Method for extracting ferrite microstructure edges based on microstructural images of steel.
The method addresses the limitations of existing methods by employing edge detection and reconstruction techniques to accurately extract and evaluate ferrite grain boundaries in steel microstructure images, improving precision in ferrite grain size determination.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- ANSTEEL BEIJING RES INST CO LTD
- Filing Date
- 2024-06-20
- Publication Date
- 2026-07-29
AI Technical Summary
Existing methods for extracting ferrite grain boundaries in steel microstructure images are inadequate, as they rely solely on grayscale intensity and lack consideration of texture patterns and edge direction, leading to unsatisfactory results and limited transfer learning ability.
A method involving edge detection, grain boundary closure, unnecessary boundary removal, and crystal grain reconstruction using multi-stage edge detection, morphological closing, and region adjacency graph analysis to accurately extract and reconstruct ferrite grain boundaries.
Enables precise extraction and evaluation of ferrite grain size in cold-rolled low-carbon steel sheets by effectively separating connected grains and reconstructing crystal structures, enhancing the accuracy of ferrite grain boundary detection.
Smart Images
Figure 2026525133000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of analysis and processing of metal microstructure images of steel, and particularly to a method for extracting ferrite grain boundaries based on a microstructure image of steel.
Background Art
[0002] Ferrite is an interstitial solid solution in which carbon is dissolved in α-Fe, and is generally represented by the symbol F. It has a body-centered cubic lattice structure, and its carbon solubility is very low. At room temperature, only 0.0008% of carbon can be dissolved, and the maximum carbon solubility at 727 °C is 0.02%. Austenite with a hypoeutectoid composition precipitates and forms ferrite by primary crystallization. Ferrite has smooth grain boundaries, and there are almost no twins or slip lines in the crystal. It is light green and shiny, darkens after deep etching, and exists in the form of plates, blocks, needles or meshes in steel. Such ferrite is called primary ferrite or structurally free ferrite. Depending on the formation conditions, primary ferrite has various forms such as equiaxed, spindle-shaped, serrated and needle-shaped. Ferrite is also the matrix of pearlite structure and is the main constituent phase in the hot rolling (annealing) and annealing structures of carbon steel and low alloy steel. The composition and structure of ferrite have a significant impact on the processing performance of steel, and in some cases, also on the service performance of steel.
[0003] The method of the present invention is mainly a method for extracting ferrite grain boundaries in a microstructure image of steel proposed based on the criteria of the morphological characteristics of grains specified in the current GB / T 4335-2013 "Method for Measuring Ferrite Grain Size of Cold-Rolled Thin Steel Sheets of Low Carbon Steel". On the one hand, the morphological characteristics of ferrite are complex, while the grain boundaries of ferrite in cold-rolled thin steel sheets of low carbon steel are clear and generally exist in the form of meshes, blocks and plates. Therefore, it is a technique for edge extraction for regions where the grain boundaries are not smooth.
[0004] Currently, most commercially available software generally performs discrimination based on the grayscale intensity of an image, but the results obtained are not satisfactory. It is necessary to consider information such as texture pattern, edge direction, and morphological pattern in addition to grayscale intensity. Some methods are limited to specific steel types, and their transfer learning ability is weak.
[0005] While there are methods that rely entirely on machine learning and deep learning, their segmentation and training processes are constrained by the training set and model, and therefore cannot fully solve the ferrite edge extraction problem. [Overview of the project] [Problems that the invention aims to solve]
[0006] The present invention provides a method for extracting ferrite structure edges based on microstructure images of steel, which enables accurate reconstruction of ferrite grains in microstructure images of steel and effective separation of two connected grains in a narrow region, thereby enabling the extraction of accurate ferrite grains in cold-rolled low-carbon steel sheets. Based on the extracted accurate ferrite grains of the cold-rolled low-carbon steel sheets, the method also provides an auxiliary method for evaluating the ferrite grain size of the cold-rolled low-carbon steel sheets. [Means for solving the problem]
[0007] To achieve the above objective, the present invention employs the following technical means.
