A method for identifying surface defects of motor bearings based on image processing

Abstract

The present invention relates to the field of image processing technology, and more specifically, to a method for identifying surface defects of motor bearings based on image processing. The method comprises: collecting each frame of bearing images, performing superpixel segmentation on the images, obtaining each superpixel block of each frame of bearing images, and obtaining characteristic values ​​of each superpixel block of each frame of bearing images according to the grayscale characteristics of each superpixel block; correcting the characteristic values ​​of each superpixel block to obtain corrected characteristic values ​​of each superpixel block of each frame of bearing images, obtaining the probability that each target area of ​​each frame of bearing images is a crack area according to the area change of the target superpixel block of each frame of bearing images in its adjacent frame of bearing images, and obtaining crack defects of each frame of bearing images according to the probability that each target superpixel block of each frame of bearing images is a crack area. The present invention obtains crack defects more accurately.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more particularly to a method for identifying surface defects of motor bearings based on image processing. Background Art

[0002] When a motor is running for an extended period, the rolling elements and raceways of the bearings are constantly subjected to alternating loads. This repeated compression and friction gradually fatigues the bearing's metal material. Microscopically, dislocations and slips occur in the metal's lattice structure. Over time, these tiny damages accumulate, and when they exceed the material's fatigue limit, microcracks develop. For example, the bearings of some large motors that operate continuously are more susceptible to fatigue cracks due to long periods of alternating stress. These cracks damage the bearing's smoothness, increasing friction between the rolling elements and raceways, and generating even more heat. If this heat cannot be dissipated promptly, the bearing temperature continues to rise, further deteriorating the bearing's operating conditions and potentially leading to bearing burnout. Therefore, crack detection on motor bearings is necessary to determine whether the bearings need replacement.

[0003] When collecting each frame of bearing images, the influence of lighting will cause the existence of light shadow areas in each frame of bearing images. The grayscale value of the light shadow area is similar to the crack area in the bearing, and the grayscale value in the crack area is also different. Therefore, the threshold segmentation algorithm or edge detection algorithm cannot accurately segment the crack area in each frame of bearing images. Summary of the Invention

[0004] In order to solve the problem that the grayscale values ​​of the illuminated shadow area and the crack area of ​​the bearing are similar, and the grayscale values ​​in the crack area are also different, and the crack area in the bearing image cannot be accurately segmented using a threshold segmentation algorithm or an edge detection algorithm, the present invention proposes a motor bearing surface defect recognition method based on image processing, which includes the following steps:

[0005] Collect each frame of bearing image and perform superpixel segmentation to obtain each superpixel block of each frame of bearing image; record any frame of bearing image as the current frame of bearing image; obtain the corrected feature value of each superpixel block of the current frame of bearing image:

[0006] , is the corrected eigenvalue of the i-th superpixel block; is the variance of the gradient direction of all edge pixels of the i-th superpixel block; as well as is the eigenvalue of the i-th superpixel block and the number of reference superpixel blocks; as well as is the grayscale mean of all reference superpixel blocks of the i-th superpixel block and the grayscale mean of the a-th reference superpixel block; as well as Represent the gradient trend angle and trend angle of the i-th super pixel block respectively; exp() represents an exponential function with a natural constant as the base; || represents the absolute value symbol; according to the modified eigenvalue, obtain the target super pixel block of the current frame bearing image;

[0007] Get the probability that the cth target superpixel block in the current frame bearing image is a crack area:

[0008] ; is the area variance between the cth target superpixel block and its reference area; as well as is the area of ​​the cth target superpixel block and the area of ​​the dth reference area of ​​the cth target superpixel block; D is the number of reference areas of the cth target superpixel block; based on the probability, the crack defect of each frame of the bearing image is obtained.

