Roller surface defect detection method, computer equipment and storage medium

Abstract

The invention discloses a roller surface defect detection method, computer equipment and a storage medium, and relates to the technical field of nondestructive detection.The method comprises the steps that an initial detection image, collected by image collection equipment at a preset detection position, of a to-be-detected roller is obtained, the image acquisition equipment with the preset detection position forms all-directional detection on the roller to be detected; fusing the initial detection images acquired by the image acquisition devices at the same moment to obtain a panoramic fusion image of the to-be-detected roller; extracting a first gray feature of the panoramic fusion image, and obtaining a second gray feature of the standard roller image; and determining an Euclidean distance between the first gray feature and the second gray feature, and determining a surface defect detection result of the to-be-detected roller based on the Euclidean distance. According to the technical scheme provided by the invention, the accuracy of roller surface defect detection is improved, and the method can be suitable for large-scale detection.

Description

Technical Field

[0001] The present invention relates to the technical field of nondestructive testing, and more specifically to a roller surface defect detection method, computer equipment and storage medium. Background Art

[0002] Rolling bearings are precision mechanical components that convert the sliding friction between the running shaft and the shaft seat into rolling friction, thereby reducing friction losses. Rollers (such as spherical rollers) of rolling bearings play a role in bearing and transmission in bearing transmission. They are important components of rolling bearings, and the quality of their surface directly affects the life of rolling bearings. Therefore, it is very important to detect surface defects of rollers in rolling bearings.

[0003] In the related art, the surface defects of the rollers of rolling bearings can be detected by manual visual observation. This method does not require complex auxiliary equipment and has a high recognition rate for small surface defects. However, manual observation is easily affected by subjective factors, with large fluctuations, which affects the accuracy of detection. In addition, human eyes are easily fatigued and are not suitable for large-scale detection. Summary of the invention

[0004] The embodiments of the present invention provide a roller surface defect detection method, a computer device and a storage medium, which are used to solve the problem that the roller surface defect detection method through manual naked eye observation in the prior art results in low detection accuracy and is not suitable for large-scale detection.

[0005] An embodiment of the present invention provides a roller surface defect detection method, comprising:

[0006] Acquire an initial detection image of the roller to be detected captured by an image acquisition device at a preset detection position; wherein the image acquisition device at the preset detection position forms an all-round detection of the roller to be detected;

[0007] Fusing the initial detection images acquired by the image acquisition devices at the same time to obtain a panoramic fused image of the roller to be detected;

[0008] Extracting a first grayscale feature of the panoramic fusion image, and acquiring a second grayscale feature of the standard roller image;

[0009] The Euclidean distance between the first grayscale feature and the second grayscale feature is determined, and a surface defect detection result of the roller to be detected is determined based on the Euclidean distance.

[0010] Optionally, fusing the initial detection images acquired by the image acquisition devices at the same time to obtain a panoramic fused image of the roller to be detected includes:

[0011] Taking each detection image in the initial detection images captured by each of the image acquisition devices at the same time as a first detection image, determining a first inscribed square area of ​​the roller image in the first detection image and a second inscribed square area of ​​the roller image in the second detection image to be fused; wherein the first detection image and the second detection image are the initial detection images captured by two adjacent image acquisition devices at the same time;

[0012] Superimposing the first detection image and the second detection image by overlapping adjacent edges between the first inscribed square area and the second inscribed square area to obtain a superimposed image;

[0013] Determine a spatial coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image;

[0014] The first detection image and the second detection image are converted to the same coordinate system according to the spatial coordinate conversion matrix, and the overlapping areas in the superimposed images are weightedly fused in the same coordinate system;

[0015] The above process is repeated until all the initial detection images are traversed, and the finally obtained fused image is determined as the panoramic fused image of the roller to be detected.

[0016] Optionally, the determining a spatial coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image includes:

[0017] Traversing each first pixel point of the first inscribed square area in the superimposed image, and determining the first pixel point as a first feature point when a brightness value of the first pixel point is greater than a brightness threshold;

[0018] Determine, within the second inscribed square area of ​​the superimposed image, a second feature point on a target straight line that is most symmetrical with the first feature point with a common side as a symmetry axis; wherein the common side is a common side of the first inscribed square area and the second inscribed square area in the superimposed image, and the target straight line is a pixel straight line that passes through the first feature point and is perpendicular to the common side;

[0019] The first feature point and the second feature point are taken as a pair of feature point pairs, and based on the feature point pairs, a spatial coordinate transformation matrix between the first detection image and the second detection image is determined using an affine transformation principle.

[0020] Optionally, the first grayscale feature includes a first grayscale mean and a first grayscale standard deviation, and the second grayscale feature includes a second grayscale mean and a second grayscale standard deviation;

[0021] The determining of the Euclidean distance between the first grayscale feature and the second grayscale feature, and determining the surface defect detection result of the roller to be detected based on the Euclidean distance, comprises:

[0022] constructing the first grayscale mean and the first grayscale standard deviation as a first eigenvector, and constructing the second grayscale mean and the second grayscale standard deviation as a second eigenvector;

[0023] Determine the Euclidean distance between the first eigenvector and the second eigenvector using the Euclidean distance formula of vectors;

[0024] When the Euclidean distance is greater than a preset distance threshold, determining that the surface defect detection result of the roller to be detected is that a surface defect exists;

[0025] When the Euclidean distance is less than or equal to the preset distance threshold, it is determined that the surface defect detection result of the roller to be detected is that there is no surface defect.

[0026] Optionally, the first grayscale feature further includes a first grayscale histogram, and the second grayscale feature further includes a second grayscale histogram; and the roller surface defect detection method further includes:

[0027] When it is determined that the surface defect detection result of the roller to be detected is that there is a surface defect, the first grayscale histogram is compared with the second grayscale histogram, and a target grayscale value of a sudden change in the first grayscale histogram is determined by comparison;

[0028] The regional position of the target grayscale value in the panoramic fusion image is obtained, and the regional position is determined as the defect position of the roller to be detected; and the surface defect detection result of the roller to be detected includes the defect position.

