Intelligent analysis method for banded carbides in steel

Abstract

The invention discloses an intelligent analysis method for banded carbides in steel, which comprises the following steps: acquiring a high-carbon steel carbide metallographic picture set, and establishing a carbide banded segmentation data set; training the carbide banded segmentation model based on the optimized DeepLabv3 + deep learning algorithm to obtain a trained banded carbide contour extraction model; inputting the high-carbon steel carbide metallographic diagram into the trained strip-shaped carbide contour extraction model to obtain a strip-shaped carbide contour; measuring and calculating according to the profile of the banded carbide to obtain a characteristic quantitative value of the banded carbide; establishing a banded carbide rating data set based on the banded carbide characteristic quantitative numerical value and the banded carbide grade; training a banded carbide rating model according to the banded carbide rating data set to obtain a trained banded carbide rating model; and inputting the quantitative numerical values of the strip-shaped carbide characteristics into the trained strip-shaped carbide rating model to obtain the grade of the strip-shaped carbide.

Description

Technical Field

[0001] The invention belongs to the field of metallographic analysis of steel, and in particular relates to an intelligent analysis method for banded carbides in steel. Background Art

[0002] Banded carbides originate from dendrite segregation during the solidification process of high carbon steel. Subsequent hot working and cooling cause secondary carbides to aggregate in specific areas to form banded structures. Banded carbides, as one of the main structural defects in steel, have a very adverse effect on the mechanical properties and service life of high carbon steel. Therefore, efficient and accurate evaluation of banded carbides is crucial for the quality assessment and control of high carbon steel. The severity of carbide banding is usually classified into different grades. The evaluation of carbide banding grade is mainly based on manual comparison of standard atlases. At present, the main domestic and foreign standards for the rating of banded carbides in steel include GB / T18254-2016, SEP 1520-1998 and ISO 5949-1983. All three of the above are rated by manually comparing the size, shape and particle aggregation of the carbide bands in the metallographic images to be evaluated with the corresponding information in the standard atlas. Obviously, manual comparison rating has problems such as large subjective influence, poor consistency of results, high professional requirements, slow experience accumulation, low detection efficiency, and high labor intensity.

[0003] With the development of artificial intelligence technology, automatic analysis methods for the microstructure of steel materials using machine learning algorithms continue to emerge. For example, the invention with application number 202310823487.2 discloses an automatic analysis method for banded carbides in bearing steel, the invention with application number 202311123480.6 discloses an intelligent rating method for low-magnification loose images of steel, and the invention with application number 202310585920.3 discloses an intelligent rating method for low-magnification cracks in steel continuous casting billets based on machine vision. The main idea of ​​the above methods is to use deep learning methods to intelligently identify and segment the contours of organizational features such as banded carbides, looseness, and cracks, and on this basis, quantify the above typical organizations, and then establish a quantitative rating index system based on artificial experience for automatic rating. Although artificial intelligence algorithms are used in microstructure identification, the final rating process still relies heavily on experience to manually divide the rating range, and no real intelligence is achieved. For example, in the invention with application number 202310823487.2, the width of the banded profile and the particle size are averaged for level division, and then the carbide area ratio is considered for correction (when the carbide ratio is less than 30%, the level is reduced by 0.5, when the ratio is 30-60%, the result remains unchanged, and when the ratio is greater than 60%, the level is increased by 0.5). For another example, in the invention with application number 202311123480.6, the loose score Tz is calculated based on complex parameters such as the loose range score and weight, the loose void score, and the void density score and weight, and then a loose rating index system is established based on artificial experience. Tz < 10 is level 0.5, Tz ≥ 80 is level 4, and 10 ≤ Tz < 80 is level 0.05Tz. In the invention with application number 202310585920.3, a complex formula is proposed to calculate the comprehensive crack score, and then the comprehensive score range of different crack levels is divided for rating based on artificial experience.