[0009] The innovation of the present invention lies in dividing each frame of bearing image into various super-pixel blocks, obtaining the characteristic values ​​of each super-pixel block of each frame of bearing image according to the grayscale characteristics of each illuminated shadow area and crack area, and then correcting the characteristic values ​​according to the contrast and direction characteristics of the crack area and the normal area around it to obtain the corrected characteristic values ​​of each super-pixel block of each frame of bearing image, and then merging the super-pixel blocks to merge the crack area as completely as possible, thereby solving the incomplete segmentation caused by the different grayscales of the crack area. Next, according to the area change of the target super-pixel block of each frame of bearing image in its adjacent frame of bearing image, the probability that each target super-pixel block of each frame of bearing image is a crack area is obtained, and the accurate crack area is identified, avoiding the influence of illumination.

[0010] Preferably, obtaining the characteristic value of the i-th super pixel block includes:

[0011] ;

[0012] Where, Represents the feature value of the i-th superpixel block of the pin axis image of the current frame; as well as Represents the gradient mean of the pixel points on the edge of the i-th super-pixel block of the pin axis image of the current frame and the grayscale mean of the i-th super-pixel block; as well as Represents the gradient mean and grayscale mean in the current pin image; norm() represents the normalization function;

[0013] It is convenient to distinguish the crack area from the normal area based on the characteristic value.

[0014] Preferably, the step of obtaining each superpixel block of each frame of the bearing image includes:

[0015] The number of super-pixel blocks N is preset, and each frame of the bearing image is segmented into super-pixels using a super-pixel segmentation algorithm to obtain a number of super-pixel blocks.

[0016] This facilitates the subsequent merging and analysis of superpixel blocks.

[0017] Preferably, obtaining the reference superpixel block of the i-th superpixel block includes:

[0018] A preset difference threshold T is used to obtain the adjacent super-pixel blocks of the i-th super-pixel block of the current frame bearing image. If the absolute value of the difference between the characteristic values ​​of the i-th super-pixel block of the current frame bearing image and its j-th adjacent super-pixel block is greater than or equal to the difference threshold T, the j-th adjacent super-pixel block of the i-th super-pixel block of the current frame bearing image is recorded as the reference pixel block of the i-th super-pixel block of the current frame bearing image.

[0019] It is convenient to correct the characteristic value of the super pixel block according to the grayscale contrast between the super pixel block and its reference pixel block.

[0020] Preferably, the acquisition of the strike angle includes:

[0021] Obtain similar pixel blocks of the i-th superpixel block of the current frame bearing image, merge the i-th superpixel block of the current frame bearing image with its similar pixel blocks to obtain the target area, use morphological processing methods to extract the skeleton of the target area, use the least squares method to fit the pixel points on the skeleton of the target area into a straight line, and obtain the angle between the straight line and the horizontal axis as the strike angle of the i-th superpixel block of the current frame bearing image.

[0022] Preferably, obtaining the gradient trend angle of the i-th superpixel block includes:

[0023] Get the gradient directions of all edge pixels of the i-th superpixel block of the current frame bearing image, and take the angle between the gradient direction with the highest number of occurrences and the horizontal axis as the gradient direction angle of the i-th superpixel block of the current frame bearing image.

[0024] Preferably, obtaining the target superpixel block of the bearing image of the current frame includes:

[0025] A preset eigenvalue threshold T1 is set. If the corrected eigenvalue of any superpixel block of the current frame bearing image is greater than the eigenvalue threshold T1, the superpixel block is recorded as a representative superpixel block, and all adjacent representative superpixel blocks are merged and recorded as the target superpixel block of the current frame bearing image.

[0026] Preferably, obtaining the reference area of ​​the c-th target superpixel block includes:

[0027] The r frame bearing images before the current frame bearing image and the r frame bearing images after the current frame bearing image are recorded as adjacent frame bearing images of the current frame bearing image, the central pixel point of the cth target superpixel block of the current frame bearing image is corresponded to any of its adjacent frame bearing images, and the target superpixel block or superpixel block where the central pixel point is located in the adjacent frame bearing image is recorded as a reference area of ​​the cth target superpixel block of the current frame bearing image, and several reference areas of the cth target superpixel block of the current frame bearing image are obtained.

