A method for identifying primary and secondary α phases in titanium alloys

Abstract

This invention discloses a method for identifying primary and secondary α phases in titanium alloys. This method combines image processing techniques based on grayscale distribution characteristics, morphological features, and skeletal structure characteristics. It achieves high-precision identification and stable quantitative analysis of primary and secondary α phases in the microstructure of titanium alloys without requiring a large amount of labeled data, thus improving analytical efficiency and result reliability.

Description

Technical Field

[0001] This invention relates to the field of titanium alloy identification technology, specifically to a method for identifying primary α phase and secondary α phase in titanium alloys. Background Technology

[0002] The morphology, size, and distribution characteristics of the α phase in titanium alloys have a significant impact on the material's strength, plasticity, and fatigue properties. Primary α phases typically exhibit an equiaxed or blocky distribution, while secondary α phases often display slender needle-like or lamellar structures. Accurate identification and quantitative analysis of different types of α phases are crucial for studying the microstructure evolution and property control of titanium alloys. Current image analysis techniques for titanium alloy microstructures mainly include general image processing software methods and machine learning-based identification methods. On the one hand, general image processing software (such as Image-Pro Plus) typically relies on grayscale thresholding and simple morphological operations, demonstrating some ability to identify primary α phases with larger sizes and regular morphologies. However, it performs poorly in identifying secondary α phases that are small, elongated, and have low contrast, often exhibiting problems such as breakage, adhesion, or missed detection, making accurate extraction difficult. On the other hand, while machine learning or deep learning-based methods have certain advantages in complex image recognition, they typically rely on large amounts of high-quality labeled data for training, resulting in high data acquisition costs and complex model construction. Furthermore, these methods are sensitive to image quality, noise levels, and imaging conditions. When noise or contrast variations exist in microscopic images, the stability of the recognition results is poor, limiting their widespread application in engineering practice. In addition, traditional image analysis methods based on thresholding and morphological processing usually require manual setting of multiple parameters, which need to be repeatedly adjusted under different image conditions. This process is complex, has poor repeatability, and struggles to meet the alpha phase recognition needs of different scales and morphological features. Summary of the Invention

[0003] To address the aforementioned shortcomings in the prior art, the present invention provides a method for identifying primary α phase and secondary α phase in titanium alloys, which solves the problem that the prior art cannot accurately identify primary α phase and secondary α phase in titanium alloys.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for identifying primary α phase and secondary α phase in titanium alloys is provided, comprising the following steps: S1. Obtain the original microscopic image of the titanium alloy and perform image preprocessing to obtain the preprocessed grayscale image; S2. Perform α-phase preliminary segmentation on the grayscale image to obtain a binary image; S3. Perform opening, closing, and small region removal operations on the binary image to obtain the α-phase region; where the small region is a region with an area smaller than a set threshold. S4. Separate the α-phase region into multiple independent regions; S5. Extract the physical topological features of each independent region, and extract the primary α phase based on the physical topological features to obtain the primary α phase region; S6. Remove the primary α phase region from the grayscale image to obtain the candidate region; S7. Perform line structure enhancement on the candidate region to obtain the enhanced response map; S8. Extract continuous linear structures from the enhanced response map to obtain a line structure map; S9. Skeletonize the candidate region, prune the endpoints, break the bifurcation points and remove the short skeleton to obtain the secondary α phase skeleton; S10. Based on the line structure diagram and the secondary α phase skeleton, a confined reconstruction is performed to obtain the secondary α phase region.

[0005] Furthermore, the image preprocessing in step S1 is as follows: grayscale normalization and contrast enhancement are performed on the original microscopic image.

[0006] Furthermore, in step S2, the grayscale image is initially segmented into α phase by using the Otsu method to perform threshold segmentation on the grayscale image.

[0007] Furthermore, in step S3, the opening operation is: performing erosion and then dilation on the binary image using a structuring element; the closing operation is: performing dilation and then erosion on the binary image using a structuring element; and the small region removal is: marking the connected regions of the binary image, calculating the pixel area of ​​each region, and deleting regions with an area smaller than a set threshold.

[0008] Furthermore, step S4, which separates the α-phase region into multiple independent regions, includes the following sub-steps: S41. Perform Euclidean distance transformation on the α phase region to obtain the distance transformation diagram; S42. Extract local maxima points from the distance transformation graph to generate marker points; S43. Using the negative values ​​of the distance transformation map as the terrain surface, and the marker points as the initial seeds, the watershed method is used to segment the α phase region. S44. Perform boundary smoothing and small region merging on the segmentation results to obtain multiple independent regions.