[0029] Optionally, before acquiring an initial detection image of the roller to be detected acquired by an image acquisition device at a preset detection position, the roller surface defect detection method further includes:

[0030] Obtaining target detection accuracy, the field of view of the image acquisition device, and the minimum number of pixels occupied by the defect;

[0031] Based on the target detection accuracy, the field of view of the image acquisition device and the minimum number of pixels occupied by the defect, the resolution of the image acquisition device is determined using a resolution formula; the resolution formula is:

[0032] Q=(P / R)2 ×E

[0033] Among them, Q represents the resolution, P represents the field of view of the image acquisition device, R represents the target detection accuracy, and E represents the minimum number of pixels occupied by the defect.

[0034] Optionally, before acquiring an initial detection image of the roller to be detected acquired by an image acquisition device at a preset detection position, the roller surface defect detection method further includes:

[0035] Obtaining the height of the starting position of the horizontal projection of the experimental roller from the image acquisition device when the experimental roller performs horizontal projection motion, and the diameter of the experimental roller;

[0036] Based on the height and the diameter, using the relationship between time and speed in free fall motion, determine a first time taken for the experimental roller to fall from the horizontal throwing starting position to just enter the field of view of the image acquisition device, and a second time taken for the experimental roller to fall from the horizontal throwing starting position to completely pass through the field of view of the image acquisition device;

[0037] determining a time difference between the first time and the second time;

[0038] The target acquisition frame number of the experimental roller by the image acquisition device is obtained, and the ratio of the target acquisition frame number to the time difference is determined as the frame rate of the image acquisition device.

[0039] Optionally, there are six image acquisition devices at the preset detection positions, one image acquisition device is deployed at each preset detection position, all the preset detection positions are distributed on the spherical surface of the same sphere, and the shooting center of each image acquisition device at the preset detection position is oriented toward the center of the sphere;

[0040] Among them, four of the image acquisition devices are evenly distributed on a horizontal plane passing through the center of the sphere; and the remaining two image acquisition devices are deployed at the top and bottom of a vertical plane passing through the center of the sphere.

[0041] An embodiment of the present invention provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes any one of the above-mentioned roller surface defect detection methods.

[0042] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes any one of the above-mentioned roller surface defect detection methods.

[0043] The roller surface defect detection method, computer device and storage medium provided by the embodiments of the present invention use an image acquisition device with a preset detection position to perform an all-round detection on the roller to be detected, collect an all-round initial detection image of the roller to be detected, and obtain a panoramic fused image of the roller to be detected by fusing the initial detection images collected by each image acquisition device at the same time, then extract the first grayscale feature of the panoramic fused image and obtain the second grayscale feature of the standard roller image, determine the Euclidean distance between the first grayscale feature and the second grayscale feature, and determine the surface defect detection result of the roller to be detected based on the Euclidean distance. In this way, the automatic detection of roller surface defects is realized by using a machine vision detection method based on computer image processing, which effectively avoids the influence of subjective factors of manual observation, thereby improving the accuracy of roller surface defect detection, and can be applicable to large-scale detection; moreover, the roller to be detected can be detected in all directions without dead ends, further improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A schematic flow chart of a roller surface defect detection method provided by an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the deployment of image acquisition equipment at six preset detection positions provided in an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of roller images captured by six cameras at the same time according to an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of the entire falling process of the experimental roller performing horizontal projection motion provided by an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of the principle of determining the frame rate of an image acquisition device by utilizing the kinematic law of free fall motion provided by an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of the working principle of the lighting method of the low-angle annular light source provided by an embodiment of the present invention;

[0051] Figure 7 A schematic diagram of the principle of fusing initial detection images acquired by various image acquisition devices at the same time provided by an embodiment of the present invention;

[0052] Figure 8 A schematic structural diagram of a roller surface defect detection device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The roller of the bearing plays the role of load-bearing and transmission in the bearing transmission and is an important component of the bearing. Among them, the surface quality of the roller directly affects the working performance and service life of the bearing. Therefore, it is of great significance to detect the surface quality of the roller.

[0055] In the related technology, the surface defects of rollers can be detected by manual naked eye observation. The manual naked eye observation method does not need to rely on other complex auxiliary equipment and has a high recognition rate for small surface defects. However, manual naked eye observation is easily affected by subjective factors and has large fluctuations, which affects the accuracy of detection. In addition, the human eye is easily fatigued and is not suitable for large-scale detection.

[0056] With the development of computer technology, machine vision technology based on computer image processing has been widely used in non-destructive testing. It has good stability and can be applied to large-scale testing. At present, when using machine vision technology to detect roller surface defects, the roller support structure is required to assist, but the support structure will block the roller, making it impossible to perform a single full-field 360° all-round inspection of the roller. If a panoramic inspection result of the roller is required, the roller needs to be moved multiple times and multiple inspections at different angles need to be performed, which greatly reduces the efficiency of the inspection.

[0057] Based on this, the embodiment of the present invention provides an all-round and efficient roller surface defect detection method. The image acquisition devices are respectively deployed at the preset detection positions to form an all-round detection of the roller to be detected. During the detection process, the roller to be detected is exposed in a space without obstruction. The deployed image acquisition devices are used to collect the initial detection images of the roller to be detected in all directions. The panoramic fusion image of the roller to be detected is obtained by fusing the initial detection images collected by each image acquisition device at the same time, and then the first grayscale feature of the panoramic fusion image is extracted and the second grayscale feature of the standard roller image is obtained. The Euclidean distance between the first grayscale feature and the second grayscale feature is determined, and the surface defect detection result of the roller to be detected is determined based on the Euclidean distance. In this way, the machine vision detection method based on computer image processing can be used to realize the automatic detection of roller surface defects through a one-time panoramic detection, which effectively avoids the influence of subjective factors of manual observation, thereby improving the accuracy of roller surface defect detection, and there is no dead angle in a single detection, and a panoramic image of the roller can be obtained in one detection, which effectively shortens the detection time of a single roller, has high detection efficiency, good robustness, and can be applied to large-scale detection.