[0028] It is convenient to subsequently obtain the probability that each target superpixel block in each frame of the bearing surface image is a crack area based on the area change of the target superpixel block and its reference area in each frame of the bearing surface image, reflecting the area change of the target superpixel block in its adjacent frame pin shaft image.

[0029] Preferably, collecting each frame of bearing image includes:

[0030] The bearing to be tested on the motor is rotated one circle, and a video of the bearing rotation is captured using a camera, which is recorded as a bearing video. The bearing video is frame-processed to obtain an RGB image of each frame of the bearing. Each frame of the bearing RGB image is grayscaled and recorded as a bearing image of each frame.

[0031] The present invention has the following beneficial effects: the purpose of the present invention is to divide each frame of bearing image into various super-pixel blocks, and obtain the characteristic values ​​of each super-pixel block of each frame of bearing image according to the grayscale characteristics of each illuminated shadow area and crack area, and then correct the characteristic values ​​according to the contrast and direction characteristics of the crack area and its surrounding normal area to obtain the corrected characteristic values ​​of each super-pixel block of each frame of bearing image, and then merge the super-pixel blocks to merge the crack area as completely as possible, thereby solving the incomplete segmentation caused by the different grayscales of the crack area, and then according to the area change of the target super-pixel block of each frame of bearing image in its adjacent frame bearing image, obtain the probability that each target super-pixel block of each frame of bearing image is a crack area, identify the accurate crack area, and avoid the influence of light. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0033] Figure 1This is a flowchart of the steps of a method for identifying surface defects of motor bearings based on image processing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0035] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0036] See also Figure 1 , which shows a flowchart of a method for identifying surface defects of motor bearings based on image processing according to an embodiment of the present invention, the method comprising the following steps:

[0037] S001. Collect each frame of bearing image.

[0038] In the embodiment of the present invention, the bearing to be tested on the motor is rotated one circle, and a video of the bearing rotation is captured by a camera, which is recorded as a bearing video. The bearing video is frame-processed to obtain an RGB image of each frame of the bearing. Each frame of the bearing RGB image is grayscaled and recorded as a bearing image of each frame. It should be noted that the bearing images of adjacent frames are continuous.

[0039] S002. Perform superpixel segmentation on each frame of the bearing image to obtain each superpixel block of each frame of the bearing image, and obtain the characteristic value of each superpixel block of each frame of the bearing image based on the grayscale characteristics of each superpixel block; correct the characteristic value of each superpixel block to obtain the corrected characteristic value of each superpixel block of each frame of the bearing image.

[0040] It should be noted that when collecting each frame of bearing image, due to the influence of illumination, there will be an illumination shadow area in each frame of bearing image collected, and the illumination shadow area has a grayscale value similar to the crack area in the bearing, and the grayscale value in the crack area is also different. Therefore, the threshold segmentation algorithm or the edge detection algorithm cannot accurately segment the crack area in the bearing image. Therefore, in an embodiment of the present invention, a larger number of super-pixel blocks are first set, and super-pixel segmentation is performed on each frame of bearing image, and the crack area is divided into each super-pixel block as much as possible. Subsequently, the features of each super-pixel block are analyzed and merged, and the complete crack area is merged together. Finally, the accurate crack area is obtained based on the area performance of the crack area in the adjacent frame bearing image.

[0041] In the embodiment of the present invention, the number of super pixel blocks N is preset, and each frame of the bearing image is subjected to super pixel segmentation using a super pixel segmentation algorithm to obtain a plurality of super pixel blocks;

[0042] In an embodiment of the present invention, the preset number of superpixel blocks N=100. In other embodiments, the implementer may preset the value of the number of superpixel blocks N according to the specific implementation situation. It should be noted that when presetting the number of superpixel blocks, a larger number needs to be preset to divide the crack area into each superpixel block as much as possible, so as to facilitate the subsequent merging and analysis of each superpixel block.

