|
| Progress in automatic analysis technology for sinter microstructure images |
|
Received:October 28, 2024
Revised:December 20, 2024
Accepted:February 06, 2025
Published Online:April 30, 2026
|
| View Full Text View/Add Comment Download reader |
| DOI：10.3969/j.issn.1005-7854.2026.02.009 |
| KeyWord:sinter;automatic microstructure image analysis;mineral phase identification;mineral phase structure;artificial intelligence |
|
| 1.BGRIMM Technology Group, Beijing 100160, China;2.State Key Laboratory of Science and Technology of Mineral Processing, Beijing 102628, China |
| Hits: 32 |
| Download times: 5 |
| Abstract: |
| The mineralogical composition and microstructure of sinter directly influence its key properties such as strength, reducibility, and permeability, which in turn determine the stability and efficiency of blast furnace operation. Due to the complex microstructural characteristics of sinter formed during high-temperature reactions and multi-phase evolution—including the coexistence of multiple mineral phases, blurred boundaries, continuously varying colors, and multi-scale structural superposition—traditional manual identification methods relying on scanning electron microscopy (SEM) and polarizing microscopy are significantly limited in terms of quantification, efficiency, and repeatability. To meet the industrial demand for high efficiency and repeatable quantification in quality control, automated micro-image analysis technology based on computer image processing has gradually become an important research direction. This paper systematically reviews the research progress in automatic analysis technology for sinter microstructure images. It summarizes the technical principles, application scenarios, and typical recognition outcomes of traditional methods for mineral phase identification and microstructure characterization, including color threshold segmentation, fuzzy C-means clustering, region growing, watershed transformation, statistical hypothesis testing, multifractal analysis, and morphological processing. Studies indicate that threshold segmentation can effectively handle tasks with distinct color differences, such as separating magnetite and hematite, but its accuracy significantly decreases when mineral phase colors shift due to solid solution composition effects. Region growing and fuzzy clustering offer certain advantages in handling weak boundaries and transitional regions, but they are sensitive to parameter selection and seed point settings. Statistical discriminant methods can identify characteristic structures like reticular or needle-plate morphologies, but they generally rely on manually defined rules, making them difficult to adapt to diverse samples. Fractal theory can be used for the quantitative characterization of pore structure complexity, but it involves high computational costs and is sensitive to preprocessing quality. Overall, while traditional methods perform stably under single or regular feature conditions, their generalization capability and robustness remain limited when faced with the complex microstructure images of sinter, characterized by multi-scale features, diverse morphologies, and high noise levels. Building on this, the paper further discusses the application progress and development directions of sample learning-based image analysis methods in this field. Related research shows that such methods can form more adaptive discriminant features from sample data, showing potential in handling color fluctuations, weak boundaries, complex textures, and morphological variations within the same mineral phase. They offer new technical pathways for the automatic identification of typical mineral structures such as fibrous/plate-like calcium ferrites and skeletal hematite. Future research may focus on constructing multi-source image datasets, introducing mineralogical prior constraints, establishing models linking microstructure to macroscopic properties, and developing on-site online analysis, thereby supporting the digitization and intelligentization of the sintering process. |
| Close |
|
|
|