[0058] Combine the following Figure 1 to Figure 7 The roller surface defect detection method of the present invention is described.

[0059] Figure 1 A schematic diagram of a roller surface defect detection method provided by an embodiment of the present invention, referring to Figure 1 As shown, the roller surface defect detection method may include the following steps 110 to 140:

[0060] Step 110: obtaining an initial detection image of the roller to be detected captured by an image acquisition device at a preset detection position;

[0061] Step 120: fusing the initial detection images acquired by each image acquisition device at the same time to obtain a panoramic fused image of the roller to be detected;

[0062] Step 130: extracting a first grayscale feature of the panoramic fusion image, and acquiring a second grayscale feature of the standard roller image;

[0063] Step 140: Determine the Euclidean distance between the first grayscale feature and the second grayscale feature, and determine the surface defect detection result of the roller to be detected based on the Euclidean distance.

[0064] The roller surface defect detection method provided by the embodiment of the present invention uses an image acquisition device at a preset detection position to perform an all-round detection on the roller to be detected, collects an all-round initial detection image of the roller to be detected, and obtains a panoramic fused image of the roller to be detected by fusing the initial detection images collected by each image acquisition device at the same time, then extracts the first grayscale feature of the panoramic fused image and obtains the second grayscale feature of the standard roller image, determines the Euclidean distance between the first grayscale feature and the second grayscale feature, and determines the surface defect detection result of the roller to be detected based on the Euclidean distance. In this way, the automatic detection of roller surface defects is realized by using a machine vision detection method based on computer image processing, which effectively avoids the influence of subjective factors of manual observation, thereby improving the accuracy of roller surface defect detection, and can be suitable for large-scale detection; moreover, the roller to be detected can be detected in all directions without dead ends, further improving the detection accuracy.

[0065] Steps 110 to 140 are described in detail below.

[0066] In step 110, the image acquisition device at the preset detection position performs an all-round detection of the roller to be detected.

[0067] Specifically, there are six image acquisition devices at the preset detection positions, one image acquisition device is deployed at each preset detection position, all preset detection positions are distributed on the spherical surface of the same sphere, and the shooting centers of the image acquisition devices at the preset detection positions are all oriented toward the center of the sphere. Among them, four image acquisition devices are evenly distributed on the horizontal plane passing through the center of the sphere; the remaining two image acquisition devices are deployed at the top and bottom of the vertical plane passing through the center of the sphere.

[0068] For example, Figure 2 A schematic diagram showing the deployment of image acquisition equipment at 6 preset detection positions is shown, referring to Figure 2 As shown, taking the example that all six image acquisition devices are cameras, the six cameras are evenly arranged, among which four cameras (camera 21, camera 22, camera 23 and camera 24) are distributed on the same horizontal plane, which passes through the center of the sphere, and the angle between two adjacent cameras is 90°, that is, the four cameras are evenly distributed on the horizontal plane passing through the center of the sphere; the remaining two cameras (camera 25 and camera 26) are distributed on the vertical plane passing through the center of the sphere, and the line connecting the two cameras is perpendicular to the horizontal plane passing through the center of the sphere. In this way, it can be ensured that the surface of the roller can be detected 360° without blind spots, and that the roller images acquired by two adjacent cameras have overlapping parts and the overlapping parts have the same area.

[0069] like Figure 3The schematic diagram shows the roller images captured by 6 cameras at the same time, wherein roller images 31-34 are images captured by cameras on the horizontal plane, corresponding to cameras 21-24 on the horizontal plane respectively. The diameter of the horizontal circular surface surrounded by these 4 cameras should not be too large, so as to ensure that when the camera takes the roller image, the roller image fills the camera's field of view, and the angles between these 4 cameras are the same; when observed from the direction perpendicular to the horizontal plane, the roller images 35 and 36 captured by cameras 25 and 26 on the vertical plane overlap, that is, cameras 25 and 26 are in symmetrical positions. In this way, it can be ensured that the overlapping areas of the roller images captured by the cameras are the same.

[0070] In one embodiment of the present invention, an infrared detector may be installed in the detection environment to trigger the image acquisition device to start acquiring the detection image of the roller to be detected. The infrared detector may be on the same plane as the image acquisition device on the horizontal plane. When the roller to be detected falls to the detection position of the infrared detector, the trigger of the infrared detector is triggered, and the image acquisition device starts to acquire the initial detection image of the roller to be detected.

[0071] In one embodiment of the present invention, before obtaining the initial detection image of the roller to be detected by the image acquisition device at the preset detection position in step 110, the resolution of the image acquisition device can be determined according to the field of view of the selected image acquisition device and the detection accuracy to be achieved. Specifically, the following steps 1011 to 1012 can be included:

[0072] Step 1011: Obtain target detection accuracy, field of view of the image acquisition device, and the minimum number of pixels occupied by the defect;

[0073] Step 1012: Based on the target detection accuracy, the field of view of the image acquisition device and the minimum number of pixels occupied by the defect, the resolution of the image acquisition device is determined using a resolution formula; the resolution formula may be the following formula (1):

[0074] Q=(P / R) 2 ×E (1)

[0075] Among them, Q represents the resolution, P represents the field of view of the image acquisition device, R represents the target detection accuracy, and E represents the minimum number of pixels occupied by the defect.

[0076] In step 1011, the target detection accuracy is the measurement accuracy that needs to be achieved when detecting the roller to be detected. The field of view of the image acquisition device refers to the range that can be covered by the lens of the image acquisition device, representing the maximum range that the image acquisition device can observe. The minimum number of pixels occupied by a defect refers to the minimum number of pixels occupied by the area of ​​the defect, for example, the area of ​​a defect can be at least 3 or at least 4 pixels.