[0043] It should be noted that the grayscale value of the known crack area is smaller than that of the normal area, and the gradient value of the edge of the crack area is larger than that of the normal area. Therefore, based on this feature, the feature value of each superpixel block of each frame of the bearing image is first obtained.

[0044] Record any frame bearing image as the current frame bearing image, and obtain the feature value of each superpixel block of each frame bearing image:

[0045] ;

[0046] Where, Represents the feature value of the i-th superpixel block of the bearing image of the current frame; Represents the gradient mean of the pixel points on the edge of the i-th superpixel block of the bearing image of the current frame; Represents the mean gradient in the bearing image of the current frame; Represents the grayscale mean of the bearing image of the current frame; Represents the grayscale mean of the i-th superpixel block of the current frame bearing image; norm() represents the normalization function;

[0047] Represents the grayscale feature of the i-th superpixel block of the bearing image of the current frame. It is known that the grayscale value of the crack area is smaller than that of the normal area. Therefore, When the value of is large, it means that the grayscale value of the i-th superpixel block is small. At this time, the possibility that the i-th superpixel block is a crack area is large, that is, the eigenvalue of the i-th superpixel block is large;

[0048] represents the gradient feature of the pixel point in the i-th superpixel block. It is known that the gradient value of the crack defect edge is larger than that of the normal area. Therefore, if The larger the value and the larger the eigenvalue, the greater the possibility that the i-th superpixel block is a crack area, and the larger the eigenvalue of the i-th superpixel block.

[0049] It should be noted that the above steps obtain the characteristic value of each superpixel block of each frame of the bearing image, that is, the larger the characteristic value of the superpixel block, the more likely the superpixel block is to be a crack area. However, when obtaining each frame of the bearing image, the influence of illumination will cause an illuminated shadow area to appear in each frame of the bearing image, and the grayscale values ​​of the illumination-affected area and the crack area are similar. Therefore, it is difficult to distinguish the illuminated shadow area from the crack area based solely on the characteristic value of the superpixel block.

[0050] It should be further explained that the grayscale contrast between the known illuminated shadow area and the normal area around it is relatively low, especially in the illuminated shadow area, the grayscale change is relatively gentle, and the edge shape of the illuminated shadow area is usually irregular. Affected by the illumination angle and the surrounding environment, the gradient direction of the edge pixel points of the illuminated shadow area is relatively scattered, and there is no obvious unified direction; while the grayscale contrast between the crack area and the normal area around it is high, which can clearly highlight the shape and boundary of the crack, and the gradient direction of the edge pixel points of the crack area is relatively consistent. Therefore, in an embodiment of the present invention, first, based on the eigenvalue difference between the superpixel blocks of each frame of the bearing image, the reference pixel block of each superpixel block of each frame of the bearing image is obtained, that is, the superpixel block and its reference pixel block are not similar. Therefore, if the grayscale contrast between any superpixel block of any frame of the bearing image and its reference superpixel block is greater and the eigenvalue of the superpixel block is greater, it means that the superpixel block of the frame of the bearing image is more likely to be a crack area;

[0051] Since the gradient direction of the edge of the crack area is usually perpendicular to the direction of the crack, this is because at the edge of the crack, the change in grayscale value is mainly along the direction perpendicular to the crack. It is necessary to analyze the direction of the crack area and compare it with the gradient direction of the edge of the crack area. Therefore, first, based on the difference in characteristic values ​​between the superpixel blocks of each frame of the bearing image, the similar pixel blocks of each superpixel block of each frame of the bearing image are obtained, that is, the similarity between the superpixel block and its similar pixel block. Then, based on the superpixel block and its similar pixel block, the direction of the superpixel block is analyzed to identify the crack area.

