear propagation potential map is an image composed of the linear propagation potential of all pixels.

[0050] Furthermore, morphological closing operations are performed on the binarized image, and the structuring element for the morphological closing operation is selected as... A rectangular kernel is then constructed. Connectivity analysis is performed, calculating the area and aspect ratio of each connected component. Components with areas smaller than [a certain value] are filtered out. Pixels or aspect ratio smaller than The remaining connected components are the final determined areas of brake pad surface damage.

[0051] For example, Figure 4 This is a schematic diagram of the damage identification results. The diagram shows the final crack detection results after processing.

[0052] This invention also discloses a machine vision-based brake pad surface damage detection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the machine vision-based brake pad surface damage detection method according to this invention.

[0053] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A machine vision-based method for detecting surface damage on brake pads, characterized in that, include: Acquire and preprocess an image of the brake pad surface; Calculate the gradient magnitude and gradient direction angle for each pixel in the preprocessed image; For any target pixel, an interaction window is set, and the structural saliency of the target pixel is obtained based on the gradient magnitude and gradient direction angle between the target pixel and the neighboring pixels within the interaction window. Obtain the local principal direction of the target pixel, and perform path search on the structure saliency map along the positive and negative directions of the local principal direction. Based on the structural saliency of the pixels on the search path and the angular consistency of the local principal direction, obtain the linear propagation potential of the target pixel. The linear propagation potential energy map is thresholded using the Otsu's method to obtain a binarized image. Morphological closing operations and connected component filtering are then performed on the binarized image to obtain the damaged area on the brake pad surface.

2. The machine vision-based brake pad surface damage detection method according to claim 1, characterized in that, The process of acquiring and preprocessing the image of the brake pad surface includes: Using a Gaussian kernel size of Standard deviation is A Gaussian filter is used to smooth the image.

3. The machine vision-based brake pad surface damage detection method according to claim 1, characterized in that, The structural saliency satisfies the expression: ； in, Indicates coordinates as The structural saliency of the target pixel; Representing the interaction window Remove target pixels The total number of neighboring pixels outside of the specified range; Represents the interaction window; This represents the neighboring pixels within the interaction window; This represents the gradient magnitude of the target pixel. Represents neighboring pixels The gradient magnitude; Indicates the gradient direction angle of the target pixel; Represents neighboring pixels The gradient direction angle; It is a constant used to prevent the denominator from being zero.

4. The machine vision-based brake pad surface damage detection method according to claim 3, characterized in that, The constant used to prevent the denominator from being zero is set as follows: The size of the interaction window is set to... .

5. The machine vision-based brake pad surface damage detection method according to claim 1, characterized in that, The local principal direction of the target pixel is the direction perpendicular to the gradient direction angle of the target pixel.

6. The machine vision-based brake pad surface damage detection method according to claim 5, characterized in that, The linear propagation potential energy satisfies the expression: ； in, Indicates coordinates as The linear propagation potential energy of the target pixel; Indicates the structural saliency of the target pixel; Indicates the length of a one-sided search path; Indicates the first position on the search path along the positive direction of the local principal direction. 1 pixel; Indicates the first position on the search path in the opposite direction of the local principal direction. 1 pixel; Represents pixels Structural salience; Represents pixels Structural salience; The angle representing the local principal direction of the target pixel; Represents pixels The angle of the local principal direction; Represents pixels The angle of the local principal direction; The base of the natural logarithm; Represents the absolute value symbol.

7. The machine vision-based brake pad surface damage detection method according to claim 6, characterized in that, The length of the one-sided search path is set to Pixel.

8. The machine vision-based brake pad surface damage detection method according to claim 1, characterized in that, The step of thresholding the linear propagation potential map to obtain a binarized image based on the maximum inter-class variance method includes: The optimal segmentation threshold of the linear propagation potential map is calculated using the Otsu's method. Pixels in the linear propagation potential map that are greater than the optimal segmentation threshold are marked as candidate damage points, and pixels that are less than or equal to the optimal segmentation threshold are marked as background, thus obtaining a binarized image.

9. The machine vision-based brake pad surface damage detection method according to claim 1, characterized in that, The morphological closing operation and connected component filtering of the binarized image include: Select Perform morphological closing operations on the rectangular kernel; filter out areas smaller than Pixels or aspect ratio smaller than Connected components.

10. A machine vision-based brake pad surface damage detection system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the machine vision-based brake pad surface damage detection method according to any one of claims 1-9.

Images