View in EDS

DINOv2 rocks geological image analysis: Classification, segmentation, and interpretability

Saved in:
Bibliographic Details
Title: DINOv2 rocks geological image analysis: Classification, segmentation, and interpretability
Authors: Florent Brondolo, Samuel Beaussant
Source: Journal of Rock Mechanics and Geotechnical Engineering, Vol 17, Iss 11, Pp 6853-6867 (2025)
Publisher Information: Elsevier, 2025.
Publication Year: 2025
Collection: LCC:Engineering geology. Rock mechanics. Soil mechanics. Underground construction
Subject Terms: Computer vision, Micro-computed tomography (μCT), DINOv2, Vision transformers (ViTs), Segmentation, Classification, Engineering geology. Rock mechanics. Soil mechanics. Underground construction, TA703-712
Description: Recent advancements in computer vision have significantly improved image analysis tasks. However, deep learning models often struggle when applied to domains outside their training distribution, such as geosciences, where domain-specific data can be scarce. This study examines the classification, segmentation, and interpretability of computed tomography (CT) scan images of rock samples, with a focus on the application of modern computer vision techniques to geoscientific tasks. We compare various segmentation methods to assess their efficacy, efficiency, and adaptability in geological image analysis. The methods evaluated include Otsu thresholding, clustering techniques (k-means and fuzzy C-means (FCM)), a supervised machine learning approach (random forest), and deep learning models (UNet, ResNet152, and DINOv2), using ten binary sandstone datasets and three multi-class carbonate datasets. DINOv2 was selected for its promising results in feature extraction and its potential applicability in geoscientific tasks, prompting further assessment of its interpretability and effectiveness in processing CT-scanned rock data. For classification, a non-fine-tuned DINOv2 demonstrates strong performance in classifying rock images, even when the CT scans are likely outside its training set. In segmentation tasks, thresholding and clustering techniques, though computationally efficient, produce subpar results despite pre-processing efforts. In contrast, supervised methods achieve better performance without pre-processing. Although deep learning methods require greater computational resources, they demand minimal intervention and offer superior generalization. A LoRA fine-tuned DINOv2 excels in out-of-distribution segmentation and outperforms other methods in multi-class tasks, even with limited data. Notably, the segmentation masks generated by DINOv2 often appear more accurate than the original targets, based on visual inspection.
Document Type: article
File Description: electronic resource
Language: English
ISSN: 1674-7755
Relation: http://www.sciencedirect.com/science/article/pii/S1674775525002677; https://doaj.org/toc/1674-7755
DOI: 10.1016/j.jrmge.2025.01.057
Access URL: https://doaj.org/article/5c07fbb16e9c45f299524897b9166d4f
Accession Number: edsdoj.5c07fbb16e9c45f299524897b9166d4f
Database: Directory of Open Access Journals
Description
Abstract:Recent advancements in computer vision have significantly improved image analysis tasks. However, deep learning models often struggle when applied to domains outside their training distribution, such as geosciences, where domain-specific data can be scarce. This study examines the classification, segmentation, and interpretability of computed tomography (CT) scan images of rock samples, with a focus on the application of modern computer vision techniques to geoscientific tasks. We compare various segmentation methods to assess their efficacy, efficiency, and adaptability in geological image analysis. The methods evaluated include Otsu thresholding, clustering techniques (k-means and fuzzy C-means (FCM)), a supervised machine learning approach (random forest), and deep learning models (UNet, ResNet152, and DINOv2), using ten binary sandstone datasets and three multi-class carbonate datasets. DINOv2 was selected for its promising results in feature extraction and its potential applicability in geoscientific tasks, prompting further assessment of its interpretability and effectiveness in processing CT-scanned rock data. For classification, a non-fine-tuned DINOv2 demonstrates strong performance in classifying rock images, even when the CT scans are likely outside its training set. In segmentation tasks, thresholding and clustering techniques, though computationally efficient, produce subpar results despite pre-processing efforts. In contrast, supervised methods achieve better performance without pre-processing. Although deep learning methods require greater computational resources, they demand minimal intervention and offer superior generalization. A LoRA fine-tuned DINOv2 excels in out-of-distribution segmentation and outperforms other methods in multi-class tasks, even with limited data. Notably, the segmentation masks generated by DINOv2 often appear more accurate than the original targets, based on visual inspection.
ISSN:16747755
DOI:10.1016/j.jrmge.2025.01.057