[0052] In an embodiment of the present invention, a difference threshold value T is preset, and the adjacent superpixel blocks of the i-th superpixel block of the current frame bearing image are obtained. If the absolute value of the difference between the characteristic values ​​of the i-th superpixel block of the current frame bearing image and its j-th adjacent superpixel block is less than the difference threshold value T, the j-th adjacent superpixel block of the i-th superpixel block of the current frame bearing image is recorded as the similar pixel block of the i-th superpixel block of the current frame bearing image. Otherwise, the j-th adjacent superpixel block of the i-th superpixel block of the current frame bearing image is recorded as the reference pixel block of the i-th superpixel block of the current frame bearing image. In an embodiment of the present invention, the preset difference threshold value T=0.2. In other embodiments, the implementer may preset the value of the difference threshold value T according to the specific implementation method.

[0053] The i-th superpixel block of the current frame bearing image is merged with its similar pixel blocks to obtain the target area. The skeleton of the target area is extracted using the morphological processing method. The pixel points on the skeleton of the target area are fitted into a straight line using the least squares method. The angle between the straight line and the horizontal axis is obtained as the direction angle of the i-th superpixel block of the current frame bearing image. It should be noted that if the superpixel block has no similar pixel block, only the superpixel block needs to be analyzed.

[0054] Get the gradient directions of all edge pixels of the i-th superpixel block of the current frame bearing image, and take the angle between the gradient direction with the highest number of occurrences and the horizontal axis as the gradient direction angle of the i-th superpixel block of the current frame bearing image.

[0055] Get the corrected feature value of each superpixel block of the current frame bearing image:

[0056] ;

[0057] Where, Represents the corrected eigenvalue of the i-th superpixel block of the bearing image of the current frame; Represents the feature value of the i-th superpixel block of the bearing image of the current frame; Represents the grayscale mean of the a-th reference superpixel block of the i-th superpixel block of the bearing image of the current frame; The grayscale mean of all reference superpixel blocks representing the i-th superpixel block of the current frame bearing image; The number of reference superpixel blocks representing the i-th superpixel block of the current frame bearing image; Represents the variance of the gradient direction of all edge pixels of the i-th superpixel block of the current frame bearing image; Represents the gradient direction angle of the i-th superpixel block of the bearing image of the current frame; Represents the direction angle of the i-th superpixel block in the current frame bearing image; exp() represents an exponential function with a natural constant as the base.

[0058] The larger the value of , the more likely the i-th superpixel block in the current frame bearing image is to be a crack area, and the larger its corrected eigenvalue is. Represents the grayscale contrast between the i-th superpixel block of the current frame bearing image and its reference superpixel block. The larger its value, the more likely the i-th superpixel block of the current frame bearing image is to be a crack area, and therefore the larger its corrected eigenvalue;

[0059] The smaller is , the more consistent the gradient direction of the edge pixels of the ith superpixel block of the current frame bearing image is, indicating that the ith superpixel block of the current frame bearing image is more likely to be a crack area, so its corrected eigenvalue is larger;

[0060] Represents the trend feature of the i-th superpixel block in the bearing image of the current frame. The smaller the value, the more likely the i-th superpixel block is to be a crack area.

[0061] A preset eigenvalue threshold T1 is set. If the corrected eigenvalue of any super-pixel block of the current frame bearing image is greater than the eigenvalue threshold T1, it means that the super-pixel block of the current frame bearing image is likely to be a crack area, and the super-pixel block is recorded as a representative super-pixel block, and all adjacent representative super-pixel blocks are merged and recorded as a target super-pixel block. In an embodiment of the present invention, the preset eigenvalue threshold T1=0.6. In other embodiments, the implementer can preset the value of the eigenvalue threshold T1 according to the specific implementation method.

[0062] At this point, the target superpixel block of each frame of the bearing image is obtained.

[0063] S003. Obtain the probability that each target superpixel block in each frame of the bearing image is a crack area based on the area change of the target superpixel block in each frame of the bearing image in its adjacent frame of the bearing image.