[0008] A method for extracting ferrite microstructure edges based on microstructural images of steel, including the following steps. Step S1 for detecting edges: Edge detection is performed on the microstructure image of the cold-rolled low-carbon steel sheet, grain edge information is detected in the ferrite microstructure image, and a grain edge image is generated; Step S2 for closing grain boundaries: The grain edge image is sequentially treated by connecting broken edges, filling in gaps, and thinning grain boundaries to generate a grain-processed image; Step S3 to remove unnecessary boundaries: Relabel the crystal grains in the image using the region expansion method, define the outline of the labeled region as the crystal grain boundary, and select and label unlabeled pixel points one by one until all labeling is complete; Step S4 for reconstructing crystal grains: Based on the network topology of the image, unconnected portions of the labeled image are detected, and grain boundary edges are determined and reaggregated for the unclosed portions of the crystal grains. Using a region adjacency graph, the regions corresponding to the nodes and adjacent regions are connected by arcs, and while each vertex of the crystal grain remains unchanged, each grain boundary arc connected to that vertex is repeatedly scanned to determine whether the characteristics of a completely closed crystal grain are met, and the reconstruction of the crystal grain is completed. Step S5 involves inspecting the reconstruction and completing the extraction of ferrite tissue edges and subsequent reconstruction; and Step S6 generates an edge image of the ferrite phase and saves the resulting image data.
[0009] Furthermore, in step S1, which detects the edge, crystal grain edge information is detected using the multi-stage edge detection Canny algorithm.
[0010] Furthermore, step S2, which closes the grain boundaries, specifically includes the following steps. Step S2.1 involves processing the edge image using a morphological closing method, further thinning the edges, and reconnecting the broken edges; Step S2.2 involves calculating the area of each hole and, based on a threshold, filling holes with an area smaller than the threshold using a hole-filling method; and Step S2.3 involves thinning the grain boundaries using a skeletalization method.
[0011] Furthermore, the scanning model when repeatedly scanning each grain boundary arc connected to the vertex in step S4 is as follows. G=(V,E) (1) "In the formula, V is the vertex of a crystal grain having an unclosed portion, and E is the grain boundary arc connected to that vertex."
[0012] Furthermore, step S5, which involves inspecting the reconstruction and completing the extraction of the ferrite tissue edge and subsequent reconstruction, specifically includes the following steps: Step S5.1: Aggregating adjacent grain clusters of each grain to form a network topology; Step S5.2: Calculate the average grayscale of each crystal grain; Step S5.3 detects whether or not each crystal grain contains a crystal grain boundary; Step S5.4 involves determining that adjacent crystal grains that satisfy both of the following conditions belong to the same crystal grain: 1) the average grayscale is below a threshold, and 2) the image boundary is not included in the crystal grain. Re-reconstruction of crystal grains: If it is found that the above inspection conditions are met but the reconstruction is not correctly labeled, step S5.5 is performed to change the grain boundaries of the adjacent crystal grains to be identical; and Step S5.6: Output and save the image result after the processing described above. [Effects of the Invention]
[0013] Compared to the prior art, the present invention offers the following beneficial effects. 1) By reconstructing the crystal structure and utilizing the topology and pixel changes of the image, a method of scanning, reconstructing, analyzing, and re-reconstructing can be used to accurately reconstruct ferrite crystal grains in microstructural images of steel. 2) By effectively separating two crystal grains connected in a narrow region, it is possible to extract ferrite crystal grains from cold-rolled low-carbon steel sheets with precision. 3) Based on the accurately extracted ferrite grain size of the cold-rolled low-carbon steel sheets, the ferrite grain size of the cold-rolled low-carbon steel sheets is evaluated as an auxiliary method. [Brief explanation of the drawing]
[0014] [Figure 1]It is a diagram schematically showing a flowchart of the method of the present invention.
Embodiments for Carrying Out the Invention
[0015] Hereinafter, specific embodiments of the present invention will be further described with reference to the drawings. The following examples are implemented based on the technical means of the present invention and show detailed implementation methods and specific operation processes, but do not limit the protection scope of the present invention. The methods used in the following examples are conventional methods unless otherwise specified.
