A method for identifying microstructure images of titanium alloy forgings

By segmenting and analyzing the connectivity domain features under different pixel thresholds in the microstructure images of titanium alloy forgings, combined with the convolutional neural network, the problem of inaccurate phase distribution quantification in microstructure recognition is solved, and more accurate microscopy is achieved Organizational feature identification and evaluation.

CN118658160BActive Publication Date: 2024-11-22BAOJI TOPUDA TITANIUM IND CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411139498.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-22
Estimated Expiration
2044-08-20

AI Technical Summary

Technical Problem

During the microstructure identification process of titanium alloy forgings, grain analysis in different regions may lead to inaccurate quantification results of phase distribution, affecting the identification and evaluation of microstructure.

Method used

By segmenting microstructure images under different pixel thresholds, analyzing the pixel distribution, shape characteristics and spatial distribution of the connecting domain, screening out the optimal segmentation situation, determining the final pixel threshold for microstructure recognition, and combining with convolutional neural network for β phase and the distribution degree analysis of α phase.

Benefits of technology

It improves the accuracy of judgment of different phase boundaries, ensures accurate identification and evaluation of microstructure characteristics, and enhances the reliability of the identification process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118658160B_ABST
    Figure CN118658160B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of image recognition and processing technology, and specifically to a method for identifying microstructure images of titanium alloy forgings. The method divides a microstructure image into sub-images through different preset pixel thresholds, and obtains the microstructure characteristics of each sub-image according to the pixel distribution and shape edge characteristics of the connected domain in each sub-image, as well as the local spatial distribution of each connected domain; the final pixel threshold is determined according to the microstructure characteristics of the sub-image under each preset pixel threshold, and the microstructure is identified with the microstructure characteristics under the final pixel threshold. The present invention combines the distribution characteristics of different phases in the image, screens out the optimal boundary through multi-segmentation analysis, and performs feature analysis, so that the quantitative feature identification of different phases is more accurate and the identification of microstructure is more reliable.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of image recognition processing, and in particular to a method for recognizing microstructure images of titanium alloy forgings. Background Art

[0002] Titanium alloy forgings refer to titanium alloy parts made by forging. Forging is a manufacturing process that changes the physical shape of metal by applying pressure. This process can not only enhance the mechanical properties of the material, but also improve the microstructure. The performance of titanium alloys corresponding to different microstructures is different. Identifying the microstructure in the material through microstructure images can help determine the type of phase in the titanium alloy and quantify the distribution and size of the phase, so as to evaluate the mechanical properties and performance of the material.

[0003] When the microstructure of titanium alloy forgings is obtained by microscopic image identification, different types of microstructures have different image characteristics. The microstructure of titanium alloy is mainly composed of α phase and β phase. Since the phases can be transformed during the forging process of the forging itself, the distribution boundaries between different phases are not obvious enough. In the identification process, errors may occur in the grain analysis of different regions, which will lead to inaccurate quantitative results of phase distribution and affect the identification and evaluation of the microstructure. Summary of the invention

[0004] In order to solve the technical problem that in the prior art, errors may occur in the analysis of grains in different regions during the identification process, thereby leading to inaccurate quantitative results of phase distribution, the purpose of the present invention is to provide a method for microstructure image recognition of titanium alloy forgings, and the technical scheme adopted is as follows:

[0005] The present invention provides a method for identifying a microstructure image of a titanium alloy forging, the method comprising:

[0006] Acquire a microstructure image of a sample titanium alloy forging; segment the microstructure image into two sub-images at each preset pixel threshold;

[0007] In each sub-image, the microstructure characteristics of each sub-image are obtained according to the uniform distribution of pixels and the complexity of shape boundaries of each connected domain, as well as the compact distribution of local connected domains of each connected domain;

[0008] Based on the difference in microstructural features between the two sub-images at each preset pixel threshold, the final pixel threshold of the microstructural image is screened out; at the final pixel threshold, microstructural recognition is performed based on the microstructural features of the two sub-images corresponding to the microstructural image.

[0009] Furthermore, the method for acquiring the microstructure characteristics includes:

[0010] In each sub-image, the pixel features of each connected domain are obtained according to the numerical distribution and chaotic variation of the overall pixel values ​​of the pixels in each connected domain;

[0011] In each sub-image, the shape feature of each connected domain is obtained according to the boundary prominence in each connected domain and the shape distribution of the connected domain;

[0012] In each sub-image, the spatial features of each connected domain are obtained according to the distribution ratio and distribution proximity of other connected domains in the local range of each connected domain;

[0013] The pixel features, shape features and spatial features of each connected domain in each sub-image are combined to obtain the local features of each connected domain in each sub-image; the microstructure features of each sub-image are obtained by combining the local features of all connected domains in each sub-image.

