Dynamic gradient fusion method with local spatial and multi-scale frequency transformations for transferable adversarial attacks

doi:10.1016/j.eswa.2026.131677

Dynamic gradient fusion method with local spatial and multi-scale frequency transformations for transferable adversarial attacks

2026-02-16

0

PRE

AI

DOI：10.1016/j.eswa.2026.131677

Original

Original request for help

Save

Abstract

En 中文

• Applies saliency-aware local spatial transformations to enhance input diversity. • Introduces multi-scale frequency-domain transformation to expand spectrum saliency coverage. • Designs an adaptive gradient fusion strategy based on loss differences to boost transferability.

Keywords:

saliency-aware local spatial transformations

multi-scale frequency-domain transformation

adaptive gradient fusion strategy

transferable adversarial attacks

Dynamic gradient fusion method with local spatial and multi-scale frequency transformations for transferable adversarial attacks

Abstract

Journal

Organization