[0009] Furthermore, the physical topological features in step S5 include: area, aspect ratio, eccentricity, solidity, skeleton length, average thickness, and thinness.

[0010] Furthermore, in step S7, a Meijering filter is used to enhance the line structure of the candidate region. Furthermore, in step S8, the hysteresis threshold segmentation method is used to extract continuous linear structures from the enhanced response map.

[0011] Further, in step S9, skeletonization is as follows: keeping the candidate region topology unchanged, the boundary pixels of the candidate region are gradually removed using an iterative erosion method, shrinking the candidate region into a skeleton with a single pixel width; endpoint pruning is as follows: based on the path length from the skeleton endpoint to the nearest branch point or end point, redundant branches of the skeleton are pruned; branch point disconnection is as follows: the branch points of the skeleton are detected and the skeleton is disconnected into multiple skeleton segments at the branch points; short skeleton removal is as follows: the length of all skeleton segments is counted and skeleton segments with a length less than a set threshold are deleted.

[0012] Further, in step S10, the confined reconstruction based on the line structure diagram and the secondary α phase skeleton is performed as follows: the candidate region is transformed by Euclidean distance and the width of the secondary α phase is calculated based on the secondary α phase skeleton and the line structure diagram; reconstruction is performed in the neighborhood of the secondary α phase skeleton according to the width of the secondary α phase.

[0013] The beneficial effects of this invention are as follows: 1. It has achieved accurate identification of the fine and elongated structures of the secondary α phase, solving the problems of missed detection, false detection and inaccurate boundary identification, and meeting the needs of fine quantitative analysis.

[0014] 2. It reduces the requirements for image quality, has strong generalization ability under different imaging conditions, is simple in process, has good repeatability, and is easy to apply stably in practical engineering.

[0015] 3. By identifying the morphological differences between primary and secondary α phases in titanium alloys through layered identification, accurate classification and independent quantitative statistics of primary and secondary α phases were achieved. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method. Figure 2 The labeled results are from Experiment 1; Figure 3 The annotation results are for Experiment 2; Figure 4 The labeled results are for comparison experiments. Detailed Implementation

[0017] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0018] like Figure 1 As shown, a method for identifying primary α phase and secondary α phase in titanium alloys is provided, comprising the following steps: S1. Obtain the original microscopic image of the titanium alloy and perform image preprocessing to obtain the preprocessed grayscale image; S2. Perform α-phase preliminary segmentation on the grayscale image to obtain a binary image; S3. Perform opening, closing, and small region removal operations on the binary image to obtain the α-phase region; where the small region is a region with an area smaller than a set threshold. S4. Separate the α-phase region into multiple independent regions; S5. Extract the physical topological features of each independent region, and extract the primary α phase based on the physical topological features to obtain the primary α phase region; S6. Remove the primary α phase region from the grayscale image to obtain the candidate region; S7. Perform line structure enhancement on the candidate region to obtain the enhanced response map; S8. Extract continuous linear structures from the enhanced response map to obtain a line structure map; S9. Skeletonize the candidate region, prune the endpoints, break the bifurcation points and remove the short skeleton to obtain the secondary α phase skeleton; S10. Based on the line structure diagram and the secondary α phase skeleton, a confined reconstruction is performed to obtain the secondary α phase region.

[0019] In this embodiment, image preprocessing includes grayscale normalization and contrast enhancement of the original microscopic image. Grayscale normalization involves mapping the 16-bit grayscale image to 8 bits using a percentile cropping method (lower limit 0.5% quantile value, upper limit 99.5% quantile value) to eliminate the influence of extreme bright and dark spots and improve consistency between different images. Contrast enhancement employs the CLAHE method (Contras Limited Adaptive Histogram Equalization) to improve the contrast between the α phase and the β matrix while preserving local detail structure. The block size is 8×8, and the contrast limiting factor is 2.0.

[0020] In this embodiment, the Otsu method is used to perform preliminary α-phase segmentation on the grayscale image to extract a binary image, followed by morphological opening and closing operations and small region removal. Opening operation: A structuring element (preferably a 3×3 or 5×5 square structuring element) is used to perform erosion followed by dilation on the binary image to remove isolated noise points and small pseudo-structures. Closing operation: A structuring element is used to perform dilation followed by erosion on the binary image to fill small holes within the α-phase region and improve region connectivity. Small region removal: Connected regions in the binary image are marked, the pixel area of ​​each region is calculated, and regions with an area smaller than a set threshold (less than 50 pixels or the corresponding actual size threshold) are deleted to remove noise interference. Through these operations, a continuous α-phase region with less noise is obtained.