[0077] For example, take the detection of a roller with a size of 15*15mm and the image acquisition device is a camera. Assuming that the target detection accuracy is 0.01mm and the camera's field of view is 18*18mm, the minimum resolution of the camera should be: (18 / 0.01)*(18 / 0.01)=1800*1800=3240000. If one pixel corresponds to one defect to be detected, the minimum resolution of the camera can be 3.2 million pixels. However, if one pixel is selected to correspond to one defect, the system will be extremely unstable, because any interfering pixel point may be mistaken for a defect. Therefore, in order to improve the accuracy and stability of the system, the area of ​​the defect can be taken to be more than 3 to 4 pixels. At this time, the camera resolution needs to be more than 320*3=9.6 million pixels, that is, the minimum cannot be less than 9.6 million pixels. For example, a 10 million pixel camera can be used.

[0078] In one embodiment of the present invention, in order to detect the roller to be detected by high-precision non-contact measurement, a camera can be used to capture the surface features of the roller to be detected in a high-speed motion state. The core of this method is that in order to avoid the collision between the falling roller and the image acquisition device on the vertical plane, an auxiliary tool such as a circular arc guide groove can be used to induce the roller to do a horizontal projection motion, that is, to give the roller a certain initial velocity in the horizontal direction, and at the same time, the roller is only subjected to the action of gravity in the vertical direction and performs a free fall motion. In this scenario, the continuous image sequence of the roller falling process can be accurately recorded by using the camera's shooting ability, and the instantaneous speed within a certain time interval can be calculated according to the speed-time relationship of the free fall motion. On this basis, in order to match the change in the roller's falling speed and ensure that the required detection accuracy can be achieved, the minimum frame rate required by the image acquisition device can be determined according to the change in the roller's falling speed. By selecting a suitable frame rate, it can be ensured that there are enough image frames to smoothly describe its motion trajectory in the fastest stage of the roller's fall, thereby avoiding the speed calculation error caused by too large an interval between frames.

[0079] Specifically, before obtaining the initial detection image of the roller to be detected by the image acquisition device at the preset detection position in step 110, the frame rate of the image acquisition device when acquiring the roller image can be determined by experiment using the kinematic law of free fall motion. Specifically, the roller surface defect detection method can also include the following steps 1021 to 1024:

[0080] Step 1021: obtaining the height of the starting position of the horizontal projection of the experimental roller from the image acquisition device when the horizontal projection motion is performed, and the diameter of the experimental roller;

[0081] Step 1022: Based on the height and the diameter, using the relationship between time and speed in free fall motion, determine the first time taken for the test roller to fall from the starting position of the horizontal projection to just enter the field of view of the image acquisition device, and the second time taken for the test roller to fall from the starting position of the horizontal projection to completely pass through the field of view of the image acquisition device;

[0082] Step 1023: Determine the time difference between the first time and the second time;

[0083] Step 1024: Obtain the target acquisition frame number of the experimental roller by the image acquisition device, and determine the ratio of the target acquisition frame number to the time difference as the frame rate of the image acquisition device.

[0084] Specifically, Figure 4 The schematic diagram of the whole process of the experimental roller falling in horizontal projection motion is shown. Figure 4 As shown, the experimental roller 40 moves along the guide groove from the P1 position to the P2 position and is thrown horizontally to start free fall motion. The P2 position is the starting position of the horizontal throw, and finally lands on the x-plane.

[0085] Combination Figure 4 , Figure 5 The schematic diagram shows the principle of determining the frame rate of the image acquisition device using the kinematic law of free fall motion, referring to Figure 5 As shown, the experimental roller falls from the horizontal throwing starting position A at a height h from the image acquisition device. According to the motion law of free fall motion, the first time t1 taken by the experimental roller to reach the position B just entering the field of view of the image acquisition device can be determined by the following formula (2):

[0086]

[0087] Here, g represents the acceleration due to gravity.

[0088] According to the law of free fall, the second time t2 taken by the experimental roller to fall from the horizontal throwing starting position A to the position C when it completely passes through the field of view of the image acquisition device can be determined by the following formula (3):

[0089]

[0090] Here, g represents the acceleration due to gravity.

[0091] The time taken by the experimental roller to pass through the image acquisition device is the time difference Δt between the first time t1 and the second time t2, which can be determined by the following formula (4):

[0092]

[0093] Assuming that the image acquisition device captures at least n frames of image within the time Δt, that is, the target acquisition frame number is n, the frame rate required by the image acquisition device can be determined by the following formula (5):

[0094]

[0095] Wherein, f represents the frame rate of the image acquisition device.

[0096] Since the rollers of bearings are usually made of metal, and the surface of metal rollers is smooth, it has a high reflectivity, almost close to a mirror surface. When light hits the metal surface, most of the light will be reflected, forming strong reflected light. This reflected light may not only form bright spots, but also reflect the images of surrounding objects, seriously affecting the accuracy of the detection results.

[0097] In view of this, in one embodiment of the present invention, a red light emitting diode (LED) diffuse light source can be selected. The wavelength of red light is longer, and compared with other colors of light, it is less likely to be reflected by the metal surface to form a strong bright spot. Moreover, the light of the diffuse light source is evenly scattered in all directions, rather than concentrated in a certain direction like mirror reflection. This scattered light can weaken the reflected light of the metal surface, making the detection target clearer.

[0098] In the embodiment of the present invention, the roller can move at high speed. When collecting images of the roller moving at high speed, it is easy to produce a smear. In view of this, in one embodiment, the light source deployed in the detection environment can be a low-angle ring-shaped high-brightness strobe light. The low-angle ring light source can enhance the texture or edge of the object surface, so that the image processing system can more easily identify subtle surface defects. The high-brightness strobe light source can provide extremely high brightness illumination in a very short time to freeze the moving object.

[0099] Specifically, Figure 6 A schematic diagram showing the working principle of the low-angle ring light source is shown. Figure 6 As shown, the LED light of this lighting method is at a 30° angle to the shooting plane, and can enhance the surface features of the object by projecting light in an oblique direction.

[0100] In step 120, the image acquisition devices at the preset detection positions can start detection at the same time, and perform image acquisition on the roller to be detected to obtain respective acquired initial detection images. When analyzing the surface defect situation of the roller to be detected based on these acquired initial detection images, since all image acquisition devices can complete the full-scale detection of the roller to be detected at the same time, all initial detection images acquired by each image acquisition device at the same time can reflect the surface defect situation of the roller to be detected, and these initial detection images at the same time can be fused together, and the surface defect situation of the roller to be detected can be obtained by further analyzing the panoramic fusion image obtained by fusion.

