A method for recognizing, positioning and measuring cracks in fluorescent magnetic powder inspection images of a train wheel set

Abstract

The application discloses a train wheelset fluorescent magnetic powder flaw detection image crack identification, positioning and measurement method, which comprises the following steps: partitioning a train wheelset and collecting a crack magnetic powder image of the train wheelset; pre-processing the collected crack magnetic powder image to highlight crack features; segmenting and extracting the crack features; and identifying, positioning and measuring the crack according to crack feature information. According to the shape characteristics of the crack, the application proposes a crack feature segmentation algorithm combining the crack area and the area of its circumscribed rectangle and the aspect ratio of the circumscribed rectangle of the connected domain, realizes accurate extraction of the wheelset crack features, and identifies, positions and measures the crack. Compared with other crack detection methods, the application not only realizes nondestructive detection of the surface crack of the train wheelset, but also effectively segments the features of the train wheelset crack, facilitates identification, positioning and measurement of the crack of the train wheelset, and thus realizes automatic wheelset crack magnetic powder flaw detection.

Description

Technical Field

[0001] This invention relates to a method for identifying, locating, and measuring surface cracks in train wheelsets, and more particularly to a method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images. Background Technology

[0002] Railway transportation plays a crucial role in the logistics industry. With economic growth, societal demand for railway services has increased significantly, placing higher demands on train maintenance. Train wheelsets, as a key component of trains, operate in extremely harsh environments and require continuous, long-term use. This leads to defects such as cracks on the wheelset surface. Without proper maintenance, the lifespan of the wheelsets can be greatly reduced, potentially even causing traffic accidents. Therefore, the identification and maintenance of surface cracks in train wheelsets are of paramount importance.

[0003] In existing technologies, crack identification in magnetic particle inspection images of train wheelsets is mostly at a semi-automated stage, relying mainly on manual inspection. This requires a lot of labor and time, and long working hours can cause eye fatigue, potentially leading to missed detections, delayed maintenance, and shortened wheelset lifespan.

[0004] Deep learning models have shown good performance in crack identification, but they require a large number of crack samples. Currently, the number of train wheelset crack samples is limited, making deep learning models unsuitable. Traditional methods are needed for train wheelset crack identification. In traditional crack identification, the key lies in the segmentation and extraction of crack features, retaining only these features. Since crack features are relatively long and thin, they are considered as long, closed regions in wheelset crack images. Crack features can be extracted by extracting connected components of the wheelset cracks. However, existing connected component filtering methods are only suitable for crack feature extraction and identification against a single background. Train wheelset surfaces are rough and complex, with interference from bright light beams, making it difficult for existing connected component filtering methods to extract wheelset crack features, leading to missed or incorrect extractions and failing to achieve the desired results. Furthermore, existing crack measurement methods are limited and lack accuracy. Therefore, it is necessary to develop a method for automatic crack defect identification, localization, and measurement on train wheelsets using fluorescent magnetic particle detection and computer vision-based technology, achieving automated identification, localization, and measurement of train wheelset cracks. Summary of the Invention

[0005] Objective: To address the difficulties in segmenting and extracting the complex and rough surface features of train wheelsets and the semi-automatic identification of wheelset cracks in existing technologies, this invention proposes a method for crack identification, localization, and measurement in train wheelset fluorescent magnetic particle inspection images. First, crack images are acquired by dividing the wheelset into sections. Then, the acquired images are preprocessed to highlight crack features, facilitating crack segmentation. Adaptive threshold segmentation and binarization are performed on the images. Morphological processing connects partially fractured cracks after binarization, forming closed regions containing crack features and other interfering features, from which connected component information is extracted. Based on the shape characteristics of the cracks, a crack feature segmentation algorithm is proposed, combining the relationship between the crack area and the area of ​​its circumscribed rectangle, and the aspect ratio of the circumscribed rectangle of the connected components. This achieves accurate extraction of wheelset crack features and further completes crack identification, localization, and measurement. An automatic crack identification and detection interface for train wheelsets is designed according to actual needs, realizing visualization and automation of automatic crack identification, localization, and measurement in train wheelset magnetic particle inspection images.

