Chengchao Zhang

Chengchao Zhang

Auto-generated profile, based on publicly available information

Chengchao Zhang

College of Intelligence and Computing, Tianjin University, Tianjin, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 December 2025 in IEEE Signal Processing Letters
Verified
Journal Article
Closed access
RWRuolong Wang
Ruolong Wang
JXJingyuan Xu
Jingyuan Xu
LZLONGGE ZHANG
LONGGE ZHANG
CZChengchao Zhang
Chengchao Zhang
WLWei Liu
Wei Liu
HLHongying Liu
Hongying Liu
Due to the exponential disparity in the magnitude of high- and low-frequency components in the image frequency domain, existing frequency-domain enhancement methods for adversarial examples often fail to balance the perturbations of high- and low-frequency information. To address this issue, we propose a novel Spectral Space Transformation and Loss adjustment (SSTL) method. Specifically, SSTL introduces a frequency-domain rotation-and-self-mixing strategy that adaptively balances perturbations across high- and low-frequency components, thereby generating more transferable adversarial examples. Furthermore, to mitigate semantic degradation during reconstruction, we design a logits fine-tuning mechanism that up-weights the logits of true labels while suppressing others, effectively preserving the original semantic information and avoiding misleading gradient propagation. Various experimental results demonstrate that SSTL effectively improves adversarial transferability while significantly preserving the semantic information of each image.