*2.4. Textural Segmentation of Partially Reacted IMDC*

Partially reacted IMDC (PR IMDC) areas segmented by porous IMDC identification should have porosity smaller in size than that of RMDC, but higher than unreacted IMDC. This is the main criterion used to identify them. The algorithms used for segmentation are the same as for bulk and porous IMDC identification, but the processing parameters are slightly adjusted to segment the correct areas. The result of bulk IMDC identification with increased dilation, erosion and scrap parameters applied for PR IMDC segmentation to the image from Figure 2 is given in Figure 7a. The result of porous IMDC identification with similarly increased parameters is provided in Figure 7b. Figure 7c shows the preliminary PR IMDC area identification obtained by combining the two identifications described above and subsequent removal of areas already identified as non-reacted IMDC. Along with the correct PR IMC identification, Figure 7c also includes some relatively thin and small RMDC areas attached to non-reacted IMDC. These areas can be removed by scrapping, leaving only the valid PR IMDC identified (see Figures 2 and 7d).

**Figure 7.** Segmentation of partially reacted IMDC: (**a**) bulk IMDC identification of PR IMDC; (**b**) porous IMDC identification of PR IMDC; (**c**) the combination of (**a**) and (**b**) with not reacted IMDC removed (blue); (**d**) PR IMDC areas (blue).

## **3. Textural Segmentation in OIA of Iron Ore Sinter**

Together with coke, iron ore sinter is the one of the major blast furnace loads. It can constitute up to 70–85% of the total ferrous burden fed to the blast furnace and its quality is also very important for stable blast furnace operation. The quality of sinter (e.g., its strength and reducibility) mainly depends on its petrology and texture, which in turn are determined by the initial ore blend, the fluxes added to the blend, and the sintering conditions. To optimize sinter quality and productivity the relationships between the initial sinter mix, sintering conditions and sinter structure, petrology and porosity must be understood [23–26].

One of the important characteristics of iron ore sinter is the quantity of primary and secondary hematite, which can provide insights into the presence of large grains of hematite in the initial iron ore blend, the degree of reaction and sintering conditions. The two types of hematite have very similar, or the same, mineral chemistry and reflectivity, so they cannot be segmented by standard scanning electron microscopy or optical image analysis methods. However, they have different morphology and phase associations which can be discriminated by OIA. Primary hematite generally does not contain inclusions of melt-precipitated phases and often preserves the morphology of the original ore particle. Secondary hematite, which mostly precipitates during cooling from the sinter melt, has crystals fully surrounded by, and/or including, other melt-precipitated sinter phases, such as Silico-Ferrites of Calcium and Aluminium (SFCA), undifferentiated glass, and larnite (di-calcium silicate). The presence

or absence of these spatially associated phases is the key feature used by Mineral4 for textural segmentation of the two types of hematite.

As a starting point for OIA processing, a map of all hematite in the sinter image is obtained by standard thresholding (Figure 1a,c and Figure 8a). A combined map of all phases which are typically associated with secondary hematite, such as SFCA, glass and larnite, is also prepared (Figure 8b). This second map is scrapped of its finest elements (to exclude the effect of imaging artefacts) and then strongly dilated to create a map of areas associated with the melt-precipitated phases (Figure 8c). Next, these areas are removed from the overall map of hematite. After additional scrap, removing undersized regions, only the areas of primary hematite are left in the map (Figure 8d). More dilation follows to compensate for loss during strong dilation of SFCA/glass areas (Figure 8e). Finally the map is masked with the original hematite map (Figure 8f) and fine/undersized objects are removed from it. The resulting identification of primary and, by exclusion, secondary hematite is shown in Figure 9.

**Figure 8.** Textural segmentation of primary hematite in iron ore sinter: (**a**) map of all hematite in the image; (**b**) map of SFCA and all phases darker than SFCA; (**c**) map of areas associated with glass and SFCA; (**d**) hematite map without areas associated with melt-precipitated phases; (**e**) map of possible areas for primary hematite; (**f**) identified areas of primary hematite including smaller grains to be removed later based on size.

**Figure 9.** Mineral map for the image in Figure 1a produced by Mineral4 software during automated image analysis: primary hematite—light blue, secondary hematite—dark blue, magnetite—magenta, platy SFCA-I—light green, prismatic/dense SFCA—olive, glass—dark green, larnite—cyan, porosity and epoxy within particles—yellow.