[0006] Technical solution: The present invention provides a method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images, comprising the following steps:

[0007] (1) Divide the train wheelsets into sections and collect magnetic particle images of the cracks in the train wheelsets;

[0008] (2) The acquired crack magnetic particle images are preprocessed to highlight crack features. The process is as follows:

[0009] (2.1) Perform G-channel separation and thresholding on the magnetic particle image;

[0010] (2.2) Filtering the magnetic particle image:

[0011] (3) Segment and extract crack features;

[0012] (4) Identify, locate, and measure the cracks based on the crack feature information in step (3); the process is as follows:

[0013] (4.1) Crack identification and localization: Let (X1, Y1) be the coordinates of the upper left corner of the circumscribed rectangle of the crack, and (X2, Y2) be the coordinates of the lower right corner of the circumscribed rectangle of the crack. Draw a detection box. Based on the crack feature extraction, obtain x, y, w, and h in the crack region, where x is the horizontal coordinate of the upper left corner of the circumscribed rectangle of each connected component, y is the vertical coordinate of the upper left corner of the circumscribed rectangle of each connected component, w is the width of the circumscribed rectangle of each connected component, h is the height of the circumscribed rectangle of each connected component, and s is the number of pixels occupied by each connected component, i.e., the area of ​​each connected component. Therefore:

[0014]

[0015] (4.2) Crack length measurement:

[0016] The actual length of the cross groove of the calibration plate of the train wheelset is known to be L. 实 Calculate the number of pixels in the cross-groove magnetic powder of the calibration plate in the crack magnetic powder image to obtain the length L of the cross-groove magnetic powder. 图 The calibration coefficient γ for the length of the magnetic powder was calculated. L :

[0017]

[0018] (4.3) Based on w and h obtained from the identified crack magnetic particle image, the number of pixels W occupied by the crack in the magnetic particle image is obtained. 图 The actual length W of the crack was calculated. 实 :

[0019] W 实 =W 图 ×γ L (8).

[0020] In step (1), when dividing the train wheelset into zones, the train wheelset is divided into three zones: left wheel, middle axle, and right wheel.

[0021] In step (1), each area of ​​the train wheelset is rotated before data is collected.

[0022] In step (1), a suspension is sprayed onto the surface of the train wheelset using a flaw detector, and then the crack images of the train wheelset are acquired using a wheelset crack image acquisition system.

[0023] In step (2), before preprocessing the image, the preprocessed image is binarized using an adaptive threshold segmentation method.

[0024] In step (2), a template is selected to traverse the pixels of the image. The gray value of the pixel at the center of the template is recorded as Q, and the weighted average value in the neighborhood is recorded as H. An offset C is introduced. When Q>HC, the gray value of the pixel is changed to obtain the image after binarization.

[0025] In step (2.1), the image pixel values ​​of the G channel are first extracted:

[0026] D = f(a, b) (1)

[0027] Where D is the grayscale value of the pixel in the G channel image, a is the x-coordinate of the pixel in the image, and b is the y-coordinate of the pixel in the image.

[0028] In step (2.2), a Gaussian template is used to scan each pixel in the image, and the value of the center pixel of the template is replaced by the weighted average gray value of the pixels in the neighborhood determined by the template.

[0029] In step (3), a crack feature segmentation algorithm combining the relationship between the crack area and the area of ​​the rectangle circumscribed by the crack and the aspect ratio of the circumscribed rectangle of the connected domain is used to screen and filter the crack features.

[0030] The crack features in step (3) include transverse crack features, longitudinal crack features, and diagonal crack features.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0032] (1) Compared with other crack detection methods, this invention not only performs non-destructive testing on the surface cracks of train wheelsets, but also effectively segments the characteristics of train wheelset cracks, which facilitates the identification, positioning and measurement of cracks in train wheelsets, thereby realizing the automation of magnetic particle testing of wheelset cracks.

[0033] (2) The wheelset crack detection interface of the present invention realizes the visualization of automatic crack identification, positioning and measurement. By collecting the wheelset cracks in sections, it is easy to quickly and accurately find the location of the crack.

[0034] (3) The present invention also has the characteristics of high processing accuracy, fast detection speed and low cost, and has advantages in the field of automated detection of magnetic particle flaws in train wheelsets. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images, as described in this invention.