[0014] Furthermore, the method for acquiring pixel features includes:

[0015] The average pixel value of all pixels in each connected domain in each sub-image is used as the pixel value distribution degree of each connected domain;

[0016] The pixel value variance of all pixels in each connected domain in each sub-image is used as the pixel chaos distribution degree of each connected domain;

[0017] The pixel features of each connected domain are obtained by combining the pixel value distribution and pixel chaos distribution of each connected domain in each sub-image. The pixel value distribution is positively correlated with the pixel features, while the pixel chaos distribution is negatively correlated with the pixel features.

[0018] Furthermore, the method for acquiring the shape feature includes:

[0019] Obtain the edge of each connected domain in each sub-image; take the total number of pixels in each connected domain in each sub-image as the distribution area of ​​each connected domain;

[0020] Combine the gradient value of each edge point on the edge of each connected domain to obtain the boundary saliency of each connected domain;

[0021] Obtain the circularity of each connected domain of each sub-image; take the ratio of the total number of edge points of each connected domain to the distribution area of ​​the connected domain as the perimeter area ratio of each connected domain;

[0022] The shape saliency of each connected domain is obtained by combining the circularity and perimeter-to-area ratio of each connected domain. The circularity is positively correlated with the shape saliency, while the perimeter-to-area ratio is negatively correlated with the shape saliency.

[0023] The shape features of each connected domain of each sub-image are obtained by combining the boundary saliency and shape saliency of each connected domain of each sub-image. Both the boundary saliency and the shape saliency are positively correlated with the shape features.

[0024] Furthermore, the method for acquiring the spatial features includes:

[0025] Get the center point of each connected domain in each sub-image;

[0026] In each sub-image, the total number of center points within a preset local range of each connected domain is counted to obtain local distribution data of each connected domain;

[0027] The local position distance of each connected domain is obtained by combining the distance between the center point of each connected domain and other center points within the preset local range;

[0028] The spatial characteristics of each connected domain are obtained by combining the local distribution quantity and local position distance of each connected domain. The local distribution quantity is positively correlated with the spatial characteristics, while the local position distance is negatively correlated with the spatial characteristics.

[0029] Furthermore, the method for obtaining the final pixel threshold comprises:

[0030] The microstructure feature difference between the two sub-images segmented by each preset pixel threshold is used as the feature difference degree of each preset pixel threshold;

[0031] The preset pixel threshold of the maximum feature difference is used as the final pixel threshold.

[0032] Furthermore, the microstructure recognition is performed according to the microstructure features of the two sub-images corresponding to the microstructure image under the final pixel threshold, including:

[0033] Under the final pixel threshold, the sub-image with the lowest average pixel value among the sub-images is used as the β phase analysis image, and the other sub-image is used as the α phase analysis image;

[0034] The microstructure characteristics of the β phase analysis image are negatively correlated and normalized to obtain the β phase characteristic index of the microstructure image; the proportion of the number of pixels in the connected domain of the β phase analysis image in the microstructure image is used as the β phase proportion index; the β phase characteristic index and the β phase proportion index are combined to obtain the β phase distribution degree of the microstructure image;

[0035] The microstructure characteristics of the α phase analysis image are normalized to obtain the α phase characteristic index of the microstructure image; the proportion of the number of pixels in the connected domain of the α phase analysis image in the microstructure image is used as the α phase proportion index; the α phase characteristic index and the α phase proportion index are combined to obtain the α phase distribution degree of the microstructure image;

[0036] Microstructure identification is performed based on the distribution of β phase and α phase.

[0037] Furthermore, the method for obtaining the local position distance includes:

[0038] The distance between the center point of each connected domain and each other center point within a preset range is used as the proximity distance between the center point of each connected domain and each other center point;

[0039] The sum of the proximity distances between the center point of each connected domain and all other center points is taken as the local position distance of each connected domain.

[0040] Furthermore, the segmenting of the microstructure image into two sub-images at each preset pixel threshold comprises:

[0041] For any preset pixel threshold, an image composed of pixel points in the microstructure image whose pixel values ​​are less than the preset pixel threshold is used as a sub-image, and an image composed of the remaining pixel points is used as another sub-image.

[0042] Furthermore, the microstructure identification based on the β phase distribution and the α phase distribution includes:

[0043] The β phase distribution and the α phase distribution are input into the trained convolutional neural network to output the microstructure recognition results.

[0044] The present invention has the following beneficial effects:

[0045] The present invention performs connected domain analysis of grains by segmenting sub-images at different pixel thresholds, taking into account the overall distribution differences of α-phase and β-phase, and improves the judgment of boundaries between different phases through multi-threshold segmentation analysis. The microstructural features are characterized from three aspects: the grayscale distribution and shape characteristics of the connected domain itself in each sub-image, and the spatial distribution of each connected domain, to obtain the microstructural features of each sub-image, thereby reflecting the microstructural feature characterization in each segmented image, and then determining the optimal segmentation from multiple segmentation results, determining the final pixel threshold for separate analysis of different phases, and performing microstructural identification with the microstructural features at the final pixel threshold. The present invention combines the distribution characteristics of different phases in the image, screens out the optimal boundary through multi-segmentation analysis for feature analysis, makes the quantitative feature identification of different phases more accurate, and makes the identification and evaluation of microstructures more reliable.