[0064] It should be noted that it is known that the target superpixel block in each frame of the bearing image may be a crack area, and the target superpixel block is obtained by analyzing the grayscale features in the image and then setting a threshold. If the target superpixel block in the bearing image is directly used as the crack area, it is inaccurate. In addition, since each frame of the bearing image is acquired when the bearing is rolled, and the crack appears or disappears when the bearing rotates, the area of ​​the crack area in the consecutive frames of the bearing image is inconsistent. However, the illumination of each frame of the bearing image is consistent. Because the area change of the illuminated shadow area in each frame of the bearing image is small, the probability that each target superpixel block in each frame of the bearing image is a crack area is obtained first according to the area change of each target superpixel block in each frame of the bearing image in its adjacent frame of the bearing image.

[0065] In an embodiment of the present invention, the r frame bearing images before the current frame bearing image and the r frame bearing images after the current frame bearing image are recorded as adjacent frame bearing images of the current frame bearing image, and the central pixel point of the cth target superpixel block of the current frame bearing image is corresponded to any of its adjacent frame bearing images, and the target superpixel block or superpixel block where the central pixel point is located in the adjacent frame bearing image is recorded as a reference area of ​​the cth target superpixel block of the current frame bearing image. Similarly, several reference areas of the cth target superpixel block of the current frame bearing image are obtained.

[0066] It should be noted that the preset number of adjacent frame images r=4. In other embodiments, implementers may preset the value of the number of adjacent frame images r according to specific implementation conditions.

[0067] Get the probability that each target superpixel block in the current frame bearing image is a crack area:

[0068] ;

[0069] Where, Represents the probability that the cth target superpixel block in the current frame bearing image is a crack area; Represents the variance of the area of ​​the cth target superpixel block in the current frame bearing image and its reference area; Represents the area of ​​the cth target superpixel block in the current frame bearing image; represents the area of ​​the reference region of the cth target superpixel block in the current frame bearing image; D represents the number of reference regions of the cth target superpixel block in the current frame bearing image; Represents the area change rate of the cth target superpixel block of the current frame bearing image in its adjacent frame bearing image. The larger the value, the greater the probability that the cth target superpixel block of the current frame bearing image is a crack area; It represents the area consistency of the cth target superpixel block of the current frame bearing image in its adjacent frame bearing image. The larger the value, the greater the probability that the cth target superpixel block of the current frame bearing image is a crack area.

[0070] S004. Obtain the crack defect of each frame of the bearing image based on the probability that each target superpixel block in each frame of the bearing image is a crack area.

[0071] In an embodiment of the present invention, the preset probability threshold T3=0.7. If the probability that any target super-pixel block of any frame bearing image is a crack area is greater than or equal to the probability threshold T3, the target super-pixel block of the frame bearing image is a crack area, and the staff is notified to replace the current bearing in time. In other embodiments, the implementer may preset the value of the probability threshold T3 according to the specific implementation method.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying surface defects of motor bearings based on image processing, characterized in that: include: Collect each frame of bearing image and perform superpixel segmentation to obtain each superpixel block of each frame of bearing image; record any frame of bearing image as the current frame of bearing image; obtain the corrected feature value of each superpixel block of the current frame of bearing image: , is the corrected eigenvalue of the i-th superpixel block; is the variance of the gradient direction of all edge pixels of the i-th superpixel block; as well as is the eigenvalue of the i-th superpixel block and the number of reference superpixel blocks; as well as is the grayscale mean of all reference superpixel blocks of the i-th superpixel block and the grayscale mean of the a-th reference superpixel block; as well as Represent the gradient trend angle and trend angle of the i-th super pixel block respectively; exp() represents an exponential function with a natural constant as the base; || represents the absolute value symbol; according to the modified eigenvalue, obtain the target super pixel block of the current frame bearing image; Get the probability that the cth target superpixel block in the current frame bearing image is a crack area: ; is the area variance between the cth target superpixel block and its reference area; as well as is the area of ​​the cth target superpixel block and the area of ​​the dth reference area of ​​the cth target superpixel block; D is the number of reference areas of the cth target superpixel block; Based on the probability, obtaining crack defects in each frame of the bearing image; The collecting of each frame of bearing image comprises: The bearing to be tested on the motor rotates one circle, and a camera is used to capture a video of the bearing rotation, which is recorded as the bearing video. The bearing video is frame-processed to obtain an RGB image of each frame of the bearing. Each frame of the RGB image of the bearing is grayscaled and recorded as the bearing image of each frame. The acquisition of the strike angle includes: Obtain similar pixel blocks of the i-th superpixel block of the current frame bearing image, merge the i-th superpixel block of the current frame bearing image with its similar pixel blocks to obtain the target area, use morphological processing methods to extract the skeleton of the target area, use the least squares method to fit the pixel points on the skeleton of the target area into a straight line, and obtain the angle between the straight line and the horizontal axis as the strike angle of the i-th superpixel block of the current frame bearing image; The acquisition of the gradient trend angle of the i-th superpixel block includes: Get the gradient directions of all edge pixels of the i-th superpixel block of the current frame bearing image, and take the angle between the gradient direction with the highest number of occurrences and the horizontal axis as the gradient direction angle of the i-th superpixel block of the current frame bearing image.