[0036] Figure 2 This is a diagram showing the train wheelset partitioning of the present invention;

[0037] Figure 3 This is a model diagram of the wheelset flaw detector of the present invention;

[0038] Figure 4 This is a crack diagram of fluorescent magnetic powder on a train wheelset according to the present invention;

[0039] Figure 5 This is a G-channel separation image of the crack image of the present invention;

[0040] Figure 6 This is a diagram illustrating the threshold processing of the present invention;

[0041] Figure 7 This is the Gaussian smoothing plot of the present invention;

[0042] Figure 8 This is a flowchart of the connected component filtering algorithm used in this invention;

[0043] Figure 9 This is the adaptive threshold segmentation image used in this invention;

[0044] Figure 10This is a morphological processing diagram of the present invention;

[0045] Figure 11 This is an image of the crack aspect ratio obtained by the present invention;

[0046] Figure 12 This is a crack segmentation diagram obtained by the present invention;

[0047] Figure 13 This is the crack identification diagram obtained by the present invention;

[0048] Figure 14 This is a schematic diagram of the dimensions of the A1 type calibration plate of the present invention;

[0049] Figure 15 This is a magnetic particle calibration diagram for train wheelsets according to the present invention;

[0050] Figure 16 This is a classification diagram of train wheelset cracks according to the present invention;

[0051] Figure 17 This is a diagram of the interface for identifying, locating, and measuring wheelset cracks according to the present invention. Detailed Implementation

[0052] like Figures 1 to 17 As shown, the specific implementation steps of the method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to the present invention are as follows:

[0053] (1) Divide the train wheelsets into sections and collect magnetic particle images of the cracks in the train wheelsets;

[0054] Because train wheelsets are cylindrical and the cracks in the wheelsets are relatively small compared to the overall crack pattern, the overall image acquisition effect is poor. Therefore, this invention divides the wheelset into three regions: the left wheel, the central axle, and the right wheel, denoted as regions A, B, and C, respectively. Crack images are acquired sequentially for each region, rotated 120°, and then acquired again. The entire wheelset acquisition area is denoted as A-1, A-2, A-3, B-1, B-2, B-3, C-1, C-2, and C-3. Taking angle 1 as an example, the wheelset partition diagram is as follows: Figure 2 As shown.

[0055] After being divided into sections, the wheelsets to be tested are placed on a conveyor belt and fed into the flaw detector. The flaw detector model is as follows: Figure 3 As shown, 1 is the spraying system, 2 is the wheel axle positioning and rotation system, 3 is the loading and unloading mechanism, 4 is the wheelset crack image acquisition system, and 5 is the operation control console. The flaw detector is started to spray magnetic suspension liquid onto the wheelset surface, causing magnetic powder to accumulate in the crack area. Under ultraviolet light, this forms a brighter magnetic mark than in the non-cracked area. Then, the wheelset crack image acquisition system is used to acquire magnetic powder images of the wheelset cracks. The acquired crack images are shown below. Figure 4As shown, regions A-1, B-1, and C-1 are respectively. It can be seen from the figure that the surface of the wheelset is relatively rough and there is interference from the bright line bundles. The circled part is the crack feature of the wheelset.

[0056] (2) The acquired crack magnetic particle images are preprocessed to highlight crack features. The process is as follows:

[0057] (2.1) Perform G-channel separation and thresholding on the magnetic particle image:

[0058] Perform G-channel separation processing on the acquired RGB image; the separation image is shown below. Figure 5 As shown, in order to overcome the technical defect of poor contrast between cracks and background in the G-channel processing effect of the prior art, this invention proposes to perform thresholding processing on the G-channel processed image. First, the pixel values ​​of the G-channel image are extracted, and the pixel values ​​are represented as follows:

[0059] D = f(a, b) (1)

[0060] Where D is the grayscale value of the pixel in the G channel image, and a and b are the pixel coordinates of the image.

[0061] By observing the grayscale values ​​of image pixels, the grayscale value at the crack feature is greater than 95. Utilizing this characteristic, when the image grayscale value is greater than 95, its original grayscale value is returned; otherwise, 0 is returned. After processing, as shown... Figure 6 As shown, after processing, the contrast between the crack and the background is enhanced, and the crack features are more obvious.