[0101] In one embodiment, step 120 fuses the initial detection images acquired by each image acquisition device at the same time to obtain a panoramic fused image of the roller to be detected, which may include the following sub-steps:

[0102] Taking each detection image in the initial detection images collected by each image acquisition device at the same time as a first detection image, determining a first inscribed square area of ​​the roller image in the first detection image and a second inscribed square area of ​​the roller image in the second detection image to be fused; wherein the first detection image and the second detection image are initial detection images collected by two adjacent image acquisition devices at the same time;

[0103] Superimposing the first detection image and the second detection image by overlapping adjacent edges between the first inscribed square area and the second inscribed square area to obtain a superimposed image;

[0104] Determine a spatial coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image;

[0105] The first detection image and the second detection image are converted to the same coordinate system according to the spatial coordinate conversion matrix, and the overlapping areas in the superimposed images are weightedly fused in the same coordinate system;

[0106] The above process is repeated until all the initial detection images at that moment are traversed, and the finally obtained fused image is determined as the panoramic fused image of the roller to be detected.

[0107] Specifically, Figure 7 The schematic diagram shows the principle of fusing the initial detection images collected by each image acquisition device at the same time, referring to Figure 7 As shown, combined Figure 2 and Figure 3Taking the fusion of roller images 31 and 35 in the initial detection images captured by adjacent cameras 21 and 25 as an example, the first inscribed square area 75 of the roller image 35 and the second inscribed square area 71 of the roller image 31 can be determined first, and the roller image 35 and the roller image 31 are superimposed by overlapping the adjacent edges of the first inscribed square area 75 and the second inscribed square area 71 to obtain Figure 7 The first inscribed square area 75 and the second inscribed square area 71 in the superimposed image both have overlapping areas with their respective adjacent roller images, and similar feature point pairs between the two roller images can be determined from the overlapping areas. The roller images 31 and 35 are converted to the same coordinate system using the correspondence between the two feature points in the feature point pairs, and the roller images 31 and 35 are fused in the same coordinate system.

[0108] Specifically, determining the space coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image may include:

[0109] Traversing each first pixel point of the first inscribed square area in the superimposed image, and determining the first pixel point as a first feature point when the brightness value of the first pixel point is greater than a brightness threshold;

[0110] Determine, within the second inscribed square region of the superimposed image, a second feature point on the target straight line that is most symmetrical to the first feature point with the common side as the axis of symmetry; wherein the common side is the common side of the first inscribed square region and the second inscribed square region in the superimposed image, and the target straight line is a pixel straight line that passes through the first feature point and is perpendicular to the common side;

[0111] The first feature point and the second feature point are taken as a pair of feature point pairs, and based on the feature point pairs, the spatial coordinate conversion matrix between the first detection image and the second detection image is determined using the affine transformation principle.

[0112] For example, combined with Figure 7, each pixel point in the first inscribed square area 75 can be traversed, and the first pixel point M1 can be taken as an example. If the brightness value of the first pixel point M1 is greater than the brightness threshold, it indicates that the first pixel point M1 is a pixel point with a relatively large brightness, that is, it indicates that it is a pixel point in the overlapping area between the first inscribed square area 75 and the roller image 31, and the first pixel point M1 is determined as the first feature point. Then, taking the common side F1F2 of the first inscribed square area 75 and the second inscribed square area 71 as the axis of symmetry, a similar feature point that is in the second inscribed square area 71 and is most symmetrical to the first feature point M1 is searched on a pixel straight line (i.e., a target straight line) passing through the first feature point M1 and perpendicular to the common side F1F2, and a second feature point is obtained, such as determining the second feature point M2. Specifically, the second feature point can be found using an image matching algorithm. For example, the similarity between each pixel point on the target straight line within the second inscribed square area 71 and the first feature point M1 can be calculated, and the pixel point on the target straight line that has a similarity greater than a preset similarity threshold and is most symmetrical to the first feature point M1 with the common edge F1F2 as the axis of symmetry is determined as the matching second feature point M2.

[0113] Assuming that the coordinates of the extracted first feature point M1 are (x1, y1, z1), and the coordinates of the matched second feature point M2 are (x2, y2, z2), the spatial coordinate transformation matrix between the first detection image and the second detection image can be determined according to the following formula (6) by using the affine transformation principle, that is, the radial transformation matrix. The spatial coordinate transformation matrix includes the rotation matrix R and the translation matrix T. Formula (6) is:

[0114]

[0115] After the spatial coordinate conversion matrix is ​​obtained, the first detection image and the second detection image in the three-dimensional space can be converted to the same coordinate system according to the spatial coordinate conversion matrix to complete the unified coordinate transformation.

[0116] Based on the same coordinate system, a weighted fusion algorithm can be used to perform weighted fusion on the overlapping areas in the superimposed images, so that the roller image 31 and the roller image 35 can be smoothly fused. Specifically, for the overlapping area between the roller image 31 and the roller image 35, each pixel position in the overlapping area can be traversed, and the pixel value U1 of the pixel position in the roller image 31 and the pixel value U2 of the pixel position in the roller image 35 can be taken respectively, and then the pixel value U1 and the pixel value U2 are weighted averaged to obtain the fused pixel value. In this way, the pixel values ​​in the overlapping area will gradually transition, thereby achieving smooth fusion.

[0117] The above fusion process is performed on the roller images 31 to 36, and finally a panoramic fusion image of the roller to be detected can be obtained.

[0118] In one embodiment, before the initial detection images acquired by each image acquisition device at the same time are fused in step 120, the initial detection images may be binarized first, or the initial detection images may be denoised and binarized in sequence, and then the processed initial detection images may be fused.