[0021] In this embodiment, Euclidean distance transformation and watershed method are used to separate mutually adhered α-phase particles. Specifically, it includes: (1) performing Euclidean distance transformation on the α-phase region so that the value of each pixel represents its distance to the nearest background pixel, thereby forming a local maximum region at the center of each particle and obtaining a distance transformation map. (2) extracting the local maximum points in the distance transformation map as the center position of the potential particles, which are used to generate marker points. (3) using the negative values ​​of the distance transformation map as the terrain surface, and using the marker points as the initial seeds, the watershed method is used to segment the α-phase region. (4) performing boundary smoothing and small region merging on the segmentation results to avoid over-segmentation, thereby separating the mutually contacting α-phase particles into multiple independent regions.

[0022] In this embodiment, each independent region is identified based on the connected component labeling method. The physical topological features of each independent region are calculated, including: area, aspect ratio, eccentricity, solidity, skeleton length, average thickness, and elongation. These features are used to filter out candidate phases with elongated structures or those that do not conform to the topological features of the primary α phase, thereby obtaining blocky or equiaxed primary α phase regions. The primary α phase regions are then removed from the grayscale image to obtain candidate regions.

[0023] In this embodiment, the Meijering filter is used to enhance the line structure of the candidate region and Gaussian smoothing is applied to enhance the secondary alpha phase structure while suppressing noise. The Meijering filter is a multi-scale line structure enhancement method. Proposed by Erik Meijering, it is a multi-scale ridge enhancement method used to highlight linear or tubular structures in image processing and is widely used in biomedical image analysis, such as the detection and segmentation of blood vessels, nerve fibers, or microtubules. Here, it is used to enhance the elongated dark bar structures in the candidate region, making the secondary alpha phase more prominent.

[0024] In this embodiment, a hysteresis thresholding method is used to extract continuous linear structures from the enhanced response map. Specifically, a high threshold and a low threshold are set to divide the pixels in the enhanced response map into strong response pixels, weak response pixels, and background pixels. Strong response pixels are directly retained as reliable seed points for the linear structure, weak response pixels are selected as candidate pixels, and background pixels are directly discarded. Weak response pixels connected to strong response pixels are considered valid parts of the linear structure and retained, while isolated weak response pixels are discarded as noise. This method avoids the linear structure breakage problem caused by single thresholding and effectively suppresses noise interference, ultimately extracting a continuous, complete, and topologically accurate linear structure.

[0025] In this embodiment, the accurate location of the secondary α-phase skeleton is obtained through skeletonization, endpoint pruning, bifurcation point disconnection, and short skeleton removal. Skeletonization involves refining the candidate region while maintaining its topology, using an iterative erosion method to gradually remove boundary pixels, shrinking the candidate region to a single-pixel-width skeleton. Endpoint pruning involves calculating the path length from the skeleton endpoint to the nearest bifurcation point or endpoint; based on Burgers orientation relation, variant selection effect, lowest interface energy and least resistance path, and diffusion-controlled one-dimensional growth, redundant branches of the skeleton are pruned (usually within a few pixel units to avoid incorrectly pruning the basketweave structure). Burgers orientation relation: The phase transition from body-centered cubic β-phase to close-packed hexagonal α-phase must follow (0001). a / / {110} (Parallel base plane), <11 0> a / / <1 1> (Parallel directions). This indicates that the α phase can only nucleate on the {110} crystal plane of the β phase, and along... <111> Crystalline growth. Theoretically, there are only 12 possible α variants within a single β grain. Once an orientation is selected, the growth direction is locked. Variant selection effect: The 1-2 α variants with the lowest energy precipitate preferentially during nucleation, suppressing other variants. The first precipitated α plates create an elastic strain field within the β matrix, repelling the nucleation of other α plates with the same orientation in the same region. All α plates within the same cluster have completely consistent orientations, without branching crystallographic conditions. Lowest interface energy and least resistance path: The α / β interface is a coherent / semi-coherent interface, with the lowest strain energy and interface energy. Along <111> During growth (in the close-packed direction of the β phase), atomic arrangement matching is optimal, interfacial energy is lowest, and resistance is minimal. Lateral growth is much slower than longitudinal growth, resulting in needle-like and plate-like morphologies with high aspect ratios. Diffusion-controlled one-dimensional growth: Secondary α precipitation is a diffusion-type phase transition (redistribution of solutes such as Al, V, and Mo). The longitudinal (long axis) solute diffusion path is the shortest and fastest. Lateral (sideways) expansion and new nucleus formation are suppressed due to high interfacial energy and high strain field. Basketball net structure: Multi-oriented secondary α phase bundles intersect and cut each other, not as single branches, but as multiple intersecting branches forming a basketball net-like appearance. Bifurcation point disconnection: Detect the bifurcation points of the skeleton (i.e., pixels whose length is greater than the length of redundant branches in the endpoint pruning and whose number of connected neighbors is greater than 2) and disconnect the skeleton into multiple skeleton segments at the bifurcation points to avoid erroneous connections between different secondary α phases. Short skeleton removal: Count the length of all skeleton segments and delete skeleton segments whose length is less than a set threshold. Images smaller than 10 pixels often represent the remaining portion after pruning and branching, and are not the actual secondary α phase. In this embodiment, a threshold of 10 pixels is set to preserve the continuous structure with actual physical meaning. After the above processing, a continuous, independent, and less noisy secondary α phase skeleton structure can be obtained for subsequent confined width reconstruction and calculation.