Example
[0016] FIG. 1 is a flowchart of the method of the present invention. The method for extracting ferrite grain boundaries based on the microstructure image of steel according to the present invention is used to measure the ferrite crystal grain size of cold-rolled thin sheets of low-carbon steel with a carbon content of less than 0.2%, and can accurately reconstruct the ferrite grains in the microstructure image of steel. Using a model, it is determined whether the grain boundaries are closed or not and whether they are multi-sided, the grains are reconstructed, and finally, the extraction and construction of the ferrite grain boundaries are completed. [[ID=I9]]
[0017] The method of the present invention uses the metal microstructure image of cold-rolled thin sheets of low-carbon steel as the data of this implementation. The method of the present invention includes the following steps.
[0018] Step S1 of detecting edges: Using the multi-stage edge detection canny algorithm, the grain boundary information of the ferrite microstructure image is detected for the metal microstructure image of the cold-rolled sheet of low-carbon steel, the noise of the grain image is removed using a Gaussian filter, the gradient width and direction of each pixel point in the image are calculated using the finite difference of the first-order partial derivative, the image is convolved using the first-order difference convolution template, non-maximum suppression is performed on the gradient width to obtain a thinned edge, the edge is detected and connected using the dual-threshold algorithm to generate a grain boundary image, and the preprocessed data is locally saved; Step S2 of closing grain boundaries: For the grain edge image, sequentially perform connection of discontinuous edges, hole filling, and thinning of grain boundaries to generate a grain processed image; Step S3 of removing unnecessary boundaries: In general edge detection algorithms, some false edges are detected. In this method, the grains in the image are relabeled using the region expansion method. Taking the points inside the grains as seed points, and recursively relabeling the pixel points inside the grains according to the region expansion method, the outer contour of the labeled region becomes the boundary of the grain. Then, until all relabeling is completed, select unlabeled pixel points and relabel them one by one; Step S4 of reconstructing grains: Detect the unconnected parts of the labeled image based on the network topology of the image, perform judgment and re-aggregation of grain boundary edges on the unclosed parts of the grains, use the region adjacency graph to connect the regions corresponding to the nodes and the adjacent regions with arcs, and repeatedly scan each grain boundary arc connected to the vertex in the situation where each vertex of the grain remains unchanged to determine whether it meets the characteristics of a completely closed grain, and complete the reconstruction of the grain; Step S5 of inspecting the reconstruction, extracting the ferrite structure edges, and completing the reconstruction again; and Step S6 of generating an edge image of the ferrite phase and saving the resulting image data.
Example
[0019] This example is the same as Example 1 except for the following points. In this embodiment, closing grain boundaries specifically includes the following: A determination of grain closing is made for the grain edge image, and for grain boundaries that are not closed, the edge image is processed using a morphological closing method. Here, a 2x2 size convolutional structure is used to enlarge the edges in the image and achieve the effect of edge expansion. Subsequently, the edges are thinned. In this way, broken edges can be reconnected. If there are discontinuities in the edges, the area of each hole is calculated, and based on a threshold, holes with an area smaller than the threshold are filled using a hole-filling method. Finally, to solve the problem of the grain edges becoming excessively thick due to the above processing, the grain boundaries are thinned using a skeletalization method to generate a grain processing image, which is then saved locally.
[0020] Reconstructing crystal grains specifically involves the following: detecting unconnected portions of the labeled image based on the image's network topology; determining and re-aggregating grain boundary edges in the unclosed portions of the crystal grains; connecting corresponding regions and adjacent regions with arcs using a region adjacency graph; and, while each vertex of the crystal grain remains unchanged, repeatedly scanning each grain boundary arc connected to a vertex to determine whether it satisfies the characteristics of a fully closed crystal grain, thereby completing the crystal grain reconstruction. The scanning model is as follows. G=(V,E) (2) "In the formula, V is the vertex of a crystal grain having an unclosed portion, and E is the grain boundary arc connected to that vertex." [Examples]
[0021] This embodiment is the same as Embodiment 2, except for the following points. In this embodiment, inspecting the reconstruction and completing the extraction and re-reconstruction of the ferrite tissue edge includes the following steps. Step S5.1: Aggregating adjacent grain clusters of each grain to form a network topology; Step S5.2: Calculate the average grayscale of each crystal grain; Step S5.3 detects whether or not each crystal grain contains a crystal grain boundary; Step S5.4 involves determining that adjacent crystal grains that satisfy both of the following conditions belong to the same crystal grain: 1) the average grayscale is below a threshold, and 2) the image boundary is not included in the crystal grain. Re-reconstruction of crystal grains: If it is found that the above inspection conditions are met but the reconstruction is not correctly labeled, step S5.5 is performed to change the grain boundaries of the adjacent crystal grains to be identical; and Step S5.6: Output and save the image result after the processing described above.
