5:15 PM - 7:15 PM
[SVC31-P02] Modelling the Internal Structure of Pumice Using the FRST and Watershed Methods
Keywords:Pumice, Bubble Network, Image Processing
Pumice is a pyroclastic material characterized by high porosity and low density (0.2-0.8 g/cm³) in dry conditions. When submerged in fluid, it absorbs water, causing its density to increase. Previous studies on water infiltration properties have demonstrated that the absorption rate decreases over time, with 22-64% of pore space becoming water-filled within five minutes. These absorption properties vary based on pumice size and fluid temperature; larger samples and colder water result in slower absorption rates, allowing the pumice to remain buoyant longer. Understanding these water absorption properties is crucial for predicting depositional mechanisms and transportability in pumice-laden flows.
Since the internal structure of pumice is closely linked to its water absorption properties, developing a quick and efficient method to analyze its internal structure is essential. Several methods exist for studying internal structures: CT scanning technology provides detailed 3D data on sample microstructure, including bubble networks and bubble walls, while Scanning Electron Microscopy (SEM) enables 2D analysis of surface textures and microstructures. Although digital rock physics has proven effective for analyzing overall 3D structure with high accuracy and resolution, the required equipment, such as synchrotron facilities and SEM instruments, is large-scale and impractical for on-site use. Additionally, these methods are time-consuming and costly. Therefore, the authors propose a simplified technique for immediate on-site evaluation of bubble structures (e.g., permeation pathways) using cross-section photographs, focusing on simulating water absorption. This approach aims to rapidly assess the connectivity and cross-sectional structures of volcanic ejecta, enabling efficient data collection on pumice's water absorption properties.
The internal structure of pumice was segmented into individual bubbles by combining the Fast Radial Symmetry Transform (FRST) and Watershed algorithm. FRST is an effective image processing technique for identifying circular or radial structures in images and is notable for its computational efficiency with large image datasets. While this technique is widely used in medical image processing for detecting blood vessels and cells, its application in volcanological fields remains limited. This study contributes to the field by verifying FRST's effectiveness in detecting bubbles in volcanic ejecta, including pumice.
To evaluate segmentation results, the method was iteratively repeated (n=1,2,...10). Additional tests assessed whether segmentation could be achieved using either FRST or Watershed alone.
The results showed that while large bubbles were effectively detected using the proposed method, small bubbles required 2-3 iterations of the combined FRST and Watershed approach for detection. However, bubbles became over-segmented and undetectable beyond 9-10 iterations. Detection accuracy was lower for bubbles near image boundaries and for very small bubbles.
Since FRST tends to segment bubbles into circular or elliptical shapes, excessive iterations led to over-segmentation, reducing detection accuracy. Extremely small bubbles were undetectable due to resolution limitations. However, prior studies suggest that bubbles with small radii tend to form spherical shapes when considering the capillary number (Ca) during bubble generation. This indicates that training AI models to detect small spherical bubbles with added noise could enhance detection accuracy. Furthermore, infiltration simulations were conducted using a diffusion equation based on randomly generated boundaries and average density data. Over time, variability in average density results increased, even for identical samples. This variability reflects uncertainties in internal structure, demonstrating that: interconnected pores sometimes allow deeper water infiltration, while disconnected pores can halt infiltration near the surface. This inconsistent infiltration behavior results in slower deposition and facilitates long-distance transport.
Since the internal structure of pumice is closely linked to its water absorption properties, developing a quick and efficient method to analyze its internal structure is essential. Several methods exist for studying internal structures: CT scanning technology provides detailed 3D data on sample microstructure, including bubble networks and bubble walls, while Scanning Electron Microscopy (SEM) enables 2D analysis of surface textures and microstructures. Although digital rock physics has proven effective for analyzing overall 3D structure with high accuracy and resolution, the required equipment, such as synchrotron facilities and SEM instruments, is large-scale and impractical for on-site use. Additionally, these methods are time-consuming and costly. Therefore, the authors propose a simplified technique for immediate on-site evaluation of bubble structures (e.g., permeation pathways) using cross-section photographs, focusing on simulating water absorption. This approach aims to rapidly assess the connectivity and cross-sectional structures of volcanic ejecta, enabling efficient data collection on pumice's water absorption properties.
The internal structure of pumice was segmented into individual bubbles by combining the Fast Radial Symmetry Transform (FRST) and Watershed algorithm. FRST is an effective image processing technique for identifying circular or radial structures in images and is notable for its computational efficiency with large image datasets. While this technique is widely used in medical image processing for detecting blood vessels and cells, its application in volcanological fields remains limited. This study contributes to the field by verifying FRST's effectiveness in detecting bubbles in volcanic ejecta, including pumice.
To evaluate segmentation results, the method was iteratively repeated (n=1,2,...10). Additional tests assessed whether segmentation could be achieved using either FRST or Watershed alone.
The results showed that while large bubbles were effectively detected using the proposed method, small bubbles required 2-3 iterations of the combined FRST and Watershed approach for detection. However, bubbles became over-segmented and undetectable beyond 9-10 iterations. Detection accuracy was lower for bubbles near image boundaries and for very small bubbles.
Since FRST tends to segment bubbles into circular or elliptical shapes, excessive iterations led to over-segmentation, reducing detection accuracy. Extremely small bubbles were undetectable due to resolution limitations. However, prior studies suggest that bubbles with small radii tend to form spherical shapes when considering the capillary number (Ca) during bubble generation. This indicates that training AI models to detect small spherical bubbles with added noise could enhance detection accuracy. Furthermore, infiltration simulations were conducted using a diffusion equation based on randomly generated boundaries and average density data. Over time, variability in average density results increased, even for identical samples. This variability reflects uncertainties in internal structure, demonstrating that: interconnected pores sometimes allow deeper water infiltration, while disconnected pores can halt infiltration near the surface. This inconsistent infiltration behavior results in slower deposition and facilitates long-distance transport.