[0119] For example, the initial detection image can be denoised by using a median filtering method. Specifically, for each pixel in the initial detection image, the pixel values ​​in a window of a preset size in its neighborhood can be sorted, and then the original pixel value of the pixel can be replaced with the sorted median value, thereby achieving the purpose of denoising. For example, a window of an odd size can be selected, all pixels in the window can be sorted by grayscale value, the grayscale value of the pixel in the center of the window can be replaced with the median value obtained above, and then the window can be moved to the next pixel position, and the above steps can be repeated until all pixels in the initial detection image are processed.

[0120] For the binarization process, the method of automatically determining the grayscale threshold can be used to implement the binarization process of the initial detection image. Specifically, the grayscale value of each pixel in the initial detection image can be compared with the preset grayscale threshold. If the grayscale value of the pixel is greater than the preset grayscale threshold, the pixel is determined as the foreground. If the grayscale value of the pixel is less than or equal to the preset grayscale threshold, the pixel is determined as the background. In this way, the binarization of the initial detection image can be implemented. Among them, the preset grayscale threshold can be an optimal grayscale threshold determined by an automatic threshold selection algorithm, and the automatic threshold selection algorithm is, for example, the maximum between-class variance (OTSU) algorithm.

[0121] In step 130, the first grayscale feature may include a first grayscale mean and a first grayscale standard deviation, and the second grayscale feature may include a second grayscale mean and a second grayscale standard deviation. Alternatively, the first grayscale feature may include a first grayscale mean, a first grayscale standard deviation, and a first grayscale histogram, and the second grayscale feature may include a second grayscale mean, a second grayscale standard deviation, and a second grayscale histogram. The standard roller image is an image of a standard roller without defects.

[0122] Among them, the grayscale mean can reflect the level of grayscale value on the roller surface. Each pixel in the image can be traversed to extract its grayscale value, and the grayscale values ​​of all pixels can be accumulated during the traversal process. Finally, the accumulated total grayscale value is divided by the total number of pixels to obtain the grayscale mean.

[0123] The grayscale standard deviation can measure the degree of dispersion of the grayscale value of the roller surface. The smaller the grayscale standard deviation, the more uniform the grayscale distribution, and the more likely the roller surface is to be defect-free. Specifically, assuming that the size of an image is M×N, and the grayscale value of the pixel in the i-th row and j-th column is f(i, j), the grayscale standard deviation s of the image can be determined according to the following formula (7):

[0124]

[0125] in, is the grayscale mean of the image.

[0126] The grayscale histogram can intuitively display the distribution of grayscale values ​​on the roller surface. A defect-free roller usually has a relatively smooth and continuous grayscale histogram.

[0127] For the roller to be detected, the grayscale value of each pixel in the panoramic fusion image can be extracted, and the average value of these grayscale values ​​can be calculated to obtain a first grayscale mean value. Based on the first grayscale mean value, the corresponding first grayscale standard deviation can be determined using the above formula (7). In addition, the grayscale values ​​of the panoramic fusion image can be subjected to grayscale statistics to obtain a first grayscale histogram.

[0128] For a standard roller, a grayscale surface of the standard roller image can be constructed under the same lighting environment as the roller to be detected, and key features with scale invariance such as the second grayscale mean, the second grayscale standard deviation and the second grayscale histogram can be extracted from the constructed grayscale surface.

[0129] Specifically, a large number of standard roller images of standard defect-free rollers can be collected as sample images under different lighting conditions, shooting angles and backgrounds when the standard roller is stationary, and then grayscale histogram analysis is performed on these sample images to understand the grayscale distribution characteristics of the standard roller through the grayscale histogram. According to the grayscale histogram analysis, an imaging model that describes the grayscale characteristics of the standard roller using a probability distribution function of grayscale values ​​or a template based on image features is established to obtain a standard roller imaging model. When constructing a grayscale surface, the grayscale image of the standard roller image can be regarded as a grayscale value distribution on a three-dimensional plane, where the x-axis and y-axis represent image coordinates and the z-axis represents grayscale values. In order to construct a smooth grayscale surface of the standard roller image, the grayscale image of the standard roller image can be first filtered based on the standard roller imaging model to reduce the influence of noise on the surface construction, and then the filtered grayscale values ​​are surface fitted using a polynomial fitting method to obtain a grayscale surface of the standard roller image.

[0130] In one embodiment, in step 140, the first grayscale feature includes a first grayscale mean and a first grayscale standard deviation, and the second grayscale feature includes a second grayscale mean and a second grayscale standard deviation. Step 140 determines the Euclidean distance between the first grayscale feature and the second grayscale feature, and determines the surface defect detection result of the roller to be detected based on the Euclidean distance, which may include the following steps 141 to 144.

[0131] Step 141: construct the first grayscale mean and the first grayscale standard deviation into a first eigenvector, and construct the second grayscale mean and the second grayscale standard deviation into a second eigenvector.

[0132] Step 142: Determine the Euclidean distance between the first eigenvector and the second eigenvector using the Euclidean distance formula of the vector.

[0133] Specifically, the Euclidean distance between the first eigenvector and the second eigenvector can be determined according to the following Euclidean distance formula (8):

[0134]

[0135] in, Represents the first eigenvector of n dimensions and the second eigenvector of n dimension The Euclidean distance, a i and b i express and The value in the i-th dimension.

[0136] Step 143: When the Euclidean distance is greater than the preset distance threshold, it is determined that the surface defect detection result of the roller to be detected is that a surface defect exists.

[0137] Step 144: when the Euclidean distance is less than or equal to the preset distance threshold, determining that the surface defect detection result of the roller to be detected is no surface defect.

[0138] The Euclidean distance between the first eigenvector and the second eigenvector reflects the distance between the two vectors. The first eigenvector represents the panoramic fusion image, and the second eigenvector represents the standard roller image. When the Euclidean distance between the two vectors is smaller, it means that the panoramic fusion image and the standard roller image are more similar, and the roller to be detected is closer to the standard roller; conversely, when the Euclidean distance is larger, it means that the difference between the panoramic fusion image and the standard roller image is greater, and the probability that the roller to be detected has surface defects is greater. Based on this, a preset distance threshold can be set to determine which differences can be regarded as real defects. The preset distance threshold can be determined based on experience or by experiment.