Claims
1. Step S1 for detecting edges: Edge detection is performed on the microstructure image of the cold-rolled low-carbon steel sheet, crystal grain edge information is detected in the ferrite microstructure image, and a crystal grain edge image is generated; Step S2 for closing grain boundaries: The grain edge image is sequentially treated by connecting broken edges, filling in gaps, and thinning grain boundaries to generate a grain-processed image; Step S3 to remove unnecessary boundaries: Relabel the crystal grains in the image using the region expansion method, define the outline of the labeled region as the crystal grain boundary, and select and label unlabeled pixel points one by one until all labeling is complete; Step S4 for reconstructing crystal grains: Based on the network topology of the image, unconnected parts of the labeled image are detected, and grain boundary edges are determined and re-aggregated for the unclosed parts of the crystal grains. Using a region adjacency graph, the regions corresponding to the nodes and adjacent regions are connected by arcs, and while each vertex of the crystal grain remains unchanged, each grain boundary arc connected to the vertex is repeatedly scanned to determine whether or not the characteristics of a completely closed crystal grain are met, and the reconstruction of the crystal grain is completed. Step S5 involves inspecting the reconstruction and completing the extraction of ferrite tissue edges and subsequent reconstruction; and A method for extracting ferrite structure edges based on a microstructure image of steel, characterized by including step S6 of generating an edge image of the ferrite phase and saving the resulting image data.
2. The method for extracting ferrite structure edges based on a microstructure image of steel according to claim 1, characterized in that, in step S1 for detecting the edge, crystal grain edge information is detected using a multi-stage edge detection canny algorithm.
3. Step S2, which closes the aforementioned grain boundary, specifically, Step S2.1 involves processing the edge image using a morphological closing method, further thinning the edges, and reconnecting the broken edges; Step S2.2: Calculate the area of each hole, and based on the threshold, fill the holes whose area is smaller than the threshold using the hole-filling method; and A method for extracting ferrite structure edges based on a microstructure image of steel according to claim 1, characterized by including step S2.3 of thinning the grain boundaries using a skeletalization method.
4. The method for extracting ferrite structure edges based on a microstructure image of steel according to claim 1, characterized in that the scanning model when repeatedly scanning each grain boundary arc connected to the vertex in step S4 is as follows. G=(V,E) (1) "In the formula, V is the vertex of a crystal grain having an unclosed portion, and E is the grain boundary arc connected to that vertex."
5. Step S5, which involves inspecting the reconstruction and completing the extraction and re-reconstruction of the ferrite tissue edge, specifically, Step S5.1: Aggregating adjacent grain clusters of each grain to form a network topology; Step S5.2: Calculate the average grayscale of each crystal grain; Step S5.3: Detect whether or not each crystal grain contains a crystal grain boundary; Step S5.4 involves determining that adjacent crystal grains that satisfy both of the following conditions belong to the same crystal grain: 1) the average grayscale is below a threshold, and 2) the image boundary is not included in the crystal grain. Re-reconstruction of crystal grains: If it is found that the above inspection conditions are met but the reconstruction is not correctly labeled, step S5.5 is performed to change the grain boundaries of the adjacent crystal grains to be identical; and The method for extracting ferrite structure edges based on a microstructure image of steel according to claim 1, characterized by including step S5.6 of outputting and saving the image result after the processing.