[0004] In summary, efficient and accurate evaluation of banded carbides is crucial for the quality assessment and control of high carbon steel. However, the existing manual comparison rating method has the problems of large subjective influence, high professional requirements and low detection efficiency. Although the automatic identification of tissues using artificial intelligence algorithms has promoted the automated development of carbide evaluation, the final rating process still mainly relies on the experience-based quantitative rating index system, the rating process of banded carbides is limited in intelligence, and the accuracy of automatic rating needs to be further improved. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides an intelligent analysis method for banded carbides in steel, which solves the problems of the existing manual comparison and rating method, such as large subjective influence, high professional requirements and low detection efficiency.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: an intelligent analysis method of banded carbides in steel, comprising the following steps:

[0007] S1. Obtain a metallographic picture set of high carbon steel carbides, mark the carbide band contours, and establish a carbide band segmentation data set;

[0008] S2. Perform data enhancement on the carbide band segmentation data set, train the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm, and obtain a trained band carbide contour extraction model;

[0009] S3, inputting the metallographic image of high carbon steel carbide into the trained band carbide contour extraction model to obtain the contour of the band carbide;

[0010] S4, measuring and calculating according to the profile of the band-shaped carbide to obtain a quantitative value of the characteristics of the band-shaped carbide;

[0011] S5. Establishing a banded carbide rating data set based on the banded carbide characteristic quantitative values ​​and banded carbide grades;

[0012] S6. Training a banded carbide rating model according to the banded carbide rating data set to obtain a trained banded carbide rating model;

[0013] S7. Input the quantified numerical values ​​of the characteristics of the banded carbides into the trained banded carbide rating model to obtain the grade of the banded carbides.

[0014] Further: S1 comprises the following sub-steps:

[0015] S11, collecting a number of metallographic pictures of high carbon steel carbides and establishing a metallographic picture set of high carbon steel carbides;

[0016] S21. Based on the metallographic image set of high carbon steel carbides, LabelMe is used to annotate each carbide band contour in the image, and the annotation file of the metallographic image is obtained to form a carbide band segmentation dataset.

[0017] Further: In S2, the method for performing data enhancement on the carbide band segmentation data set is specifically:

[0018] The carbide band segmentation dataset is enhanced by horizontal mirror image flipping and vertical mirror image flipping, where the image rotation angles include 90 degrees, 180 degrees and 270 degrees.

[0019] Further: in said S2, the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm includes an encoder and a decoder connected to each other;

[0020] The encoder includes a MobileNetv2 network, a first attention mechanism module, an ASPP dilated convolution spatial pyramid pooling module, a first convolution module, and a second attention mechanism module connected in sequence;

[0021] The decoder includes a second convolution module, a Concat layer, a third convolution module, and a first upsampling layer connected in sequence. The Concat layer is also connected to the second attention mechanism module through the second upsampling layer.

[0022] Among them, the first convolution module and the second convolution module are both 1×1 convolution, the third convolution module is 3×3 convolution, the first attention mechanism module and the second attention mechanism module both introduce the coordinate attention mechanism, and the first upsampling layer and the second upsampling layer both use the CARAFE upsampling algorithm.

[0023] Further: In said S2, the method for obtaining the trained band carbide contour extraction model is specifically as follows:

[0024] S21, inputting the carbide band segmentation dataset after data enhancement into the MobileNetv2 network to obtain low-level features and high-level features;

[0025] S22, inputting the high-level features into the ASPP dilated convolution spatial pyramid pooling module through the first attention mechanism module, performing 1×1 convolution operation, 3×3 dilated convolution operation with rate 6, 3×3 dilated convolution operation with rate 12, 3×3 dilated convolution operation with rate 18 and global average pooling respectively through the ASPP dilated convolution spatial pyramid pooling module to obtain 5 outputs, connecting the 5 outputs and sequentially inputting them into the first convolution module and the second attention mechanism module to obtain an intermediate feature map;

[0026] S23. Input the low-level features into the second convolution module to adjust the dimension to obtain the first feature map, input the intermediate feature map into the second upsampling layer to obtain the second feature map, connect the first feature map and the second feature map through the Concat layer, and the connected feature map passes through the third convolution module and the first upsampling layer in turn to obtain the output result of the band carbide contour extraction model, optimize the output result through the joint loss function of Cross-Entropy loss and Dice loss, and obtain the trained band carbide contour extraction model.

