.png){kind=icon}
.png){kind=icon}
Return
PSR-Net: process-state consistency verification for industrial image tampering detection and localization
.png){kind=icon}
PRE
AI
J
B
W
DOI：10.1007/s00371-026-04642-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Industrial visual monitoring systems increasingly support process awareness, remote inspection, and safety–critical alarming. However, visually plausible tampering may evade conventional image-forensics methods when local texture, noise, and boundary artifacts are weak, while still contradicting synchronized equipment states or sensor readings. We propose PSR-Net, a process-state consistency verification framework for industrial image tampering detection and localization. PSR-Net encodes image-region features and synchronized process-state vectors into a shared representation space, uses a process-state re-indexing module to estimate region-level process relevance, and reconstructs the visual representation expected under the current operating condition. The residual between actual and reconstructed region features is then used as evidence for image-level attack detection and region-level tampering localization. We construct ICS-Tamper, a process-synchronized industrial image tampering dataset containing 13,500 image-state sample groups, including 8,100 normal samples and 5,400 tampered samples from six tampering types. On the ICS-Tamper test set, PSR-Net achieves 94.08% Precision, 92.74% Recall, and 93.41% F1-score, improving over the strongest baseline by 3.77, 4.12, and 3.95 percentage points, respectively. Ablation, process-inconsistent subset analysis, complexity comparison, and visualization results support the roles of process-state re-indexing and residual reconstruction. Our code is publicly available at https://github.com/cdt-zhangwei .
Keywords:
Industrial visual security
Image forensics
Tampering localization
Process-state consistency
Multimodal anomaly detection
Industrial control systems
Trustworthy visual perception
Journal
T
IF:
0
Papers:
369
Citations:
0