[0062] (2.2) Filtering the magnetic particle image:

[0063] In this embodiment, a 5×5 Gaussian template is selected to scan each pixel in the image, and the value of the center pixel of the template is replaced by the weighted average gray value of the pixels in the neighborhood determined by the template. The effect is as follows. Figure 7 As shown in the figure, most of the interference information has been filtered out, making the crack features more prominent.

[0064] (3) Segment and extract crack features:

[0065] After the above processing, interference information is reduced and crack features are highlighted. Observing the characteristics of the crack in the processed image, the crack features and other interference features will form a closed region. Therefore, by observing the difference between crack features and other interference features, this invention proposes a crack feature segmentation algorithm that combines the relationship between the crack area and the area of ​​its circumscribed rectangle and the aspect ratio of the circumscribed rectangle of the connected component. This algorithm filters out interference information and retains crack information. Since the connected component processing object is a binary image, after image preprocessing, the image is first binarized. To improve the processing effect, morphological processing is then performed. The processing flow is as follows: Figure 8 As shown, the specific operation steps are as follows:

[0066] (3.1) Since the wheelset cracks are relatively small and have uneven brightness compared to the whole image, using Otsu's binarization method would lose some features of the cracks. Therefore, this invention uses an adaptive threshold segmentation method to perform binarization on the preprocessed image, as follows:

[0067] A 21×21 template is selected and iterated through each pixel of the image. The grayscale value of the center pixel of the template is denoted as Q, and the weighted average value of the neighborhood is denoted as H. To improve the processing effect, an offset C is introduced. When Q > HC, the grayscale value of the pixel is changed to white; otherwise, it is changed to black. This yields a binarized image, and the processing effect is as follows. Figure 9 As shown, the crack features are relatively well preserved, with a small number of cracks showing signs of fracture.

[0068] (3.2) To address the issue of some cracks breaking apart, morphological closing operations were further performed on the image to connect the broken cracks, as shown in the following figure. Figure 10 As shown, by Figure 10 It can be seen that the fracture crack has been completely connected.

[0069] (3.3) This invention uses Python and OpenCV development environment to perform connected component processing on the processed image. The cv2.connectedComponentsWithStats function in OpenCV is used to extract all connected components in the image. After processing, the connected component information is obtained: the x, y, w, h and s information corresponding to each connected component.

[0070] Where x is the x-coordinate of the top-left corner image pixel of the bounding rectangle of each connected component, y is the y-coordinate of the top-left corner image pixel of the bounding rectangle of each connected component, w is the width and height of the bounding rectangle of each connected component, h is the width and height of the bounding rectangle of each connected component, and s is the number of pixels occupied by each connected component, i.e., the area of ​​each connected component.

[0071] (3.4) Calculate the aspect ratio P of the bounding rectangle of each connected component in the image. That is, when the bounding rectangle w>h, the aspect ratio P = h / w, and conversely, P = w / h. The calculation formula is expressed as:

[0072]

[0073] Where h and w are the height and width of the bounding rectangle of each connected component; taking A-1 as an example, calculate the aspect ratio of all bounding rectangles of connected components, and the results are as follows. Figure 11 As shown; where the horizontal axis represents the number of connected components and the vertical axis represents the aspect ratio.

[0074] (3.5) Due to the slender shape of the crack, its area relative to its circumscribed rectangle is relatively small. After multiple experiments, the selection rule was determined as follows:

[0075] Step 1. Obtain the information of x, y, w, h, s, and P for each connected component in the above image.

[0076] Step 2. Execute s < w * h * 0.3.

[0077] Step 3. If P ≤ 0.3, output; if P > 0.3, execute s < w * h * 0.15 and then output.

[0078] Step 4. Output the information of x, y, w, h, and s.

[0079] After screening, the information of the screened connected components is obtained. Create a template with the same size as the black and the captured image. Overlay the connected component information with the template to obtain an accurate segmentation map of the crack. The segmentation effect is as shown in Figure 12 Shown, and by comparison, it can be concluded that the segmentation effect is better than the original image.