[0027] Further: in said S4, the quantitative values ​​of the characteristics of the banded carbides include the maximum band width of the banded carbides, the number of bands of the banded carbides, the total length of the banded carbides, the maximum particle size, the average particle distance and the sum of the number of particles and the particle distance;

[0028] The maximum band width of the band-like carbides is measured by a linear scanning method, wherein the maximum band width of the band-like carbides is extracted from the outline of the band-like carbides in a metallographic image of high carbon steel carbides at a magnification of 100 times.

[0029] Further: carbide particle diameter, number and distance are quantified by the minimum circumscribed circle method, wherein the maximum particle size, average particle distance and the sum of particle number and particle distance are extracted from the outline of the banded carbides in the metallographic image of high carbon steel carbides at 500 times magnification.

[0030] Furthermore: before quantifying the diameter, number and distance of carbide particles by the minimum circumscribed circle method, each particle is framed by a minimum circumscribed rectangle according to the outline of the banded carbide, and if the aspect ratio of the rectangle of the framed particle is greater than a preset aspect ratio threshold, the particle is removed.

[0031] Further: In S6, the band carbide rating model includes Random Forest, Catboost, Decision Tree and Gradient Boosting Decision Tree classifiers.

[0032] The beneficial effects of the present invention are:

[0033] (1) The present invention aims to solve the problems of large subjective influence, high professional requirements and low detection efficiency in the analysis of banded carbides in high carbon steel, and provides an intelligent analysis method for banded carbides in steel. Compared with the current intelligent evaluation method for microstructure in steel, the present invention not only uses a deep learning algorithm to realize the intelligent segmentation of banded carbides, but also uses a machine learning classifier to realize the intelligent classification of the grade of banded carbides, thereby greatly improving the detection efficiency and accuracy of carbide structure in steel, and significantly improving the degree of intelligence.

[0034] (2) The present invention designs an intelligent analysis method for banded carbides in steel. The overall framework adopts a full-process automated evaluation method of intelligent segmentation of banded carbides → quantitative analysis of carbide characteristics → intelligent calculation of banded carbide grade. This method can greatly improve the recognition and rating accuracy of banded carbide structures in high-carbon steel, reduce the detection cost, and ultimately achieve efficient and accurate evaluation of banded carbides in high-carbon steel. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The figure is a schematic flow chart of an intelligent analysis method of banded carbides in steel according to the present invention.

[0036] Figure 2 It is a schematic diagram of the intelligent segmentation of band-shaped carbides in the present invention.

[0037] Figure 3 It is a schematic diagram of the quantitative details of the ribbon-shaped carbides in the present invention.

[0038] Figure 4 This is a graph showing the prediction accuracy results of the four embodiments of the present invention in the test set. DETAILED DESCRIPTION

[0039] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0040] Embodiment 1:

[0041] like Figure 1 As shown, in one embodiment of the present invention, an intelligent analysis method for banded carbides in steel comprises the following steps:

[0042] S1. Obtain a metallographic picture set of high carbon steel carbides, mark the carbide band contours, and establish a carbide band segmentation data set;

[0043] S2. Perform data enhancement on the carbide band segmentation data set, train the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm, and obtain a trained band carbide contour extraction model;

[0044] S3, inputting the metallographic image of high carbon steel carbide into the trained band carbide contour extraction model to obtain the contour of the band carbide;

[0045] S4, measuring and calculating according to the profile of the band-shaped carbide to obtain a quantitative value of the characteristics of the band-shaped carbide;

[0046] S5. Establishing a banded carbide rating data set based on the banded carbide characteristic quantitative values ​​and banded carbide grades;

[0047] S6. Training a banded carbide rating model according to the banded carbide rating data set to obtain a trained banded carbide rating model;

[0048] S7. Input the quantified numerical values ​​of the characteristics of the banded carbides into the trained banded carbide rating model to obtain the grade of the banded carbides.

[0049] In this embodiment, S1 to S3 belong to the method of intelligent segmentation of banded carbides, S4 belongs to the method of quantitative analysis of carbide characteristics, and S5 to S7 belong to the method of intelligent calculation of banded carbide levels.

