基于语言模型多比特嵌入的不可感知分层水印嵌入方法
By employing a language model-based multi-bit embedding method and utilizing instruction logic trees and dynamic hash values to calculate dynamic sampling offset operators, the problem of logical integrity and traceability reliability in watermark embedding in industrial instruction sequences is solved, achieving efficient and reliable multi-bit watermark embedding and traceability in short-sequence text.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI PUDONG CRYPTOZOOLOGY RESEARCH INSTITUTE
- Filing Date
- 2026-01-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to embed highly reliable, imperceptible, and layered traceability-supporting multi-bit watermarks into industrial instruction sequences while maintaining logical integrity, leading to failures in traceability information extraction or causing logical collapse of industrial modeling scripts.
By employing a language model-based multi-bit embedding method, a dynamic sampling offset operator is calculated using the hierarchical dependency relationship of the instruction logic tree and dynamic hash values. This constructs a probabilistic mapping relationship between watermark bits and key instruction fields, performs dimension remapping correction and sampling space partitioning modulation, and ensures that the watermark information is embedded within a high-entropy semantic range.
It enables rapid accumulation of statistical significance in short-sequence text, ensuring traceability reliability, preventing watermark signals from interfering with industrial design logic, and improving the attack resistance and accuracy of industrial digital assets.
Smart Images
Figure CN121997303B_ABST