Intelligent rating method for network carbides of high-carbon-chromium bearing steel
Through the image segmentation technology trained by U2Net network model and mixed loss function, the problem of the mesh carbide rating of high-carbon chromium bearing steel depends on manual experience, and efficient and accurate intelligent rating is achieved, which improves the accuracy of ratings and data traceability.
Patent Information
- Application Number
- CN202510356746.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-07-11
AI Technical Summary
The mesh carbide rating of high-carbon chromium bearing steel in the prior art depends on the experience of professionals, resulting in inaccurate results, time-consuming and untraceable data, and the deep learning model does not capture small targets and boundary details information sufficiently, resulting in inaccurate ratings.
The U2Net network model is used for image segmentation, combined with mixed loss function training, and the mesh carbide mask diagram is extracted through the image segmentation model, and the boundaries are optimized using image expansion and corrosion operations, and intelligent rating is achieved by combining hit-and-hit transformation and post-processing functions.
It realizes efficient, accurate and fast mesh carbide ratings, reduces dependence on professionals, and improves rating accuracy and data traceability.
Smart Images
Figure CN120299037A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of metal microstructure analysis, and particularly to an intelligent rating method for network carbide in high-carbon chromium bearing steel. Background Art
[0002] High-carbon chromium bearing steel is used to manufacture bearing rings, tools and wear-resistant parts, and has the advantages of high hardness, wear resistance, fatigue resistance, corrosion resistance, etc. However, in the production process, its cleanliness, carbide distribution uniformity, tissue defects and smelting process requirements are extremely strict, and it is one of the steel types with the strictest quality control in steel production. Network carbide refers to a tissue feature in which carbides are distributed along grain boundaries or certain specific regions in the form of a network or chain, showing a continuous or semi-continuous distribution. The network carbide structure often leads to uneven distribution of carbides in the metal matrix. This network structure usually appears when the casting, forging or heat treatment process is improper. The network carbide structure is often an undesirable tissue morphology, which has a negative impact on the toughness and fatigue performance of the material. Network carbide rating is an important index for evaluating the performance of high-carbon chromium bearing steel.
[0003] At present, metallographic inspection mainly relies on metallographic inspection personnel with professional experience and qualifications to observe and evaluate the grade using an optical microscope. This traditional metallographic inspection mode has the following problems: 1) High professional threshold: The professional requirements for inspectors are high, and it takes 3 - 5 years to train a professional metallographic inspector, with a relatively long training cycle.
[0004] 2) Strong dependence on subjective experience: The rating result of network carbide depends on the personal experience of metallographic inspection personnel. Due to strong subjectivity, the rating result of network carbide is inaccurate, has a large deviation, and the rating takes a long time.
[0005] 3) Process data cannot be traced. Manual inspection only retains the inspection results of the specimens and does not store process data such as images and intermediate inspection data. If the customer has any objections to the inspection rating later, it is necessary to remake the specimens, take pictures and rate them again, which greatly increases the workload of the inspection personnel.
[0006] Network carbide usually appears at the grain boundaries, presenting a continuous or semi - continuous chain - like, network - like, or even island - like distribution form. The network contours vary in size, and the morphological distribution is intricate. There is a problem of unbalanced distribution between the foreground network carbide information and the background matrix information in the image (the proportion of foreground network pixels is relatively small). Image segmentation models based on deep learning rely on a large amount of high - quality training data. Insufficient training data or low - quality annotation, especially for some unseen data, will affect the generalization ability of the model, and it is not sufficient to capture boundary detail information such as small targets, linear, and chain - like, resulting in inaccurate target segmentation edges and missing segmentation regions. Therefore, seeking an efficient, accurate, and fast intelligent method for rating network carbide is beneficial to the development of metallographic inspection and analysis. Summary of the Invention
[0007] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide an intelligent rating method for network carbide in high - carbon chromium bearing steel.
[0008] To solve the above - mentioned technical problems, the technical solution of the present invention is as follows: An intelligent rating method for network carbide in high - carbon chromium bearing steel, comprising: Collecting network carbide images of high - carbon chromium bearing steel specimens and making a data set; Training an image segmentation model based on the data set; Obtaining the network carbide image to be detected of the target high - carbon chromium bearing steel and extracting the mask image of the network carbide in the image through the image segmentation model; Rating the network carbide based on the structural data of the network carbide in the mask image.
