Recent advancements in computer vision have significantlyimproved image analysis tasks.However,deep learning models often struggle when applied to domains outside their training distribution,such as geosciences,where ...Recent advancements in computer vision have significantlyimproved image analysis tasks.However,deep learning models often struggle when applied to domains outside their training distribution,such as geosciences,where domain-specificdata 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 geoscientifictasks.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 geoscientifictasks,prompting further assessment of its interpretability and effectiveness in processing CT-scanned rock data.For classification,a non-fine-tunedDINOv2 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-tunedDINOv2 excels in out-ofdistribution 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.展开更多
Critical Infrastructures(CIs),which serve as the foundation of our modern society,are facing increasing risks from cyber threats,physical attacks,and natural disasters.Additionally,the interdependencies between CIs th...Critical Infrastructures(CIs),which serve as the foundation of our modern society,are facing increasing risks from cyber threats,physical attacks,and natural disasters.Additionally,the interdependencies between CIs through-out their operational lifespan can also significantly impact their integrity and safety.As a result,enhancing the resilience of CIs has emerged as a top priority for many countries,including the European Union.This involves not only understanding the threats/attacks themselves but also gaining knowledge about the areas and infrastruc-tures that could potentially be affected.A European Union-funded project named PRECINCT(Preparedness and Resilience Enforcement for Critical INfrastructure Cascading Cyber-Physical Threats),under the Horizon 2020 program,tries to connect private and public stakeholders of CIs in a specific geographical area.The key objec-tive of this project is to establish a common cyber-physical security management approach that will ensure the protection of both citizens and infrastructures,creating a secure territory.This paper presents the components of PRECINCT,including a directory of PRECINCT Critical Infrastructure Protection(CIP)blueprints.These blueprints support CI communities in designing integrated ecosystems,operating and replicating PRECINCT components(or toolkits).The integration enables coordinated security and resilience management,incorporating improved’installation-specific’security solutions.Additionally,Serious Games(SG),and Digital Twins(DT)are a significant part of this project,serving as a novel vulnerability evaluation method for analysing complicated multi-system cascading effects in the PRECINCT Living Labs(LLs).The use of SG supports the concentrated advancement of innovative resilience enhancement services.展开更多
文摘Recent advancements in computer vision have significantlyimproved image analysis tasks.However,deep learning models often struggle when applied to domains outside their training distribution,such as geosciences,where domain-specificdata 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 geoscientifictasks.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 geoscientifictasks,prompting further assessment of its interpretability and effectiveness in processing CT-scanned rock data.For classification,a non-fine-tunedDINOv2 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-tunedDINOv2 excels in out-ofdistribution 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.
基金funded by the European Commission,Horizon 2020 research and innovation programme under grant agreement No.101021668.
文摘Critical Infrastructures(CIs),which serve as the foundation of our modern society,are facing increasing risks from cyber threats,physical attacks,and natural disasters.Additionally,the interdependencies between CIs through-out their operational lifespan can also significantly impact their integrity and safety.As a result,enhancing the resilience of CIs has emerged as a top priority for many countries,including the European Union.This involves not only understanding the threats/attacks themselves but also gaining knowledge about the areas and infrastruc-tures that could potentially be affected.A European Union-funded project named PRECINCT(Preparedness and Resilience Enforcement for Critical INfrastructure Cascading Cyber-Physical Threats),under the Horizon 2020 program,tries to connect private and public stakeholders of CIs in a specific geographical area.The key objec-tive of this project is to establish a common cyber-physical security management approach that will ensure the protection of both citizens and infrastructures,creating a secure territory.This paper presents the components of PRECINCT,including a directory of PRECINCT Critical Infrastructure Protection(CIP)blueprints.These blueprints support CI communities in designing integrated ecosystems,operating and replicating PRECINCT components(or toolkits).The integration enables coordinated security and resilience management,incorporating improved’installation-specific’security solutions.Additionally,Serious Games(SG),and Digital Twins(DT)are a significant part of this project,serving as a novel vulnerability evaluation method for analysing complicated multi-system cascading effects in the PRECINCT Living Labs(LLs).The use of SG supports the concentrated advancement of innovative resilience enhancement services.
