Image Privacy Item Recognition Based on Hybrid Model of Hierarchical Feature Recognition and ViT

With the development of artificial intelligence technology, privacy risk detection has become particularly important in scenarios such as intelligent monitoring and identity authentication. However, existing technologies have shortcomings in complex scenarios and global feature processing, resulting in low detection accuracy in some cases. This paper proposes a hybrid model that combines CNN, OCNN and Transformer models to extract features and achieves higher detection accuracy. This method innovatively combines the advantages of different feature extraction methods and improves the ability to identify privacy risks. Experimental results show that the proposed method outperforms existing technologies on multiple test sets, not only improving detection accuracy but also reducing false alarm rates.