[0009] As a preferred scheme of the intelligent rating method for network carbide in high - carbon chromium bearing steel of the present invention, wherein: the collecting of network carbide images of high - carbon chromium bearing steel specimens includes: Polishing the high - carbon chromium bearing steel specimens and etching them with 4% nitric acid alcohol solution after polishing; Collecting network carbide images of high - carbon chromium bearing steel specimens using an optical microscope.
[0010] As a preferred scheme of the intelligent rating method for network carbide in high - carbon chromium bearing steel of the present invention, wherein: the making of the data set includes: Screening the collected network carbide images; Annotating the network targets in the network carbide images; Dividing the annotated network carbide images into a training set and a test set according to a preset ratio.
[0011] As a preferred embodiment of the intelligent rating method for the network carbide of high-carbon chromium bearing steel according to the present invention, wherein: the data set includes images of network carbide of all grades.
[0012] As a preferred embodiment of the intelligent rating method for the network carbide of high-carbon chromium bearing steel according to the present invention, wherein: the training of the image segmentation model based on the data set includes: Select the U2Net network model as the basic image segmentation model; Train the basic image segmentation model with the training set, and adjust the hyperparameters of the basic image segmentation model based on the hybrid loss function during the training process; Test the trained basic image segmentation model with the test set.
[0013] As a preferred embodiment of the intelligent rating method for the network carbide of high-carbon chromium bearing steel according to the present invention, wherein: the hybrid loss function is: ; wherein, BCELoss is the binary cross-entropy loss, which is used to calculate the difference between the predicted probability value and the true label, and for each pixel position (i,j), the definition of the binary cross-entropy loss is: BCE(pij, gij)=−[gij·log(pij)+(1−gij)·log(1−pij)]; DiceLoss is the loss function based on the Dice coefficient, which is used to measure the similarity of two sets, and its definition is: 。
[0014] As a preferred embodiment of the intelligent rating method for the network carbide of high-carbon chromium bearing steel according to the present invention, wherein: the obtaining of the image of the network carbide to be detected of the target high-carbon chromium bearing steel and the extraction of the mask image of the network carbide in the image by the image segmentation model includes: Obtain several images of the network carbide to be detected of the target high-carbon chromium bearing steel; Input the image to be detected into the image segmentation model, and use the mask image output by the image segmentation model as the rough mask image of the network carbide; Perform an image erosion operation on the rough mask image to remove the noise information of the miscellaneous points in the rough mask image; Use the image dilation operation to connect the broken linear, chain-like, and network-like contours in the rough mask image, and use the thinning algorithm to extract the skeleton line, remove the edge pixels of the rough mask image, and extract the structure and edge information of the network carbide until the fine mask image of the network carbide is obtained.
[0015] As a preferred solution of the intelligent rating method for the network carbide of high-carbon chromium bearing steel described in the present invention, wherein: the rating of the network carbide based on the structural data of the network carbide in the mask graph includes: Traverse each contour target in the fine mask graph of the network carbide, and sequentially record the structural information of each contour target; Based on the hit-or-miss transformation of different kernel functions, determine whether there is a closed network in the fine mask graph of the network carbide, and obtain the number p of chain-like contours and the number q of folding angles. If there is a closed network, the current fine mask graph is rated as level 4. If there is no closed network, calculate the rating score of the current fine mask graph through formula (1), and determine the rating level of the current fine mask graph based on the rating score; The formula (1) is: , where grade is the rating score of the current fine mask graph, is the total number of pixels of the skeleton lines of all carbide networks in the current fine mask graph; x is the width value of the current fine mask graph, y is the height value of the current fine mask graph; p is the number of chain-like contours, is the contribution degree of the number of chain-like contours; q is the number of folding angles, is the contribution degree of the number of folding angles; n is the number of network carbides in the current fine mask graph; If , the rating level of the current fine mask graph is level 0; if , the rating level of the current fine mask graph is level 1; if , the rating level of the current fine mask graph is level 2; if , the rating level of the current fine mask graph is level 2.5; wherein, , , are the level thresholds of the rating function.