[0046] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of a titanium alloy forging microstructure image recognition method provided by one embodiment of the present invention;

[0049] Figure 2 A microstructure image provided by an embodiment of the present invention;

[0050] Figure 3 A flow chart of a method for acquiring microstructure characteristics provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a titanium alloy forging microstructure image recognition method proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0052] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] The specific scheme of the titanium alloy forging microstructure image recognition method provided by the present invention is described in detail below with reference to the accompanying drawings.

[0054] See also Figure 1 , which shows a flow chart of a titanium alloy forging microstructure image recognition method provided by an embodiment of the present invention, the method comprising the following steps:

[0055] S1: Acquire a microstructure image of a sample titanium alloy forging; segment the microstructure image into two sub-images at each preset pixel threshold.

[0056] In the embodiment of the present invention, the process of acquiring the microstructure image of the titanium alloy forging sample mainly includes: cutting: cutting a representative small sample from the parent material, and the size of the small sample is usually convenient for handling and observation. Embedding: embedding the small sample into a plastic or metal base suitable for microscopic observation to ensure that the sample is stable and easy to manipulate. Grinding and polishing: using sandpaper to gradually grind the sample surface from coarse to fine, and then using polishing paste to polish until a smooth mirror effect is obtained. Etching: selecting a suitable etchant according to the material of the sample, applying the etchant evenly on the sample surface, controlling the action time, and then washing with water or other appropriate liquids, and finally drying. Etching is to treat the sample surface by chemical or electrochemical methods so as to more clearly show the details of the microstructure under a microscope. Microscopic observation: using an optical microscope or a scanning electron microscope for observation, adjusting the appropriate magnification, finding the area of ​​interest for detailed observation, and the magnification that can be selected in this embodiment is 20 times. Image acquisition and analysis: taking photos of the microstructure under a microscope, transmitting the acquired photos to the data processing terminal for analysis, performing image preprocessing on the photos, and obtaining a microstructure image.

[0057] It should be noted that the image preprocessing process may specifically include image grayscale processing, filtering and denoising processing, and background removal processing. It should be noted that the image preprocessing process is a technical means well known to those skilled in the art. Specifically, grayscale weighting method may be used for grayscale conversion, bilateral filtering method for filtering and denoising, and background difference method for background removal, etc., which will not be described in detail here. Figure 2 , which shows a microstructure image provided by an embodiment of the present invention.

[0058] The phase distribution of the titanium alloy microstructure, that is, the grain distribution, is related to the performance of titanium alloy forgings. Due to the different alloy materials such as iron, copper or aluminum, the alloy properties will be different. The proportion of other elements added to the base metal, such as carbon, nickel and titanium, will significantly affect the properties of the alloy. Different elements have different effects on the hardness, toughness and corrosion resistance of the alloy. Heat treatment processes such as annealing, quenching and tempering, as well as other processing techniques such as forging and rolling, will change the crystal structure and grain size of the alloy, thereby affecting its mechanical properties and other physical properties.

[0059] The microstructure of titanium alloy is mainly composed of α phase and β phase. α phase is the main component of titanium alloy. Its content affects the corrosion resistance and ductility of titanium alloy. It is a stable phase in titanium alloy. β phase is an unstable phase. Its content affects the strength and hardness of titanium alloy. β phase dissolves at high temperature and gradually precipitates α phase for transformation as the temperature decreases. α alloy is mainly composed of α phase, with stable structure and excellent mechanical properties, and has good corrosion resistance and high temperature resistance. β alloy is mainly composed of β phase, with high strength and hardness, and has sufficient toughness and fatigue resistance. α+β alloy is the most common titanium alloy. While maintaining good mechanical properties, it also maintains a certain processing plasticity.

[0060] In order to describe the boundaries of different phases in the microstructure image more accurately, the image is segmented. Since the α phase and the β phase have different grayscale expressions in the image, the grayscale of the α phase is usually higher than that of the β phase, and in the image where the α phase and the β phase coexist, the characteristics of each phase are different. Therefore, the image is divided into two parts for analysis through the pixel threshold. At each preset pixel threshold, the microstructure image is segmented into two sub-images.

[0061] In some achievable ways of the embodiments of the present invention, for any preset pixel threshold, an image composed of pixels in the microstructure image whose pixel values ​​are less than the preset pixel threshold is used as a sub-image, and an image composed of the remaining pixels is used as another sub-image. In the embodiments of the present invention, the specific process of segmenting to obtain a new image may be: establishing two new images of the same size as the microstructure image, traversing the pixels in the microstructure image, analyzing the relationship with the preset pixel threshold, and when the pixel value is less than the preset pixel threshold, copying and storing the corresponding pixel in the corresponding position of the new image to obtain a sub-image, and similarly allocating the pixel values ​​greater than or equal to the preset pixel threshold to another new image to obtain another sub-image.

