Compartir
Título
Classifying Screenshots of Industrial Control System Using Transfer Learning and Fine-Tuning
Autor
Facultad/Centro
Área de conocimiento
Es parte de
Investigación en Ciberseguridad. Actas de las VI Jornadas Nacionales (JNIC2021 LIVE)
Cita Bibliográfica
Vasco-Carofilis, R. A., Blanco-Medina, P., Jáñez-Martino, F., Bennabhaktula, G. S., Fidalgo, E., Prieto Castro, A., & Fidalgo, V. (2021). Classifying Screenshots of Industrial Control System Using Transfer Learning and Fine-Tuning. En Investigación en Ciberseguridad (pp. 151-152). https://doi.org/10.18239/JORNADAS_2021.34.35
Editorial
Ediciones de la Universidad de Castilla-La Mancha
Fecha
2021-06
Resumen
[EN] Industrial control systems are heavily dependant on security and monitoring protocols. For this purpose, monitoring tools take screenshots of control panels for later analysis.Classifying these screenshots into specific groups can be a time-consuming process, but it is crucial for the security tasks performed by manual operators. To solve this problem, we propose a pipeline based on deep learning to classify snapshots of industrial control panels into three categories: Internet Technologies (IT), Operation Technologies (OT), and others. We compare the results obtained with transfer learning and finetuning on nine convolutional neuronal networks pre-trained with the ImageNet dataset, testing them on a custom CRitical INFrastructure dataset (CRINF-300). Inception-ResNet-V2 obtains the best learning result with an F1-score of 98.32% on CRINF-300, while MobileNet-V1 obtained the best performance-speed tradeoff
Materia
Palabras clave
ID proyecto
- This work was supported by the framework agreement between the Universidad de Leon and INCIBE (Spanish ´ National Cybersecurity Institute) under Addendum 01. We acknowledge NVIDIA Corporation with the donation of the TITAN Xp and Tesla K40 GPUs used for this research.
URI
Versión del editor
Aparece en las colecciones
- Capítulos de monografías [1431]
Ficheros en el ítem
Tamaño:
265.4
xmlui.dri2xhtml.METS-1.0.size-kilobytes
Formato:
Adobe PDF
Descripción:
Aportación congreso publicada