融合检测、消除与对抗训练的软测量对抗防御方法及系统
By integrating detection and adversarial training, adversarial examples are processed hierarchically and a SPAS-FAAT model is constructed. This solves the problem of poor defense of the DLSS model when facing highly covert adversarial attacks, and achieves high-precision adversarial robustness and detection accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAN UNIV OF TECH
- Filing Date
- 2026-05-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing DLSS models have poor defense against strong, covert adversarial attacks. Active defense suffers from overfitting due to adversarial robustness, while reactive defense struggles to identify strong, covert adversarial attacks. This results in decreased prediction accuracy when the model is subjected to adversarial attacks, failing to meet the high-precision measurement requirements of industry.
A method combining detection and adversarial training is adopted. The detector classifies the test samples, eliminates the perturbation of adversarial samples with large perturbations, and trains the feature anchoring of adversarial samples with small perturbations to construct a SPAS-FAAT model. The sample classification is performed by combining forward prediction and backward reconstruction residual calculation matrix norm to form a deep learning model with graded response.
It effectively identifies and eliminates highly covert adversarial attacks, ensuring that the model maintains high-precision prediction when facing adversarial attacks, balancing defense efficiency and measurement speed, and improving the adversarial robustness and detection accuracy of the DLSS model.
Smart Images
Figure CN122137687B_ABST