[0062] In a specific implementation of an embodiment of the present invention, the selection range of the preset pixel threshold is between 50 and 150, and the pixel value is selected every 10. The implementer of the specific selection method can limit it according to the specific implementation scenario, and no limitation is made here.

[0063] S2: In each sub-image, the microstructure characteristics of each sub-image are obtained according to the uniform distribution of pixels and the complexity of shape boundaries of each connected domain, as well as the compact distribution of local connected domains of each connected domain.

[0064] In view of the characteristic differences of different microstructures of titanium alloys, the characteristics of each sub-image are characterized. Under ideal circumstances, one sub-image after segmentation represents the characteristics of the α phase and the other represents the characteristics of the β phase. Based on the difference between the two phases, the more ideal the segmentation state is, the greater the difference in the overall microstructure characteristics of the image. Therefore, for each sub-image, the pixel value shape represented by the connected domain and the different situations in the global spatial distribution are analyzed. It should be noted that the acquisition of the connected domain in the image is a technical means well known to those skilled in the art, such as obtaining the connected domain by using regional growing, etc., which will not be elaborated or limited here.

[0065] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining microstructure characteristics can be found in Figure 3 , which shows a flow chart of a method for acquiring microstructure characteristics provided by an embodiment of the present invention, the method comprising the following steps:

[0066] S201: In each sub-image, according to the numerical distribution and chaotic variation of the overall pixel values ​​of the pixels in each connected domain, the pixel features of each connected domain are obtained.

[0067] The α phase usually represents a stable crystal structure and chemical composition in the image. The pixel value of the connected domain represented is high and the distribution is relatively uniform. It is the part with lighter brightness in the microstructure. The β phase contains more alloy elements or other microstructural features in the image, which will lead to lower pixel values ​​and fluctuating distribution. It is the part with darker brightness in the microstructure. Therefore, feature analysis is first performed based on the distribution of pixel values.

[0068] Preferably, in an embodiment of the present invention, the method for acquiring pixel features includes:

[0069] First, the mean pixel value of all pixels in each connected domain in each sub-image is taken as the pixel value distribution degree of each connected domain. The mean value reflects the overall distribution degree of pixel values ​​in the connected domain. The higher the pixel value distribution degree, the more significant the α phase feature is and the weaker the β phase feature is.

[0070] Secondly, the pixel value variance of all pixels in each connected domain in each sub-image is used as the pixel chaos distribution degree of each connected domain. The variance reflects the numerical chaos degree of the pixel values ​​in the connected domain. When the pixel chaos distribution degree is smaller, the α phase feature is more significant and the β phase feature is weaker. In other embodiments of the present invention, the standard deviation or coefficient of variation can also be used to reflect the distribution chaos of the numerical values, which will not be elaborated here.

[0071] Finally, the pixel features of each connected domain are obtained by combining the pixel value distribution and pixel chaos distribution of each connected domain in each sub-image. The pixel value distribution is positively correlated with the pixel features, and the pixel chaos distribution is negatively correlated with the pixel features. The larger the pixel value distribution, the smaller the pixel chaos distribution, and the larger the pixel features, which reflects that the α phase features in the connected domain are more significant and the β phase features are weaker.

[0072] In an embodiment of the present invention, the product of the value after negative correlation mapping and normalization of the pixel chaos distribution and the pixel value distribution is used as the pixel feature. In other embodiments of the present invention, the ratio of the pixel value distribution to the pixel chaos distribution can also be used as the pixel feature, which is not limited here.

[0073] In a specific implementation of the embodiment of the present invention, the expression of pixel features is:

[0074] ; In the formula, Expressed as The pixel features of the connected components, Expressed as The pixel value distribution of a connected domain, Expressed as The degree of pixel chaos distribution of a connected domain, Represents an exponential function with a natural constant as its base. It is expressed as an exponential function with a natural constant as the base for negative correlation mapping and normalization, so that the pixel chaos distribution degree is negatively correlated with the pixel characteristics.

[0075] At this point, the analysis of pixel distribution in the connected domain is completed.

[0076] S202: In each sub-image, according to the boundary prominence in each connected domain and the shape distribution of the connected domain, a shape feature of each connected domain is obtained.

[0077] In the microstructure, the shape of the α phase is closer to spherical or equiaxed, and is closer to a circle than the elongated strip of the β phase. At the same time, since the α phase represents a more stable crystal structure, its boundary is simpler and clearer, while the boundary of the more unstable β phase is more complex and more blurred. Therefore, the analysis can be carried out based on the shape distribution of the connected domain and the significance of the boundary.