[0016] As a preferred solution of the intelligent rating method for the network carbide of high-carbon chromium bearing steel described in the present invention, wherein: after determining whether there is a closed network in the fine mask graph of the network carbide based on the hit-or-miss transformation of different kernel functions, and obtaining the number p of chain-like contours and the number q of folding angles, if there is a closed network, the current fine mask graph is rated as level 4, if there is no closed network, calculate the rating score of the current fine mask graph through formula (1), and determine the rating level of the current fine mask graph based on the rating score, further including: Determine the network carbide image with the highest rating level in the target high-carbon chromium bearing steel, and use the rating level of its corresponding fine mask graph as the network carbide rating level of the target high-carbon chromium bearing steel.
[0017] As a preferred solution of the intelligent rating method for the network carbide of high-carbon chromium bearing steel described in the present invention, where: after determining the network carbide image with the highest rating level in the target high-carbon chromium bearing steel and taking the rating level of its corresponding fine mask image as the network carbide rating level of the target high-carbon chromium bearing steel, it further includes: Storing the network carbide rating data of the target high-carbon chromium bearing steel, where the network carbide rating data of the target high-carbon chromium bearing steel includes: the target high-carbon chromium bearing steel number, the network carbide image number corresponding to the target high-carbon chromium bearing steel, the rating level corresponding to each network carbide image, and the network carbide rating level of the target high-carbon chromium bearing steel.
[0018] The beneficial effects of the present invention are: The present invention uses an image segmentation model to extract the global context information of the image, and solves the problem of unbalanced distribution of foreground network carbide and background matrix in the image through mixed Loss. First, it identifies the overall information of the image to obtain a rough mask image of the network carbide, and then optimizes the rough boundary connection of the network carbide through dilation and erosion operations of the image. The thinning algorithm is used to obtain the skeleton image of the network carbide, and a fine mask image with fine boundaries is obtained. Then, the hit-or-miss transform is used to extract the key features of the carbide network, and combined with the post-processing function of the carbide network business rules, the intelligent rating of the network carbide is realized, and the efficient, accurate, and rapid rating of the network carbide is also realized, which is beneficial to the development of metallographic inspection and analysis. Description of the Drawings
[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0020] Figure 1 It is a schematic flow chart of the intelligent rating method for the network carbide of high-carbon chromium bearing steel provided by the present invention; Figure 2 It is the original schematic diagram of the collected network carbide; Figure 3 It is the schematic diagram of the labeled network carbide image; Figure 4 It is the rough mask image of the network carbide image; Figure 5 It is the fine mask image of the network carbide image; Figure 6 It is the schematic diagram of the endpoints based on the fine mask image of the carbide network; Figure 7Schematic diagram of inflection points based on the fine mask diagram of carbide network Figure 8 Schematic diagram of the standard atlas of carbide network of Grade 1 according to national standard Figure 9 Schematic diagram of the standard atlas of carbide network of Grade 2 according to national standard Figure 10 Schematic diagram of the standard atlas of carbide network of Grade 2.5 according to national standard Figure 11 Schematic diagram of the standard atlas of carbide network of Grade 3 according to national standard Figure 12 Another process schematic diagram of the intelligent rating method for network carbide in high-carbon chromium bearing steel Detailed implementation manners
[0021] To make the content of the present invention easier to be clearly understood, the following further detailed description of the present invention is made according to the detailed implementation manners and in combination with the accompanying drawings
[0022] Figure 1 Process schematic diagram of the intelligent rating method for network carbide in high-carbon chromium bearing steel provided by the embodiment of the present application. The method specifically includes the following steps Step S101: Collect the network carbide images of high-carbon chromium bearing steel specimens and make a data set
[0023] Specifically, preprocess the high-carbon chromium bearing steel specimens, that is, polish the high-carbon chromium bearing steel specimens and etch them with 4% nitric acid alcohol solution after polishing. Then, take the network carbide images of the high-carbon chromium bearing steel specimens under an optical microscope at 200X or 500X, and set the resolution of the images to 2048*2448
[0024] It should be noted that it is required to take all regions of high-carbon chromium bearing steel carbon specimens of multiple different grades to form carbide network images covering all grades, so as to ensure the diversity and integrity of the image acquisition data
[0025] After completing the shooting and collection of the network carbide images of the high-carbon chromium bearing steel specimens, it is necessary to first preliminarily screen the collected network carbide images to remove images with surface dirt, defocus blur, and poor imaging quality. Then, use a marking tool to finely label the network targets of the carbide in the images. See Figure 2 and Figure 3 , where Figure 2 is the original network carbide image collected Figure 3 is the network carbide image after annotation. After the annotation is completed, divide the training set and the test set according to the ratio of 7:3
[0026] Step S102: Train the image segmentation model based on the data set
[0027] Specifically, first, a basic image segmentation model is selected. In this embodiment, the U2Net network model is selected as the basic image segmentation model.
