基于内嵌伪造机理的多模态协同鲁棒伪造信息检测方法

Multi-mode Collaborative Robust Forgery Detection Method Basedon Embedded Forgery Mechanism

针对深度伪造信息强对抗干扰导致的鲁棒性偏差问题,提出一种基于内嵌伪造机理的多模态协同高鲁棒性伪造信息检测方法,通过自监督预训练模型提取结合上下文信息、语义风格、深度特征的鲁棒多模态特征信息,深入研究伪造机理,进行抗干扰声学建模,实现复杂场景下的多模态协同鲁棒伪造信息检测。

Aiming at the problem of robustness bias caused by strong counter-interference of deep forgery information,the study proposes a multi-modal collaborative high-robustness forgery information detection method based on embedded forgery mechanism;extract robust multi-modal feature information combining context information,semantic style and depth features with self-supervised pre-training model;conducts in-depth study of the forgery mechanism for anti-interference acoustic modeling.The multi-mode cooperative robust forged information detection in complex scenarios is realized.