[0078] Preferably, in an embodiment of the present invention, the method for acquiring shape features includes:

[0079] First, the edge of each connected domain in each sub-image is obtained, and the total number of pixels in each connected domain in each sub-image is used as the distribution area of ​​each connected domain. Further analysis is performed based on the obtained edges and area ratios. It should be noted that the acquisition of the edges of connected domains is a technical means well known to technicians in this field and will not be elaborated here.

[0080] Furthermore, the boundary significance of each connected domain is obtained by combining the gradient value of each edge point on the edge of each connected domain. The edge clarity is reflected by the size of the edge gradient value. When the gradient value is higher, the edge boundary is clearer, the α phase feature is more significant, and the β phase feature is weaker. It should be noted that the method for obtaining the gradient value is a technical means well known to those skilled in the art, such as using the Sobel operator, etc., which will not be described in detail here.

[0081] In an embodiment of the present invention, the gradient values ​​of all edge points may be combined by accumulating them to obtain the boundary saliency. In other embodiments of the present invention, the gradient values ​​of all edge points may be combined by averaging to obtain the boundary saliency, which will not be described in detail here.

[0082] Furthermore, the circularity of each connected domain of each sub-image is obtained. The circularity reflects the degree to which the connected domain is circular. The higher the circularity, the more significant the α phase feature of the connected domain and the weaker the β phase feature. The ratio of the total number of edge points of each connected domain to the distribution area of ​​the connected domain is used as the perimeter-to-area ratio of each connected domain. The perimeter of the connected domain is reflected by the length of the edge points, and the complexity of the boundary is reflected by the perimeter-to-area ratio. When the perimeter-to-area ratio is smaller, the edge of the connected domain is simpler, which may reflect an equiaxed area, and the α phase feature is more significant and the β phase feature is weaker.

[0083] Furthermore, the shape saliency of each connected domain is obtained by combining the circularity and perimeter area ratio of each connected domain. The circularity is positively correlated with the shape saliency, and the perimeter area ratio is negatively correlated with the shape saliency. When the circularity is larger and the perimeter area ratio is smaller, it means that the shape in the connected domain is closer to a circle and the boundary distribution is simpler, reflecting that the characteristics of the α phase in the connected domain are more significant and the characteristics of the β phase are weaker. In the embodiment of the present invention, the ratio of the circularity to the perimeter area ratio is used as the shape saliency in the form of a ratio.

[0084] Finally, the shape features of each connected domain of each sub-image are obtained by combining the boundary saliency and shape saliency of each connected domain of each sub-image. The boundary saliency and shape saliency are positively correlated with the shape features. When the boundary saliency is larger and the shape saliency is larger, it means that the connected domain has clearer boundary distribution and simpler shape distribution, which reflects that the α phase feature in the connected domain is more significant and the β phase feature is weaker.

[0085] In the embodiment of the present invention, the product of the boundary saliency and the shape saliency is used as the shape feature. In other embodiments of the present invention, the sum of the boundary saliency and the shape saliency can also be used as the shape feature, which is not limited here.

[0086] At this point, the analysis of the shape edge conditions of each connected domain is completed.

[0087] S203: In each sub-image, the spatial features of each connected domain are obtained according to the distribution ratio and distribution proximity distance of other connected domains within the local range of each connected domain.

[0088] In the microstructure of titanium alloy, the α phase tends to have a higher phase structure density, so the local distribution of the connected domain in the image tends to be denser. Therefore, the spatial distribution characteristics of each connected domain can be analyzed on the global distribution of the connected domain.

[0089] Preferably, in an embodiment of the present invention, the method for acquiring shape features includes:

[0090] First, the center point of each connected domain in each sub-image is obtained, and the spatial position of the connected domain is reflected by the center point. The acquisition of the center point of the connected domain is a well-known technical means and will not be described in detail here.

[0091] Furthermore, in each sub-image, the total number of central points within a preset local range of each connected domain is counted to obtain local distribution data of each connected domain. The total number of central points distributed within the local range can be used to reflect the density of the local distribution. The larger the local distribution data, the more local connected domains are distributed, the denser the distribution of the local connected domains, the more significant the α-phase characteristics reflected by the connected domains, and the weaker the β-phase characteristics.

[0092] Furthermore, the local position distance of each connected domain is obtained by combining the distance between the center point of each connected domain and other center points within a preset local range. The degree of aggregation of the connected domain is reflected by the closeness of the distance and the distribution of the distance between the center points. In the embodiment of the present invention, the distance between the center point of each connected domain and each other center point within the preset range is used as the proximity distance between the center point of each connected domain and each other center point, and the sum of the proximity distances between the center point of each connected domain and all other center points is used as the local position distance of each connected domain. When the local position distance is smaller, it means that the distance between the center point and other center points is closer, the local connected domain distribution of the connected domain is more aggregated, the connected domain reflects the α phase characteristics more significantly, and the β phase characteristics are weaker.