[0050] The S1 comprises the following sub-steps:

[0051] S11, collecting a number of metallographic pictures of high carbon steel carbides and establishing a metallographic picture set of high carbon steel carbides;

[0052] S21. Based on the metallographic image set of high carbon steel carbides, LabelMe is used to annotate each carbide band contour in the image, and the annotation file of the metallographic image is obtained to form a carbide band segmentation dataset.

[0053] In this embodiment, 871 metallographic images of high carbon chromium bearing steel carbides are collected, and LabelMe is used to annotate each carbide band contour in the image, and the annotation files of the metallographic images are obtained to form a carbide band segmentation data set.

[0054] In S2, the method for performing data enhancement on the carbide band segmentation data set is specifically as follows:

[0055] The carbide band segmentation dataset is enhanced by horizontal mirror image flipping and vertical mirror image flipping, where the image rotation angles include 90 degrees, 180 degrees and 270 degrees.

[0056] In S2, the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm includes an encoder and a decoder connected to each other;

[0057] The encoder includes a MobileNetv2 network, a first attention mechanism module, an ASPP dilated convolution spatial pyramid pooling module, a first convolution module, and a second attention mechanism module connected in sequence;

[0058] The decoder includes a second convolution module, a Concat layer, a third convolution module, and a first upsampling layer connected in sequence. The Concat layer is also connected to the second attention mechanism module through the second upsampling layer.

[0059] Among them, the first convolution module and the second convolution module are both 1×1 convolution, the third convolution module is 3×3 convolution, the first attention mechanism module and the second attention mechanism module both introduce the coordinate attention mechanism, and the first upsampling layer and the second upsampling layer both use the CARAFE upsampling algorithm.

[0060] In this embodiment, the overall architecture of the optimized DeepLabv3+ deep learning algorithm is:

[0061] The pre-trained convolutional neural network MobileNetv2 is used as the backbone feature extraction network for optimizing DeepLabv3+. The coordinate attention mechanism (CA) is introduced. The CARAFE upsampling algorithm is used to replace the bilinear interpolation upsampling algorithm in the DeepLabv3+ decoder part. The joint loss function of Cross-Entropy loss and Dice loss is used as the training loss function of the optimized DeepLabv3+.

[0062] In S2, the method for obtaining the trained band carbide contour extraction model is specifically as follows:

[0063] S21, inputting the carbide band segmentation dataset after data enhancement into the MobileNetv2 network to obtain low-level features and high-level features;

[0064] S22, inputting the high-level features into the ASPP dilated convolution spatial pyramid pooling module through the first attention mechanism module, performing 1×1 convolution operation, 3×3 dilated convolution operation with rate 6, 3×3 dilated convolution operation with rate 12, 3×3 dilated convolution operation with rate 18 and global average pooling respectively through the ASPP dilated convolution spatial pyramid pooling module to obtain 5 outputs, connecting the 5 outputs and sequentially inputting them into the first convolution module and the second attention mechanism module to obtain an intermediate feature map;

[0065] S23. Input the low-level features into the second convolution module to adjust the dimension to obtain the first feature map, input the intermediate feature map into the second upsampling layer to obtain the second feature map, connect the first feature map and the second feature map through the Concat layer, and the connected feature map passes through the third convolution module and the first upsampling layer in turn to obtain the output result of the band carbide contour extraction model, optimize the output result through the joint loss function of Cross-Entropy loss and Dice loss, and obtain the trained band carbide contour extraction model.

[0066] In said S4, the quantitative values ​​of the characteristics of the banded carbides include the maximum band width of the banded carbides, the number of bands of the banded carbides, the total length of the banded carbides, the maximum particle size, the average particle distance, and the sum of the number of particles and the particle distance;

[0067] The maximum band width of the band-like carbides is measured by a linear scanning method, wherein the maximum band width of the band-like carbides is extracted from the outline of the band-like carbides in a metallographic image of high carbon steel carbides at a magnification of 100 times.

[0068] The carbide particle diameter, number, and distance were quantified by the least circumscribed circle method, in which the maximum particle size, average particle distance, and the sum of the number of particles and the particle distance were extracted from the outlines of the banded carbides in the metallographic image of high carbon steel carbides at 500× magnification.