Figure 9 demonstrates the false color map of all sinter phases identified by Mineral4 software during automated image analysis of the crushed sinter shown in Figure 1a. Note that the large grain of primary hematite in the bottom right corner includes some remnant kenomagnetite. Magnetite may still be present in stable ore nuclei remaining after sintering, not just as one of melt-precipitated sinter phases. Therefore, it is not included as part of the melt-precipitated phases in the map (Figure 8b). It is still clear that the majority of magnetite present in Figure 9 is melt-precipitated.

The other pertinent textural segmentation by OIA shown in Figure 9 is that of two types of SFCA: microporous platy SFCA-I (light green) and prismatic/dense SFCA (olive). These two phases have the same reflectivity but different morphology. SFCA-I has slightly higher iron contents which may be determined by SEM methods [27,28]. However, in OIA sinter characterization, textural identification is required for segmentation of different SFCA types. As SFCA-I often has fine porosity evident between adjacent plates, this can be utilized for the textural segmentation. The actual algorithm is very similar to that utilized for porous IMDC identification described above.

#### **4. Textural Segmentation in OIA of Iron Ore**

Many authors have demonstrated the importance of iron ore textural characterisation [29–36] for the optimization of downstream processing performance. Donskoi et al. [23,24] showed that the presence of textural information for parent iron ore blend allows a significant improvement in modelling of iron ore sinter quality. It also provides better modelling and deeper understanding of beneficiation processes [34,35,37].

Quantitative mineral characterization, sometimes including identification of different morphologies of the same mineral, is required to correctly texturally classify iron ores. Figure 10a shows an image of iron ore consisting of two hematite types: microplaty hematite (thin, long plates) and martite. To better understand the reactivity of such an ore, its behavior during pelletising, granulation and sintering, it is important to know the abundances of both types of hematite. Textural segmentation in this particular example can be fairly simple, e.g., initial erosion removing the fine structure of microplaty hematite, followed by compensating dilation restoring the martite grains (similar to the steps in bulk IMDC identification shown in Figure 3c,d.

**Figure 10.** Identification of martite and microplaty hematite by Mineral4: (**a**) original reflected light photomicrograph; (**b**) resulting mineral map (martite—magenta, microplaty hematite—blue; porosity—yellow).

Textural identification may also be of significant help during mineral segmentation in complex cases. Figure 11a shows part of a particle that mainly consists of siliceous goethite with some inclusions of hematite (the brighter grains) and porosity. The area in the top-left corner of the image is epoxy. Segmentation of this siliceous goethite with usual thresholding is problematic because it is rather dark, such that the reflectivity of the epoxy is within the same range as the reflectivity of the goethite. Figure 12a shows an attempted segmentation of the epoxy, which corresponds to the tall narrow peak on the reflectivity histogram (Figure 12b). During this segmentation, significant areas inside the goethite particle were also selected. The reason is that the part of the histogram corresponding to goethite is the relatively wider but lower elevation on which the epoxy peak is based. Obviously, if goethite thresholding is attempted, the whole epoxy area will be selected as well (Figure 12c). To properly segment goethite from epoxy, Mineral4 used multi-thresholding [20] with textural identification. Initially, the area of goethite with reflectivity less than that of epoxy is thresholded (Figure 12d,e). The resulting map is subjected to dilation and erosion (Figure 12f) solidifying the map (this combined image analysis operation is known as Closing), but still some goethite areas remain unselected. Next, the area of goethite with reflectivity higher than epoxy is thresholded (Figure 12g,h) and the same dilation and erosion combination is applied (Figure 12i). The two maps are then combined. After previously identified maps of hematite (corresponding to the small elevation in the right part of the reflectivity histogram), vitreous goethite and porosity are removed, the remaining map gives the final siliceous goethite identification (Figure 11b) which would not be possible to obtain without textural identification.

**Figure 11.** (**a**) Image of siliceous goethite particle with hematite inclusions; (**b**) mineral map obtained in Mineral4: siliceous goethite—olive, hematite—blue, vitreous goethite (very fine)—green, porosity—yellow.

**Figure 12.** Segmentation of siliceous goethite in Figure 11a: (**a**) thresholding of epoxy; (**b**) reflectivity histogram with thresholds corresponding to (**a**); (**c**) thresholding of goethite; (**d**) thresholding of goethite area with reflectivity less than epoxy; (**e**) reflectivity histogram with thresholds corresponding to (**d**); (**f**) the result of thresholding shown in (**d**) after some dilation and erosion; (**g**) thresholding of goethite area with reflectivity higher than epoxy; (**h**) reflectivity histogram with thresholds corresponding to (**g**); (**i**) the result of thresholding shown in (**g**) after dilation and erosion.
