Improving the resilience of insulator hydrophobicity classification systems against cyber attacks

Akash, R, Rebeiro, Chester, Haddad, Manu A. ORCID: https://orcid.org/0000-0003-4153-6146 and Sarathi, R 2025. Improving the resilience of insulator hydrophobicity classification systems against cyber attacks. Presented at: 2025 IEEE 4th International Conference on Smart Technologies for Power, Energy and Control (STPEC), Goa, India, 10-13 December 2025. 2025 IEEE 4th International Conference on Smart Technologies for Power, Energy and Control (STPEC). IEEE, 10.1109/stpec66316.2025.11490885

Full text not available from this repository.

Abstract

Condition monitoring of polymeric insulators relies on deep learning and computer vision methods, which are vulnerable to cyberattacks. This paper introduces a white-box threat model employing Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks to assess the robustness of a standard Swin Transformer and a novel Hybrid Swin Transformer (HST). This work compared four adversarial training strategies: no defence, FGSM-only, PGD-only, and a combined FGSM + PGD approach, evaluating clean and adversarial accuracy across different perturbation budgets. Adversarial imperceptibility is measured using Multiscale Structural Similarity (MS-SSIM) and Peak Signal-to-Noise Ratio (PSNR). Results show that the HST outperforms the standard Swin Transformer by 3 to 4 percentage points in robust accuracy. The mixed training method enhances robust accuracy by 3.15 percentage points at

Item Type:	Conference or Workshop Item - published (Paper)
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering ?? AHVER ??
Publisher:	IEEE
ISBN:	979-8-3315-9807-5
Last Modified:	11 May 2026 11:16
URI:	https://orca.cardiff.ac.uk/id/eprint/186921

Actions (repository staff only)

Edit Item