[0028] After that, the basic image segmentation model is trained with a training set, and the hyperparameters of the basic image segmentation model are adjusted based on a hybrid loss function during the training process. Appropriate hyperparameters are set, including batch size = 16, epoch = 1000, optimizer = Adam, lr = 0.001, eps = 1e-08. The hybrid loss Hybird Loss is used, and a multi-scale loss strategy is adopted. The losses are calculated separately for multiple output layers in the network, and the weighted sum of these losses constructs the final total loss. This multi-scale loss helps the model to perform fine feature learning at different scales, improve the segmentation accuracy, and helps to optimize the situation where the reticular contour of the carbide foreground and the proportion distribution of the matrix background are seriously unbalanced.
[0029] The above hybrid loss function is: ; where BCELoss is the binary cross-entropy loss, which is used to calculate the difference between the predicted probability value and the true label. And for each pixel position (i, j), the definition of the binary cross-entropy loss is: BCE(pij, gij) = -[gij·log(pij) + (1 - gij)·log(1 - pij)]; DiceLoss is a loss function based on the Dice coefficient, which is used to measure the similarity between two sets, and its definition is: .
[0030] After the model training is completed, the trained basic image segmentation model is tested with a test set.
[0031] Step S103: Obtain the reticular carbide image of the target high-carbon chromium bearing steel to be detected, and extract the mask image of the reticular carbide in the image through the image segmentation model.
[0032] Specifically, for the target high-carbon chromium bearing steel to be detected, first obtain the reticular carbide image therein as the input data of the above image segmentation model. The acquisition method is the same as the acquisition method in step S101 and will not be elaborated here.
[0033] It should be noted that for each target high-carbon chromium bearing steel to be detected, several reticular carbide images need to be collected.
[0034] After that, the obtained reticular carbide image is input into the image segmentation model, and the image segmentation model extracts features from the reticular carbide image to obtain the mask image corresponding to the reticular carbide image. SeeFigure 4 Use this mask image as the rough mask image for the corresponding reticular carbide image.
[0035] Subsequently, perform an image erosion operation on the rough mask image output by the image segmentation model to remove the miscellaneous noise information in the rough mask image.
[0036] Finally, use the image dilation operation to connect the discontinuous linear, chain-like, and reticular contours, and use the thinning algorithm to extract the skeleton line, remove the edge pixels of the reticular carbide rough mask image, and extract the main structure and edge information of the reticular carbide until a fine mask image of the reticular carbide is obtained. See Figure 5 。
[0037] Step S104: Rate the reticular carbide based on the structural data of the reticular carbide in the mask image.
[0038] Specifically, first, rate the reticular carbide in a single fine mask image. The grade of the reticular carbide mainly depends on the structural information of the reticular carbide, including the number of closed meshes, chain-like structures, and the number of corners, etc. The number of chain-like structures is approximately obtained through the endpoints of the skeleton graph. See Figure 6 , where the green cross stars in the figure represent the endpoints of the carbide reticulation. The number of chain-like structures of the carbide reticulation can be approximately obtained through the endpoints. The number of corners is obtained through the number of inflection points of the skeleton graph. See Figure 7 , where the green cross stars in the figure represent the inflection points of the carbide reticulation. The number of corners of the carbide reticulation can be approximately obtained through the inflection points. For the fine mask image obtained in the above step S103, traverse each contour target in the figure and record the information of each contour target in turn. Determine whether there is a closed mesh in the fine mask image of the reticular carbide through the hit-or-miss transform based on different kernel functions, and obtain the number of chain-like structures p and the number of corners q of the contour.