[0026] In this embodiment, the confined reconstruction based on the line structure diagram and the secondary α phase skeleton is as follows: the candidate region is subjected to Euclidean distance transformation and the width of the secondary α phase is calculated based on the secondary α phase skeleton and the line structure diagram; reconstruction is performed in the neighborhood of the secondary α phase skeleton according to the width of the secondary α phase.

[0027] To further illustrate the effectiveness of this method, Experiments 1 and 2 were conducted according to this method, and a comparative experiment was carried out using the image segmentation method based on grayscale histograms in Image-ProPlus software.

[0028] Experiment 1: Using scanning electron microscope (SEM) images of TiAlVMoFeCr alloy after high-pressure aging as the subject, microstructure parameters were extracted from the backscattered electron images without corrosion treatment. Experimental conditions: Heat treatment process: high-pressure aging; Imaging mode: scanning electron microscope (SEM) BSE mode; magnification: 20K. The image characteristics at this stage are: small grayscale difference between the α phase and the matrix, low contrast of the secondary α phase, and low overall image noise. Figure 2 The labeling results of Experiment 1 are shown. Under non-corrosion conditions, the secondary α phase can still be continuously extracted, and the primary α phase and secondary α phase can be effectively distinguished. The identification results are highly consistent with the manual labeling results.

[0029] Experiment 2: Using scanning electron microscope (SEM) images of TiAlVFeCr titanium alloy after atmospheric pressure aging and etching, microstructure parameters were extracted from the backscattered electron images. Experimental conditions: Heat treatment: atmospheric pressure aging; Etching treatment: chemical etching solution treatment; Imaging mode: scanning electron microscope (SEM) BSE mode; Magnification: 40K. The images at this stage showed high contrast between the α phase and the matrix, clear boundaries of the secondary α phase, and localized uneven corrosion and noise. Adaptive adjustments were made to the alloy microstructure under atmospheric pressure aging, and the threshold for removing short skeletons was appropriately lowered. For example... Figure 3 The annotation results for Experiment 2 are shown. In the corrosion image, the boundary of the secondary α phase is more clearly identified. This indicates that the method can effectively suppress the interference caused by corrosion noise. Compared with the annotation results of Experiment 1, it shows good stability under different imaging conditions, indicating that the method has good adaptability and robustness.

[0030] Comparative Experiment: Image segmentation based on grayscale histograms in Image-Pro Plus software was used to process microscopic images of titanium alloys. Processing Method: After enhancing the image using a Gauss filter, the grayscale threshold range was manually selected through the histogram interface in the image segmentation function to extract low-grayscale regions as α-phase regions and generate a binary image. Subsequently, morphological processing and region measurements were performed on the segmentation results to obtain tissue parameters. For example... Figure 4 The results shown are the annotation results of the comparative experiment. In the comparative experiment, the threshold selection relied on human experience, and different images needed to be adjusted repeatedly, resulting in poor stability. For small secondary α phases, since their grayscale is close to that of the matrix, they are difficult to distinguish in the histogram, and are prone to missed detection or breakage. The segmented images still need to be measured manually or semi-automatically.