[0069] Before quantifying the diameter, number and distance of carbide particles by the minimum circumscribed circle method, each particle is framed by a minimum circumscribed rectangle according to the outline of the banded carbide. If the aspect ratio of the rectangle of the framed particle is greater than a preset aspect ratio threshold, the particle is removed.

[0070] In this embodiment, before performing the minimum circumscribed circle method, in order to exclude the influence of the network carbide particles, each particle is first framed by a minimum circumscribed rectangle, and a judgment threshold is set: if the aspect ratio of the rectangle is greater than 1.5, it is considered to be a network carbide and is then excluded.

[0071] like Figure 4 As shown, in this embodiment, the band carbide rating model adopts Random Forest, and the accuracy of band carbide rating is 92%, the precision is 92.87%, the recall rate is 92%, and the F1 score is 92.21%

[0072] Embodiment 2:

[0073] The difference between this embodiment and embodiment 1 is that the banded carbide rating model adopts the Catboost algorithm, such as Figure 4 As shown, the accuracy of the ribbon carbide rating is 94%, the precision is 94.69%, the recall is 94%, and the F1 score is 94.19%.

[0074] Embodiment 3:

[0075] The difference between this embodiment and embodiment 1 is that the banded carbide rating model adopts Decision Tree, and the banded carbide rating accuracy (Accuracy) is 92.67%, the precision (Precision) is 93.42%, the recall rate is 92.67%, and the F1 score is 92.72%.

[0076] Embodiment 4:

[0077] The difference between this embodiment and embodiment 1 is that the band carbide rating model adopts GradientBoosting Decision Tree, and the accuracy (Accuracy) of band carbide rating is 94%, the precision (Precision) is 95.01%, the recall rate is 94%, and the F1 score is 94.31%.

[0078] The beneficial effects of the present invention are as follows: In view of the problems of large subjective influence, high professional requirements and low detection efficiency in the analysis of banded carbides in high carbon steel, the present invention provides an intelligent analysis method for banded carbides in steel. Compared with the current intelligent evaluation method for microstructure in steel, the present application not only utilizes a deep learning algorithm to realize the intelligent segmentation of banded carbides, but also utilizes a machine learning classifier to realize the intelligent classification of the grade of banded carbides, thereby greatly improving the detection efficiency and accuracy of carbide structure in steel, and significantly improving the degree of intelligence.

[0079] The present invention designs an intelligent analysis method for banded carbides in steel. The overall framework adopts a full-process automated evaluation method of intelligent segmentation of banded carbides → quantitative analysis of carbide characteristics → intelligent calculation of banded carbide levels. This method can greatly improve the recognition and rating accuracy of banded carbide structures in high-carbon steel, reduce detection costs, and ultimately achieve efficient and accurate evaluation of banded carbides in high-carbon steel.

[0080] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.

Claims

1. An intelligent analysis method for banded carbides in steel, characterized in that: The following steps are involved: S1. Obtain a metallographic picture set of high carbon steel carbides, mark the carbide band contours, and establish a carbide band segmentation data set; S2. Perform data enhancement on the carbide band segmentation data set, train the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm, and obtain a trained band carbide contour extraction model; S3, inputting the metallographic image of high carbon steel carbide into the trained band carbide contour extraction model to obtain the contour of the band carbide; S4, measuring and calculating according to the profile of the band-shaped carbide to obtain a quantitative value of the characteristics of the band-shaped carbide; S5. Establishing a banded carbide rating data set based on the banded carbide characteristic quantitative values ​​and banded carbide grades; S6. Training a banded carbide rating model according to the banded carbide rating data set to obtain a trained banded carbide rating model; S7. Input the quantified numerical values ​​of the characteristics of the banded carbides into the trained banded carbide rating model to obtain the grade of the banded carbides.

2. The intelligent analysis method of banded carbides in steel according to claim 1, characterized in that: The S1 comprises the following sub-steps: S11, collecting a number of metallographic pictures of high carbon steel carbides and establishing a metallographic picture set of high carbon steel carbides; S21. Based on the metallographic image set of high carbon steel carbides, LabelMe is used to annotate each carbide band contour in the image, and the annotation file of the metallographic image is obtained to form a carbide band segmentation dataset.