[0139] For example, the preset distance threshold can be set by calculating a certain percentile (such as the 95% percentile) of the second eigenvector corresponding to the standard roller, which means that only 5% of the second eigenvectors will exceed the preset distance threshold, thereby ensuring that the rollers that are successfully matched have a high degree of confidence.

[0140] If the Euclidean distance is greater than the preset distance threshold, it means that there is a large difference between the surface features of the roller to be detected and the surface features of the standard roller, and it can be determined that the roller to be detected has surface defects; if the Euclidean distance is less than or equal to the preset distance threshold, it means that the surface features of the roller to be detected are similar to the surface features of the standard roller, and it can be determined that the roller to be detected has no surface defects.

[0141] In one embodiment, the first grayscale feature may include a first grayscale mean, a first grayscale standard deviation, and a first grayscale histogram, and the second grayscale feature may include a second grayscale mean, a second grayscale standard deviation, and a second grayscale histogram. Accordingly, after using the Euclidean distance to identify whether the roller to be inspected has a surface defect, the position of the defect can be further located by comparing the first grayscale histogram with the second grayscale histogram. Specifically, the roller surface defect detection method may also include the following steps 145 to 146.

[0142] Step 145: When it is determined that the surface defect detection result of the roller to be detected is that there is a surface defect, the first grayscale histogram is compared with the second grayscale histogram, and the target grayscale value of the mutation in the first grayscale histogram is determined by comparison.

[0143] The grayscale histogram can intuitively display the distribution of grayscale values ​​on the roller surface. For a defect-free standard roller, the second grayscale histogram of its standard roller image is relatively smooth and continuous. By comparing the first grayscale histogram of the roller to be tested with the second grayscale histogram of the standard roller, the position where the mutation occurs in the first grayscale histogram can be intuitively detected, and the grayscale value of the position can be obtained from the first grayscale histogram to obtain the target grayscale value.

[0144] Step 146: Obtain the regional position of the target grayscale value in the panoramic fusion image, and determine the regional position as the defect position of the roller to be detected.

[0145] After determining the target grayscale value, the regional position of the target grayscale value in the panoramic fusion image can be determined. The determined regional position is the position that is significantly different from the standard roller surface, thereby locating the defect position of the roller to be inspected.

[0146] Specifically, the surface defect detection result of the roller to be detected may include the defect position, so that not only can it be detected whether there is a defect on the surface of the roller to be detected, but also the position of the surface defect can be located when the defect exists.

[0147] The roller surface defect detection method provided by the embodiment of the present invention uses an image acquisition device with a preset detection position to perform an all-round detection on the roller to be detected. First, an all-round initial detection image of the roller to be detected is acquired, and the acquired initial detection image is denoised and binarized to obtain a denoised grayscale image; then the inscribed square area of ​​each grayscale image is taken and the high-brightness feature points in the inscribed square area are extracted, and a matching algorithm is used on adjacent grayscale images to find the feature points most similar to the high-brightness feature points to form feature point pairs; then the spatial coordinate transformation matrix is ​​determined according to the corresponding relationship between the feature point pairs, and the two adjacent grayscale images are aligned to the same coordinate system using the spatial coordinate transformation matrix; then, an image fusion algorithm is used in the same coordinate system to form a natural panoramic fusion image of the roller to be detected without obvious splicing marks; the first grayscale feature of the panoramic fusion image is extracted, and the second grayscale feature is extracted from the grayscale surface of the constructed standard roller image; a matching operator is constructed based on the first grayscale feature and the second grayscale feature to compare the image difference between the roller to be detected and the standard roller, and the surface defects of the roller to be detected are identified according to the image difference. Compared with the manual observation method, this method greatly improves the detection accuracy, reduces the error of surface defect detection, and can be applied to large-scale detection. Moreover, in the process of matching feature point pairs, the method of finding similar feature points on the symmetric lines of adjacent grayscale images is used to improve the success rate and accuracy of feature point pair matching.

[0148] Based on the same inventive concept, an embodiment of the present invention provides a roller surface defect detection device. Since the principle of solving the technical problem by the device is similar to that of the roller surface defect detection method, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0149] like Figure 8 As shown, the roller surface defect detection device includes an acquisition unit 801, a fusion unit 802, an extraction unit 803 and a determination unit 804.

[0150] The acquisition unit 801 is used to acquire an initial detection image of the roller to be detected acquired by an image acquisition device at a preset detection position, wherein the image acquisition device at the preset detection position forms an all-round detection of the roller to be detected;

[0151] A fusion unit 802 is used to fuse the initial detection images collected by each image acquisition device at the same time to obtain a panoramic fusion image of the roller to be detected;

[0152] An extraction unit 803 is used to extract a first grayscale feature of the panoramic fusion image and obtain a second grayscale feature of the standard roller image;

[0153] The determination unit 804 is used to determine the Euclidean distance between the first grayscale feature and the second grayscale feature, and determine the surface defect detection result of the roller to be detected based on the Euclidean distance.

[0154] It should be understood that the units included in the above roller surface defect detection device are only logical divisions based on the functions implemented by the device. In practical applications, the above units can be superimposed or split. In addition, the functions implemented by the roller surface defect detection device provided in this embodiment correspond one-to-one to the roller surface defect detection method provided in the above embodiment. The more detailed processing flow implemented by the roller surface defect detection device has been described in detail in the above method embodiment 1 and will not be described in detail here.

[0155] Another embodiment of the present invention also provides a computer device, which includes: a processor and a memory; the memory is used to store computer program code, and the computer program code includes computer instructions; when the processor executes the computer instructions, the computer device executes each step of roller surface defect detection in the method flow shown in the above method embodiment.

[0156] Another embodiment of the present invention further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer device, the computer device executes each step of roller surface defect detection in the method flow shown in the above method embodiment.