[0093] Finally, the spatial characteristics of each connected domain are obtained by combining the number of local distributions and the local position distance of each connected domain. The number of local distributions is positively correlated with the spatial characteristics, and the local position distance is negatively correlated with the spatial characteristics. When the number of local distributions is larger and the local position distance is smaller, it means that the number of local connected domain distributions of the connected domain is larger and closer, and the α phase characteristics reflected by the connected domain are more significant, and the β phase characteristics are more significant. In an embodiment of the present invention, the product of the value of the local position distance that is negatively correlated and normalized and the number of local distributions is used as the spatial feature.

[0094] In a specific implementation of an embodiment of the present invention, the preset local range is set to a circular range with the center point of each connected domain as the center and the pixel point 25 as the side length. The implementer can adjust the specific range setting according to the specific implementation method and is not limited here.

[0095] At this point, the analysis of the spatial distribution of each connected domain is completed.

[0096] S204: combining the pixel features, shape features and spatial features of each connected domain in each sub-image to obtain the local features of each connected domain in each sub-image; combining the local features of all connected domains in each sub-image to obtain the microstructure features of each sub-image.

[0097] The three features of each connected domain are integrated, and the three features of all connected domains in each sub-image are integrated to obtain the microstructure features of each sub-image. In the embodiment of the present invention, the product of the pixel features, shape features and spatial features of each connected domain is normalized to obtain the local features of each connected domain. The microstructure features are comprehensively characterized in the form of products. When the pixel features are larger, the shape features are larger, and the spatial features are larger, that is, the local features are larger, the α-phase features of the connected domain are more significant, and the β-phase features are weaker. It should be noted that normalization is a technical means well known to those skilled in the art, and the selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0098] In other embodiments of the present invention, the sum of the pixel features, shape features and space features of each connected domain may be used as the local feature of each connected domain, and the three features may be integrated in the form of a sum, which will not be elaborated here.

[0099] In some achievable methods of the embodiments of the present invention, the cumulative value of the local features of all connected domains in each sub-image is used as the microstructure feature of each sub-image, and the feature conditions of all connected domains are combined in the form of accumulation. When the microstructure feature is larger, it means that the α phase feature in the sub-image is more significant, and conversely, when the microstructure feature is smaller, it means that the β feature in the sub-image is more significant. In other embodiments of the present invention, the average value of the local features of all connected domains can also be used as the microstructure feature of each sub-image, which will not be described in detail here.

[0100] At this point, the analysis of the microstructure features in each sub-image is completed.

[0101] S3: Based on the difference in microstructure features between the two sub-images at each preset pixel threshold, a final pixel threshold of the microstructure image is screened out; at the final pixel threshold, microstructure recognition is performed based on the microstructure features of the two sub-images corresponding to the microstructure image.

[0102] Each preset pixel threshold can divide the microstructure image into two different sub-images. The connected domain features reflected in different sub-images are also different. For each pixel threshold, the greater the difference in microstructure features between sub-images, the better the division of a-phase and β-phase under this segmentation, and the smaller the error result in the subsequent boundary analysis. Therefore, the difference in the representation features between sub-images can be analyzed under all preset pixel thresholds to screen the optimal threshold.

[0103] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the final pixel threshold includes:

[0104] The difference in microstructure characteristics between the two sub-images after segmentation by each preset pixel threshold is taken as the characteristic difference degree of each preset pixel threshold, reflecting the degree of characterization of different phase characteristics after segmentation. The preset pixel threshold of the maximum characteristic difference degree is taken as the final pixel threshold. The segmentation result of the final pixel threshold is used to further demarcate and analyze the characteristics of the two phases of the microstructure. It should be noted that, if there are extremely special cases where there are multiple preset pixel thresholds corresponding to the maximum characteristic difference degree, the smallest preset pixel threshold is taken as the final pixel threshold for segmentation analysis.

[0105] Under the segmentation of the final pixel threshold, the distribution characteristics of the phases can be quantified, which is convenient for subsequent identification and evaluation based on the characteristic distribution of the microstructure. In an embodiment of the present invention, under the final pixel threshold, the sub-image with the lowest average pixel value in the sub-image is used as the β phase analysis image, and the other sub-image is used as the α phase analysis image. The characteristics characterizing different phases are analyzed respectively through the two sub-images.

[0106] Furthermore, the microstructural features of the β-phase analysis image are negatively correlated and normalized to obtain the β-phase characteristic index of the microstructural image. The smaller the microstructural feature, the more significant the β feature. The β-phase analysis image is mainly analyzed to characterize the degree of the β-phase feature. The higher the β-phase characteristic index, the more significant the β phase in the image. The proportion of the number of pixels in the connected domain of the β-phase analysis image in the microstructural image is used as the β-phase proportion index, and the proportion of the connected domain reflects the distribution degree of the β-phase in the titanium alloy forging sample. The β-phase distribution degree of the microstructural image is obtained by combining the β-phase characteristic index and the β-phase proportion index. In an embodiment of the present invention, the product of the β-phase characteristic index and the β-phase proportion index is normalized as the β-phase distribution degree to comprehensively characterize the β-phase distribution in the microstructural image.