3. The intelligent analysis method of banded carbides in steel according to claim 1, characterized in that: In S2, the method for performing data enhancement on the carbide band segmentation data set is specifically as follows: The carbide band segmentation dataset is enhanced by horizontal mirror image flipping and vertical mirror image flipping, where the image rotation angles include 90 degrees, 180 degrees and 270 degrees.

4. The intelligent analysis method of banded carbides in steel according to claim 1, characterized in that: In S2, the carbide band segmentation model based on the optimized DeepLabv3+ deep learning algorithm includes an encoder and a decoder connected to each other; The encoder includes a MobileNetv2 network, a first attention mechanism module, an ASPP dilated convolution spatial pyramid pooling module, a first convolution module, and a second attention mechanism module connected in sequence; The decoder includes a second convolution module, a Concat layer, a third convolution module, and a first upsampling layer connected in sequence. The Concat layer is also connected to the second attention mechanism module through the second upsampling layer. Among them, the first convolution module and the second convolution module are both 1×1 convolution, the third convolution module is 3×3 convolution, the first attention mechanism module and the second attention mechanism module both introduce the coordinate attention mechanism, and the first upsampling layer and the second upsampling layer both use the CARAFE upsampling algorithm.

5. The intelligent analysis method of banded carbides in steel according to claim 4, characterized in that: In S2, the method for obtaining the trained band carbide contour extraction model is specifically as follows: S21, inputting the carbide band segmentation dataset after data enhancement into the MobileNetv2 network to obtain low-level features and high-level features; S22, inputting the high-level features into the ASPP dilated convolution spatial pyramid pooling module through the first attention mechanism module, performing 1×1 convolution operation, 3×3 dilated convolution operation with rate 6, 3×3 dilated convolution operation with rate 12, 3×3 dilated convolution operation with rate 18 and global average pooling respectively through the ASPP dilated convolution spatial pyramid pooling module to obtain 5 outputs, connecting the 5 outputs and sequentially inputting them into the first convolution module and the second attention mechanism module to obtain an intermediate feature map; S23. Input the low-level features into the second convolution module to adjust the dimension to obtain the first feature map, input the intermediate feature map into the second upsampling layer to obtain the second feature map, connect the first feature map and the second feature map through the Concat layer, and the connected feature map passes through the third convolution module and the first upsampling layer in turn to obtain the output result of the band carbide contour extraction model, optimize the output result through the joint loss function of Cross-Entropy loss and Dice loss, and obtain the trained band carbide contour extraction model.

6. The intelligent analysis method of banded carbides in steel according to claim 1, characterized in that: In said S4, the quantitative values ​​of the characteristics of the banded carbides include the maximum band width of the banded carbides, the number of bands of the banded carbides, the total length of the banded carbides, the maximum particle size, the average particle distance, and the sum of the number of particles and the particle distance; The maximum bandwidth of the band-like carbides is measured by a linear scanning method, wherein the maximum bandwidth of the band-like carbides is extracted from the outline of the band-like carbides in a metallographic image of high carbon steel carbides at a magnification of 100 times.

7. The intelligent analysis method of banded carbides in steel according to claim 6, characterized in that: The carbide particle diameter, number, and distance were quantified by the least circumscribed circle method, in which the maximum particle size, average particle distance, and the sum of the number of particles and the particle distance were extracted from the outlines of the banded carbides in the metallographic image of high carbon steel carbides at 500× magnification.

8. The intelligent analysis method of banded carbides in steel according to claim 7, characterized in that: Before quantifying the diameter, number and distance of carbide particles by the minimum circumscribed circle method, each particle is framed by a minimum circumscribed rectangle according to the outline of the banded carbide. If the aspect ratio of the rectangle of the framed particle is greater than a preset aspect ratio threshold, the particle is removed.

9. The intelligent analysis method of banded carbides in steel according to claim 1, characterized in that: In S6, the band carbide rating model includes Random Forest, Catboost, Decision Tree and Gradient BoostingDecision Tree classifiers.

Images