[0039] It should be noted that the grades of the reticular carbide include four evaluation grades: grade 1, grade 2, grade 2.5, and grade 4. The specific standard graphs are shown in Figures 8 to 11 in turn. If there is a closed mesh in one contour target, the evaluation grade of this fine mask image is grade 4. If there is no closed mesh, calculate the number of chain-like structures p, the number of corners q, and the number of pixels of all contours in turn, and calculate the evaluation grade of the fine mask image. The specific calculation method is as follows: The calculation formula is: , where grade is the rating score of the current fine mask image, is the total number of pixels of the skeleton lines of all carbide reticulations in the current fine mask image; x is the width value of the current fine mask image, y is the height value of the current fine mask image; p is the number of chain-like structures, is the contribution degree of the number of chain-like strips; q is the number of folding angles, is the contribution degree of the number of folding angles; n is the number of network carbides in the current fine mask image.
[0040] The mapping relationship of the rating function is: If , then the rating level of the current fine mask image is level 0; if , then the rating level of the current fine mask image is level 1; if , then the rating level of the current fine mask image is level 2; if , then the rating level of the current fine mask image is level 2.5; where and and are the level thresholds of the rating function, calculated from the atlas in the national standard file using the above formula.
[0041] It can be understood that the above steps calculate the rating structure of the network carbides in each fine mask image. For each target high-carbon chromium bearing steel to be detected, a number of network carbide images are collected. Among all the network carbide images, the most severe field of view and area are used as the rating basis, that is, the highest rating level among all the network carbide images is used as the carbide network level of the final specimen.
[0042] Step S105: Store the rating data of the network carbides of the target high-carbon chromium bearing steel.
[0043] Specifically, store the rating data of the network carbides of the target high-carbon chromium bearing steel in the FastDFS file server for subsequent management and tracking.
[0044] It should be noted that the above-mentioned rating data of the network carbides of the target high-carbon chromium bearing steel includes: the number of the target high-carbon chromium bearing steel, the number of the network carbide image corresponding to the target high-carbon chromium bearing steel, the rating level corresponding to each network carbide image, and the network carbide rating level of the target high-carbon chromium bearing steel.
[0045] Figure 12 This is another schematic flowchart of the intelligent rating method for network carbides of high-carbon chromium bearing steel provided by the embodiment of the present application.
[0046] Therefore, the technical solution of this application uses an image segmentation model to extract the global context information of the image, and processes the problem of the unbalanced distribution of foreground reticular carbides and background matrix in the image through a mixed Loss. First, the overall information of the image is recognized to obtain a rough mask image of the reticular carbides, and then the rough boundary joints of the reticular carbides are optimized through dilation and erosion operations of the image. The skeleton image of the reticular carbides is obtained by using a thinning algorithm to obtain a mask image with fine boundaries. After that, the key features of the carbide network are extracted by using the hit-or-miss transform, and combined with the post-processing function of the carbide network business rules to realize the intelligent rating of the reticular carbides, and also realize the efficient, accurate and rapid rating of the reticular carbides, which is beneficial to the development of metallographic inspection and analysis.
[0047] In addition to the above embodiments, the present invention may also have other implementation manners; all technical solutions formed by equivalent replacement or equivalent transformation fall within the protection scope required by the present invention.
Claims
1. An intelligent rating method for the network carbide of high-carbon chromium bearing steel, characterized in that: Including: Collecting the network carbide images of high-carbon chromium bearing steel specimens and creating a dataset; Training an image segmentation model based on the dataset; Obtaining the network carbide image to be detected of the target high-carbon chromium bearing steel and extracting the mask image of the network carbide in the image through the image segmentation model; Rating the network carbide based on the structural data of the network carbide in the mask image.
2. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 1, wherein: The collecting of the network carbide images of high-carbon chromium bearing steel specimens includes: Polishing the high-carbon chromium bearing steel specimens and etching them with a 4% nitric acid alcohol solution after polishing; Collecting the network carbide images of high-carbon chromium bearing steel specimens using an optical microscope.
3. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 1, wherein: The creating of the dataset includes: Screening the collected network carbide images; Labeling the network targets in the network carbide images; Dividing the labeled network carbide images into a training set and a test set according to a preset ratio.
4. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 3, characterized in that: The dataset includes network carbide images of all grades.
5. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 1, characterized in that: The training of the image segmentation model based on the dataset includes: Selecting the U2Net network model as the basic image segmentation model; Training the basic image segmentation model with the training set and adjusting the hyperparameters of the basic image segmentation model based on a hybrid loss function during the training process; Testing the trained basic image segmentation model with the test set.
6. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 5, characterized in that: The hybrid loss function is as follows: ; Among them, BCELoss is the binary cross-entropy loss, which is used to calculate the difference between the predicted probability value and the true label, and for each pixel position (i, j), the definition of the binary cross-entropy loss is: BCE(pij, gij)=−[gij·log(pij)+(1−gij)·log(1−pij)]; DiceLoss is a loss function based on the Dice coefficient, which is used to measure the similarity between two sets and is defined as: .
7. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 1, wherein: The obtaining of the network carbide image to be detected of the target high-carbon chromium bearing steel and the extraction of the mask image of the network carbide in the image through the image segmentation model includes: Obtaining several network carbide images to be detected of the target high-carbon chromium bearing steel; Inputting the image to be detected into the image segmentation model and taking the mask image output by the image segmentation model as the rough mask image of the network carbide; Performing an image erosion operation on the rough mask image to remove the miscellaneous noise information in the rough mask image; Using an image dilation operation to connect the broken linear, chain-like, and network-like contours in the rough mask image, and using a thinning algorithm to extract the skeleton line, removing the edge pixels of the rough mask image, and extracting the structure and edge information of the network carbide until a fine mask image of the network carbide is obtained.
8. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 7, wherein: The rating of the network carbide based on the structural data of the network carbide in the mask image includes: Traversing each contour target in the fine mask image of the network carbide and sequentially recording the structural information of each contour target; Judging whether there is a closed network in the fine mask image of the network carbide based on the hit-or-miss transform with different kernel functions, and obtaining the number p of chain-like contours and the number q of fold angles. If there is a closed network, the current fine mask image is rated as grade 4. If there is no closed network, the rating score of the current fine mask image is calculated through formula one, and the rating grade of the current fine mask image is determined based on the rating score; The first formula is as follows: , where grade is the rating score of the current fine mask image, is the total number of pixels of the skeleton lines of all carbide networks in the current fine mask image; x is the width value of the current fine mask image, y is the height value of the current fine mask image; p is the number of chains, is the contribution degree of the number of chains; q is the number of corners, is the contribution degree of the number of corners; n is the number of network carbides in the current fine mask image; If , the evaluation level of the current fine mask image is level 0; if , the evaluation level of the current fine mask image is level 1; if , the evaluation level of the current fine mask image is level 2 for the zero level; if , the evaluation level of the current fine mask image is level 2.5; where , , are the level thresholds of the rating function.
9. The intelligent rating method for the network carbide of high-carbon chromium bearing steel according to claim 8, wherein: In the above process, it is necessary to determine whether there is a closed network in the fine mask image of the reticular carbide based on the hit-or-miss transform with different kernel functions, and obtain the number of chain segments p and the number of inflection angles q of the contour. If there is a closed network, the current fine mask image is rated as level 4. If there is no closed network, the rating score of the current fine mask image is calculated through Formula 1, and after determining the rating level of the current fine mask image based on the rating score, the following steps are also included: Determine the reticular carbide image with the highest rating level in the target high-carbon chromium bearing steel, and use the rating level of its corresponding fine mask image as the reticular carbide rating level of the target high-carbon chromium bearing steel.
10. The intelligent rating method for the reticular carbide of high-carbon chromium bearing steel according to claim 9, wherein: After the above step of determining the reticular carbide image with the highest rating level in the target high-carbon chromium bearing steel and using the rating level of its corresponding fine mask image as the reticular carbide rating level of the target high-carbon chromium bearing steel, the following steps are also included: Store the reticular carbide rating data of the target high-carbon chromium bearing steel. The reticular carbide rating data of the target high-carbon chromium bearing steel includes: the number of the target high-carbon chromium bearing steel, the number of the reticular carbide image corresponding to the target high-carbon chromium bearing steel, the rating level corresponding to each reticular carbide image, and the reticular carbide rating level of the target high-carbon chromium bearing steel.