[0107] Furthermore, the microstructure characteristics of the α-phase analysis image are normalized to obtain the α-phase characteristic index of the microstructure image. The degree of characterization of the α-phase characteristics of the α-phase analysis image is analyzed. The larger the α-phase characteristic index, the more significant the α-phase in the image. The proportion of the number of pixels in the connected domain of the α-phase analysis image in the microstructure image is used as the α-phase proportion index, and the distribution degree of the α-phase in the titanium alloy forging sample is also reflected by the proportion of the connected domain. The α-phase characteristic index and the α-phase proportion index are combined to obtain the α-phase distribution degree of the microstructure image. In an embodiment of the present invention, the product of the α-phase characteristic index and the α-phase proportion index is normalized as the α-phase distribution degree to comprehensively characterize the α distribution in the microstructure image.

[0108] Finally, microstructure identification is performed based on the distribution of β phase and α phase. Microstructure identification is to evaluate the performance of titanium alloy by the distribution of β phase and α phase. In an embodiment of the present invention, the distribution of β phase and α phase can be input into a trained convolutional neural network to output the microstructure identification result. In a specific implementation of the embodiment of the present invention, an adversarial neural network can be selected as the structural main body, and sample data can be input into the neural network for training. The distribution of β phase and α phase of the microstructure image is used as sample data, and cross entropy is used as the loss function. The adaptive motion estimation algorithm is used for optimization to obtain a trained convolutional neural network. The subsequent identification process only needs to input the obtained distribution of β phase and α phase to obtain the microstructure identification result. It should be noted that the specific neural network structure, training algorithm process and adaptive motion estimation algorithm are technical means well known to those skilled in the art, and will not be described in detail here.

[0109] In other embodiments of the present invention, preliminary identification can also be performed directly based on the difference between the β phase distribution and the α phase distribution. In a specific implementation of the embodiment of the present invention, the minimum value of the β phase distribution and the α phase distribution is used as the low distribution standard. When the difference between the β phase distribution and the α phase distribution is greater than the low distribution standard, when the low distribution standard is the α phase distribution, the microstructure is characterized as a β alloy, when the low distribution standard is the β phase distribution, the microstructure is characterized as an α alloy, and when the difference between the β phase distribution and the α phase distribution is less than or equal to the low distribution standard, the microstructure is characterized as an α+β alloy. The identification process is not limited here.

[0110] It should be noted that, in order to facilitate calculation, all indicator data involved in the calculation in the embodiment of the present invention are preprocessed to eliminate the dimension effect. The specific means of eliminating the dimension effect are technical means well known to those skilled in the art and are not limited here.

[0111] In summary, the present invention performs connected domain analysis of grains by segmenting sub-images at different pixel thresholds, taking into account the overall distribution differences of α-phase and β-phase, and improves the judgment of boundaries between different phases through multi-threshold segmentation analysis. The microstructural characteristics are characterized from three aspects: the grayscale distribution and shape characteristics of the connected domain itself in each sub-image, and the spatial distribution of each connected domain. The microstructural characteristics of each sub-image are obtained, thereby reflecting the microstructural characteristics in each segmented image. The optimal segmentation is determined from multiple segmentation results, and the final pixel threshold is determined to perform separate analysis of different phases, and the microstructural characteristics under the final pixel threshold are used for microstructural identification. The present invention combines the distribution characteristics of different phases in the image, and screens out the optimal boundary through multi-segmentation analysis for feature analysis, so that the quantitative feature identification of different phases is more accurate, and the identification and evaluation of microstructures is more reliable.