[0080] (4) Identify, locate, and measure the crack based on the crack feature information in step (3). The process is as follows:

[0081] (4.1) Crack identification and location: Denote (X1, Y1) as the coordinates of the upper - left corner of the bounding rectangle of the crack, and (X2, Y2) as the coordinates of the lower - right corner of the bounding rectangle of the crack. According to the information of the crack region x, y, w, h obtained after crack feature extraction, where x is the abscissa of the upper - left corner image pixel of the bounding rectangle of each connected component, y is the ordinate of the upper - left corner image pixel of the bounding rectangle of each connected component, w is the width of the bounding rectangle of each connected component, h is the height of the bounding rectangle of each connected component, and s is the number of pixel points occupied by each connected component, that is, the area of each connected component, we get:

[0082] X1 = x (3)

[0083] Y1 = y (4)

[0084] X2 = x + w (5)

[0085] Y2 = y + h (6)

[0086] According to the information of (X1, Y1) and (X2, Y2), use the cv2.rectangle function to draw the detection red box, as shown in Figure 13 Shown. It can be seen from the figure that the detection effect is good. Locate the actual crack position of the wheel set according to the partition location in step (1).

[0087] (4.2) Crack length measurement:

[0088] First, the train wheelsets are dimensionally calibrated using calibration plates. The A1 type calibration plate has artificial crack defects in both horizontal and vertical directions, as well as a circular crack defect on the outer perimeter. These crack defects of different directions and shapes are designed to verify the direction of the magnetic field. Figure 14 As shown.

[0089] Given the actual transverse and longitudinal lengths of the artificial cross groove in the calibration piece, the actual length of the cross groove in the calibration piece is L. 实 Calculate the number of pixels of magnetic powder in the cross groove of the sample in the magnetic particle image, such as Figure 15 As shown in Table 1, after multiple calculations, the average value was taken. Ten sets of data were measured to obtain the length L of the cross-groove magnetic powder in the image. 图 Further calculations yielded the magnetic powder length calibration coefficient γ. L The calculation formula is as follows:

[0090]

[0091] Table 1

[0092]

[0093]

[0094] The calibration coefficient γ was calculated. L = 0.194 mm / pixel.

[0095] Based on the w and h information obtained from the identified crack image, the number of pixels W occupied by the crack in the image is obtained. 图 The actual length W of the crack was further calculated. 实 The calculation formula is as follows:

[0096] W 实 =W 图 ×γ L (8)

[0097] In calculating W 图 In this invention, cracks are classified into three categories based on their characteristics. The width and height of the circumscribed rectangle of the crack are approximated as the length of the crack. The classification method is as follows:

[0098] When w / h > 4, it is recorded as a transverse crack;

[0099] When w / h < 0.25, it is recorded as a longitudinal crack;

[0100] When 0.25≤w / h≤4, it is denoted as a diagonal crack.

[0101] Taking A-1 as an example, the classification diagram is as follows: Figure 16As shown in Table 2, the actual transverse crack length is calculated based on the w information, the actual longitudinal crack length is calculated based on the h information, and the actual width and height of the circumscribed rectangle of the diagonal crack are calculated based on the w and h information, and then the actual diagonal crack length is calculated using the Pythagorean theorem.

[0102] Table 2

[0103]

[0104] As a preferred approach, to achieve visualization of crack identification, location, and measurement, after completing the identification, location, and measurement of wheel set cracks, a software interface for train wheel set crack identification, location, and measurement was developed using PyQt5 and Python development platforms, and each functional module was implemented. The modules include four modules: image acquisition, crack identification, location, and measurement, data storage, and user interface. The software interface is divided into two main areas: an image display area and an operation area. The image display area includes the acquired crack image, the display of the detection results, the identification time, the current image number, the flaw detection results, the flaw detection record, the export of the flaw detection record, and the real-time time. When acquiring images, the images are saved according to the region division in step (1) based on the crack type, and the image name is named according to the acquisition region name. The flaw detection record is displayed according to the region, including the crack classification, crack length, and detection time. The flaw detection results can be exported as needed. The operation area includes workpiece information, image acquisition and detection operation buttons, the location for saving acquired images, and the location for saving detection images. The interface effect is as follows: Figure 17 As shown.