2. The method for identifying surface defects of motor bearings based on image processing according to claim 1, characterized in that: The acquisition of the characteristic value of the i-th super pixel block includes: ； Where, Represents the feature value of the i-th superpixel block of the pin axis image of the current frame; as well as Represents the gradient mean of the pixel points on the edge of the i-th super-pixel block of the pin axis image of the current frame and the grayscale mean of the i-th super-pixel block; as well as Represents the gradient mean and grayscale mean in the current pin image; norm() represents the normalization function.

3. The method for identifying surface defects of motor bearings based on image processing according to claim 1, characterized in that: The step of obtaining each superpixel block of each bearing image frame includes: The number of super-pixel blocks N is preset, and each frame of the bearing image is segmented into super-pixels using a super-pixel segmentation algorithm to obtain a number of super-pixel blocks.

4. The method for identifying surface defects of motor bearings based on image processing according to claim 1, characterized in that: The obtaining of the reference super-pixel block of the i-th super-pixel block includes: A preset difference threshold T is used to obtain the adjacent super-pixel blocks of the i-th super-pixel block of the current frame bearing image. If the absolute value of the difference between the characteristic values ​​of the i-th super-pixel block of the current frame bearing image and its j-th adjacent super-pixel block is greater than or equal to the difference threshold T, the j-th adjacent super-pixel block of the i-th super-pixel block of the current frame bearing image is recorded as the reference pixel block of the i-th super-pixel block of the current frame bearing image.

5. The method for identifying surface defects of motor bearings based on image processing according to claim 1, characterized in that: The step of obtaining a target superpixel block of the bearing image of the current frame includes: A preset eigenvalue threshold T1 is set. If the corrected eigenvalue of any superpixel block of the current frame bearing image is greater than the eigenvalue threshold T1, the superpixel block is recorded as a representative superpixel block, and all adjacent representative superpixel blocks are merged and recorded as the target superpixel block of the current frame bearing image.

6. The method for identifying surface defects of motor bearings based on image processing according to claim 1, characterized in that: Acquiring the reference area of ​​the c-th target superpixel block includes: The r frame bearing images before the current frame bearing image and the r frame bearing images after the current frame bearing image are recorded as adjacent frame bearing images of the current frame bearing image, the central pixel point of the cth target superpixel block of the current frame bearing image is corresponded to any of its adjacent frame bearing images, and the target superpixel block or superpixel block where the central pixel point is located in the adjacent frame bearing image is recorded as a reference area of ​​the cth target superpixel block of the current frame bearing image, and several reference areas of the cth target superpixel block of the current frame bearing image are obtained.

Images