[0112] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for identifying the microstructure image of titanium alloy forgings, It is characterized in that The method comprises: Acquire a microstructure image of a sample titanium alloy forging; segment the microstructure image into two sub-images at each preset pixel threshold; In each sub-image, the microstructure characteristics of each sub-image are obtained according to the uniform distribution of pixels and the complexity of shape boundaries of each connected domain, as well as the compact distribution of local connected domains of each connected domain; Based on the difference in microstructure features between the two sub-images at each preset pixel threshold, the final pixel threshold of the microstructure image is screened out; at the final pixel threshold, microstructure recognition is performed based on the microstructure features of the two sub-images corresponding to the microstructure image; The method for obtaining the microstructure characteristics comprises: In each sub-image, the pixel features of each connected domain are obtained according to the numerical distribution and chaotic variation of the overall pixel values ​​of the pixels in each connected domain; In each sub-image, the shape feature of each connected domain is obtained according to the boundary prominence in each connected domain and the shape distribution of the connected domain; In each sub-image, the spatial features of each connected domain are obtained according to the distribution ratio and distribution proximity of other connected domains in the local range of each connected domain; Combining the pixel features, shape features and spatial features of each connected domain in each sub-image, the local features of each connected domain in each sub-image are obtained; combining the local features of all connected domains in each sub-image, the microstructure features of each sub-image are obtained; The method for acquiring the pixel features includes: The average pixel value of all pixels in each connected domain in each sub-image is used as the pixel value distribution degree of each connected domain; The pixel value variance of all pixels in each connected domain in each sub-image is used as the pixel chaos distribution degree of each connected domain; The pixel features of each connected domain are obtained by combining the pixel value distribution and pixel chaos distribution of each connected domain in each sub-image. The pixel value distribution is positively correlated with the pixel features, while the pixel chaos distribution is negatively correlated with the pixel features. The method for acquiring the shape feature includes: Obtain the edge of each connected domain in each sub-image; take the total number of pixels in each connected domain in each sub-image as the distribution area of ​​each connected domain; Combine the gradient value of each edge point on the edge of each connected domain to obtain the boundary saliency of each connected domain; Obtain the circularity of each connected domain of each sub-image; take the ratio of the total number of edge points of each connected domain to the distribution area of ​​the connected domain as the perimeter area ratio of each connected domain; The shape saliency of each connected domain is obtained by combining the circularity and perimeter-to-area ratio of each connected domain. The circularity is positively correlated with the shape saliency, while the perimeter-to-area ratio is negatively correlated with the shape saliency. The shape features of each connected domain of each sub-image are obtained by combining the boundary saliency and shape saliency of each connected domain of each sub-image. The boundary saliency and shape saliency are both positively correlated with the shape features. The method for acquiring the spatial features includes: Get the center point of each connected domain in each sub-image; In each sub-image, the total number of center points within a preset local range of each connected domain is counted to obtain local distribution data of each connected domain; The local position distance of each connected domain is obtained by combining the distance between the center point of each connected domain and other center points within the preset local range; The spatial characteristics of each connected domain are obtained by combining the local distribution quantity and local position distance of each connected domain. The local distribution quantity is positively correlated with the spatial characteristics, while the local position distance is negatively correlated with the spatial characteristics.

2. According to claim 1, a titanium alloy forging microstructure image recognition method, It is characterized in that The method for obtaining the final pixel threshold comprises: The microstructure feature difference between the two sub-images segmented by each preset pixel threshold is used as the feature difference degree of each preset pixel threshold; The preset pixel threshold of the maximum feature difference is used as the final pixel threshold.

3. According to claim 1, a titanium alloy forging microstructure image recognition method, It is characterized in that The method of performing microstructure recognition according to microstructure features of two sub-images corresponding to the microstructure image under the final pixel threshold comprises: Under the final pixel threshold, the sub-image with the lowest average pixel value among the sub-images is used as the β phase analysis image, and the other sub-image is used as the α phase analysis image; The microstructure characteristics of the β phase analysis image are negatively correlated and normalized to obtain the β phase characteristic index of the microstructure image; the proportion of the number of pixels in the connected domain of the β phase analysis image in the microstructure image is used as the β phase proportion index; the β phase characteristic index and the β phase proportion index are combined to obtain the β phase distribution degree of the microstructure image; The microstructure characteristics of the α phase analysis image are normalized to obtain the α phase characteristic index of the microstructure image; the proportion of the number of pixels in the connected domain of the α phase analysis image in the microstructure image is used as the α phase proportion index; the α phase characteristic index and the α phase proportion index are combined to obtain the α phase distribution degree of the microstructure image; Microstructure identification is performed based on the distribution of β phase and α phase.

4. According to claim 1, a titanium alloy forging microstructure image recognition method, It is characterized in that The method for obtaining the local position distance includes: The distance between the center point of each connected domain and each other center point within a preset range is used as the proximity distance between the center point of each connected domain and each other center point; The sum of the proximity distances between the center point of each connected domain and all other center points is taken as the local position distance of each connected domain.

5. According to claim 1, a titanium alloy forging microstructure image recognition method, It is characterized in that The method of dividing the microstructure image into two sub-images at each preset pixel threshold comprises: For any preset pixel threshold, an image composed of pixel points in the microstructure image whose pixel values ​​are less than the preset pixel threshold is used as a sub-image, and an image composed of the remaining pixel points is used as another sub-image.

6. According to claim 3, a titanium alloy forging microstructure image recognition method, It is characterized in that The microstructure identification based on the β phase distribution and the α phase distribution includes: The β phase distribution and the α phase distribution are input into the trained convolutional neural network to output the microstructure recognition results.

Citation Information

Patent Citations

  • Method for analyzing and evaluating micro-nano particle dispersion and distribution

    CN103903266A

  • Method for analyzing pore connectivity of dense core

    CN109253956A

  • Chromium-titanium alloy treatment method and application of chromium-titanium alloy treatment method in metallographic structure display

    CN111579324A

  • Titanium alloy furniture production quality monitoring system

    CN116611748A

  • Method and device for detecting cleanliness of carbon steel raw material of bimetal composite board

    CN117095003A