Claims

1. A method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images, characterized in that: It includes the following steps: (1) Divide the train wheel set into zones and collect the crack magnetic particle images of the train wheel set; (2) Preprocess the collected crack magnetic particle images to highlight the crack features. The process is as follows: (2.1) Separate the G channel of the magnetic particle image and perform threshold processing; (2.2) Perform filtering on the magnetic particle image: (3) Segment and extract the crack features; (3.1) Use the method of adaptive threshold segmentation to perform binary processing on the preprocessed image. Specifically, traverse each pixel of the image, record the gray value of the template center pixel as Q, and the weighted average mean in the neighborhood as H. To improve the processing effect, introduce an offset C. When Q > H - C, change the gray value of this pixel to white, otherwise change it to black, thereby obtaining a binary image; (3.2) Further perform morphological closing operation on the image to connect the broken cracks; (3.3) Perform connected component processing on the processed image, extract all connected components in the image, and obtain the information of the connected components after processing: the x, y, w, h, and s information corresponding to each connected component; where x is the abscissa of the upper left corner image pixel of the bounding rectangle of each connected component, y is the ordinate of the upper left corner image pixel of the bounding rectangle of each connected component, w is the width of the bounding rectangle of each connected component, h is the height of the bounding rectangle of each connected component, and s is the number of pixel points occupied by each connected component, that is, the area of each connected component; (3.4) Calculate the aspect ratio P of the circumscribed rectangle of each connected component in the image. That is, when the circumscribed rectangle w > h, the aspect ratio P = h / w, and vice versa. The calculation formula is expressed as follows: ; where h and w are the height and width of the bounding rectangle of each connected component; taking A-1 as an example, calculate the aspect ratio of all bounding rectangles of connected components; (3.5) Determine the screening rule as follows: Step1 Obtain the x, y, w, h, s, P information of each connected component in the image; Step2 Execute s < w * h * 0.3; Step3 If P ≤ 0.3, output the screened x, y, w, h, s information; if P > 0.3, execute s < w * h * 0.15 and output the screened x, y, w, h, s information; (4) Identify, locate, and measure the cracks according to the crack feature information in step (3). The process is as follows: (4.1) Crack identification and location. Denote (X1, Y1) as the coordinates of the upper left corner of the crack bounding rectangle, and (X2, Y2) as the coordinates of the lower right corner of the crack bounding rectangle, and draw a detection box. According to the x, y, w, h in the crack region obtained after crack feature extraction, where x is the abscissa of the upper left corner image pixel of the bounding rectangle of each connected component, y is the ordinate of the upper left corner image pixel of the bounding rectangle of each connected component, w is the width of the bounding rectangle of each connected component, h is the height of the bounding rectangle of each connected component, and s is the number of pixel points occupied by each connected component, that is, the area of each connected component, we get: ； (4.2) Crack length measurement: The actual length of the cross groove of the calibration plate of the train wheelset is known to be L. 实 Calculate the number of pixels in the cross-groove magnetic powder of the calibration plate in the crack magnetic powder image to obtain the length L of the cross-groove magnetic powder. 图 The calibration coefficient γ for the length of the magnetic powder was calculated. L : （7） (4.3) Based on w and h obtained from the identified crack magnetic particle image, the number of pixels W occupied by the crack in the magnetic particle image is obtained. 图 The actual length W of the crack was calculated. 实 : （8）。 2. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (1), when dividing the train wheel set, divide the train wheel set into three zones: the left wheel, the middle axis, and the right wheel.

3. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (1), rotate each zone of the train wheel set and then collect the images.

4. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (1), spray the suspension on the surface of the train wheel set through a flaw detector, and then use the wheel set crack image acquisition system to collect the crack images of the train wheel set.

5. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (2.1), first extract the image pixel values of the G channel: D = f(a, b) (1) Where D is the grayscale value of the pixel in the G channel image, a is the x-coordinate of the pixel in the image, and b is the y-coordinate of the pixel in the image.

6. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (2.2), a Gaussian template is used to scan each pixel in the image, and the value of the center pixel of the template is replaced by the weighted average gray value of the pixels in the neighborhood determined by the template.

7. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: In step (3), a crack feature segmentation algorithm combining the relationship between the crack area and the area of ​​the rectangle circumscribed by the crack and the aspect ratio of the circumscribed rectangle of the connected domain is used to screen and filter the crack features.

8. The method for identifying, locating, and measuring cracks in train wheelsets using fluorescent magnetic particle inspection images according to claim 1, characterized in that: The crack features in step (3) include transverse crack features, longitudinal crack features, and diagonal crack features.

Images