[0157] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0158] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A roller surface defect detection method, characterized in that: include: Acquire an initial detection image of the roller to be detected captured by an image acquisition device at a preset detection position; wherein the image acquisition device at the preset detection position forms an all-round detection of the roller to be detected; Fusing the initial detection images acquired by the image acquisition devices at the same time to obtain a panoramic fused image of the roller to be detected; Extracting a first grayscale feature of the panoramic fusion image, and acquiring a second grayscale feature of the standard roller image; The Euclidean distance between the first grayscale feature and the second grayscale feature is determined, and a surface defect detection result of the roller to be detected is determined based on the Euclidean distance.

2. The roller surface defect detection method according to claim 1, characterized in that: The fusing of the initial detection images acquired by the image acquisition devices at the same time to obtain a panoramic fused image of the roller to be detected includes: Taking each detection image in the initial detection images captured by each of the image acquisition devices at the same time as a first detection image, determining a first inscribed square area of ​​the roller image in the first detection image and a second inscribed square area of ​​the roller image in the second detection image to be fused; wherein the first detection image and the second detection image are the initial detection images captured by two adjacent image acquisition devices at the same time; Superimposing the first detection image and the second detection image by overlapping adjacent edges between the first inscribed square area and the second inscribed square area to obtain a superimposed image; Determine a spatial coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image; The first detection image and the second detection image are converted to the same coordinate system according to the spatial coordinate conversion matrix, and the overlapping areas in the superimposed images are weightedly fused in the same coordinate system; The above process is repeated until all the initial detection images are traversed, and the finally obtained fused image is determined as the panoramic fused image of the roller to be detected.

3. The roller surface defect detection method according to claim 2, characterized in that: The determining of a spatial coordinate transformation matrix between the first detection image and the second detection image based on the first inscribed square area and the second inscribed square area in the superimposed image comprises: Traversing each first pixel point of the first inscribed square area in the superimposed image, and determining the first pixel point as a first feature point when a brightness value of the first pixel point is greater than a brightness threshold; Determine, within the second inscribed square area of ​​the superimposed image, a second feature point on a target straight line that is most symmetrical with the first feature point with a common side as a symmetry axis; wherein the common side is a common side of the first inscribed square area and the second inscribed square area in the superimposed image, and the target straight line is a pixel straight line that passes through the first feature point and is perpendicular to the common side; The first feature point and the second feature point are taken as a pair of feature point pairs, and based on the feature point pairs, a spatial coordinate transformation matrix between the first detection image and the second detection image is determined using an affine transformation principle.

4. The roller surface defect detection method according to claim 1, characterized in that: The first grayscale feature includes a first grayscale mean and a first grayscale standard deviation, and the second grayscale feature includes a second grayscale mean and a second grayscale standard deviation; The determining of the Euclidean distance between the first grayscale feature and the second grayscale feature, and determining the surface defect detection result of the roller to be detected based on the Euclidean distance, comprises: The first grayscale mean and the first grayscale standard deviation are constructed as a first eigenvector, and the second grayscale mean and the second grayscale standard deviation are constructed as a second eigenvector; Determine the Euclidean distance between the first eigenvector and the second eigenvector using the Euclidean distance formula of vectors; When the Euclidean distance is greater than a preset distance threshold, determining that the surface defect detection result of the roller to be detected is that a surface defect exists; When the Euclidean distance is less than or equal to the preset distance threshold, it is determined that the surface defect detection result of the roller to be detected is that there is no surface defect.

5. The roller surface defect detection method according to claim 4, characterized in that: The first grayscale feature further includes a first grayscale histogram, and the second grayscale feature further includes a second grayscale histogram; the roller surface defect detection method further includes: When it is determined that the surface defect detection result of the roller to be detected is that there is a surface defect, the first grayscale histogram is compared with the second grayscale histogram, and a target grayscale value of a sudden change in the first grayscale histogram is determined by comparison; The regional position of the target grayscale value in the panoramic fusion image is obtained, and the regional position is determined as the defect position of the roller to be detected; and the surface defect detection result of the roller to be detected includes the defect position.

6. The roller surface defect detection method according to claim 1, characterized in that: Before acquiring the initial detection image of the roller to be detected acquired by the image acquisition device at the preset detection position, the roller surface defect detection method further includes: Obtaining target detection accuracy, the field of view of the image acquisition device, and the minimum number of pixels occupied by the defect; Based on the target detection accuracy, the field of view of the image acquisition device and the minimum number of pixels occupied by the defect, the resolution of the image acquisition device is determined using a resolution formula; the resolution formula is: Q=(P / R) 2 ×E Among them, Q represents the resolution, P represents the field of view of the image acquisition device, R represents the target detection accuracy, and E represents the minimum number of pixels occupied by the defect.

7. The roller surface defect detection method according to claim 1, characterized in that: Before acquiring the initial detection image of the roller to be detected acquired by the image acquisition device at the preset detection position, the roller surface defect detection method further includes: Obtaining the height of the starting position of the horizontal projection of the experimental roller from the image acquisition device when the experimental roller performs horizontal projection motion, and the diameter of the experimental roller; Based on the height and the diameter, using the relationship between time and speed in free fall motion, determine a first time taken for the experimental roller to fall from the horizontal throwing starting position to just enter the field of view of the image acquisition device, and a second time taken for the experimental roller to fall from the horizontal throwing starting position to completely pass through the field of view of the image acquisition device; determining a time difference between the first time and the second time; The target acquisition frame number of the experimental roller by the image acquisition device is obtained, and the ratio of the target acquisition frame number to the time difference is determined as the frame rate of the image acquisition device.

8. The roller surface defect detection method according to any one of claims 1 to 7, characterized in that: There are six image acquisition devices at the preset detection positions, one image acquisition device is deployed at each preset detection position, all the preset detection positions are distributed on the spherical surface of the same sphere, and the shooting center of each image acquisition device at the preset detection position is oriented toward the center of the sphere; Among them, four of the image acquisition devices are evenly distributed on a horizontal plane passing through the center of the sphere; and the remaining two image acquisition devices are deployed at the top and bottom of a vertical plane passing through the center of the sphere.

9. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the roller surface defect detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the roller surface defect detection method according to any one of claims 1 to 7.

Images