[0031] In actual scanning electron microscope (SEM) images, problems such as noise, uneven contrast, and local blurring often exist, leading to unstable recognition results using traditional methods. This invention's recognition process does not rely on single grayscale information but combines structural information and spatial constraints, thereby reducing the impact of noise and brightness fluctuations and achieving stable recognition results even under complex imaging conditions. Employing a layered recognition strategy, unlike existing technologies that uniformly identify different types of α phases, this method effectively avoids mutual interference between primary and secondary α phases through layer-by-layer screening and progressive refinement, improving the accuracy and stability of identifying fine secondary α phases. Introducing physical topological features effectively characterizes the essential structural differences of α phases with different morphologies, resulting in higher discriminative power and robustness. Introducing spatial constraints ensures that the structure reconstruction process occurs only within the skeleton's neighborhood, effectively avoiding the problem of erroneous expansion into noisy or non-target regions, thus improving morphology reconstruction accuracy while maintaining structural continuity. A multi-scale line structure enhancement method is employed to process images, combined with a dual-threshold hysteresis segmentation method to extract continuous linear structures. This enables the extraction of slender structures under complex backgrounds and noise conditions, significantly improving the continuity and recognition integrity of the secondary α phase. Pruning techniques are used to remove short branches, break bifurcation points, and eliminate short skeletons to optimize the structural topology, effectively removing noise-induced pseudo-structures and improving the accuracy and stability of secondary α phase recognition.

Claims

1. A method for identifying primary α phase and secondary α phase in titanium alloys, characterized in that, Includes the following steps: S1. Obtain the original microscopic image of the titanium alloy and perform image preprocessing to obtain the preprocessed grayscale image; S2. Perform α-phase preliminary segmentation on the grayscale image to obtain a binary image; S3. Perform opening, closing, and small region removal operations on the binary image to obtain the α-phase region; where the small region is a region with an area smaller than a set threshold. S4. Separate the α-phase region into multiple independent regions; S5. Extract the physical topological features of each independent region, and extract the primary α phase based on the physical topological features to obtain the primary α phase region; S6. Remove the primary α phase region from the grayscale image to obtain the candidate region; S7. Perform line structure enhancement on the candidate region to obtain the enhanced response map; S8. Extract continuous linear structures from the enhanced response map to obtain a line structure map; S9. Skeletonize the candidate region, prune the endpoints, break the bifurcation points and remove the short skeleton to obtain the secondary α phase skeleton; S10. Based on the line structure diagram and the secondary α phase skeleton, a confined reconstruction is performed to obtain the secondary α phase region.

2. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, The image preprocessing in step S1 is as follows: grayscale normalization and contrast enhancement are performed on the original microscopic image.

3. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, In step S2, the grayscale image is initially segmented into α phases by using the Otsu method to perform threshold segmentation on the grayscale image.

4. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, In step S3, the opening operation is performed by eroding and then dilating the binary image using a structuring element; the closing operation is performed by dilating and then eroding the binary image using a structuring element; and the small region removal operation is performed by marking the connected regions of the binary image, calculating the pixel area of ​​each region, and deleting regions with an area smaller than a set threshold.

5. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, Step S4, which separates the α-phase region into multiple independent regions, includes the following sub-steps: S41. Perform Euclidean distance transformation on the α phase region to obtain the distance transformation diagram; S42. Extract local maxima points from the distance transformation graph to generate marker points; S43. Using the negative values ​​of the distance transformation map as the terrain surface, and the marker points as the initial seeds, the watershed method is used to segment the α phase region. S44. Perform boundary smoothing and small region merging on the segmentation results to obtain multiple independent regions.

6. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, The physical topological features in step S5 include: area, aspect ratio, eccentricity, solidity, skeleton length, average thickness, and thinness.

7. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, In step S7, the Meijering filter is used to enhance the line structure of the candidate region.

8. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, In step S8, the hysteresis threshold segmentation method is used to extract continuous linear structures from the enhanced response map.

9. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that, In step S9, skeletonization is performed as follows: keeping the topology of the candidate region unchanged, the boundary pixels of the candidate region are gradually removed by iterative erosion method, and the candidate region is shrunk to a skeleton with a width of one pixel; endpoint pruning is performed as follows: based on the path length from the skeleton endpoint to the nearest branch point or end point, redundant branches of the skeleton are pruned; branch point disconnection is performed as follows: the branch points of the skeleton are detected and the skeleton is disconnected into multiple skeleton segments at the branch points. Short skeleton removal involves calculating the length of all skeleton segments and deleting those segments whose length is less than a set threshold.

10. The method for identifying primary α phase and secondary α phase in titanium alloys according to claim 1, characterized in that... In step S10, the confined reconstruction based on the line structure diagram and the secondary α phase skeleton is performed as follows: the candidate region is transformed by Euclidean distance and the width of the secondary α phase is calculated based on the secondary α phase skeleton and the line structure diagram. Reconstruction is performed within the neighborhood of the secondary α phase skeleton according to the width of the secondary α phase.

Images