This technology develops an encrypted traffic intelligent generation system based on an LSTM-DCGAN fusion model. By converting time-series features into pseudo-image matrices and embedding LSTM layers in the DCGAN generator, a spatiotemporal joint feature architecture is formed, combined with AMSGrad optimizer to balance adversarial training, it achieves unsupervised learning and label retention, generating high-fidelity traffic data (97% feature fidelity) containing TLS1.3/QUIC and 23 other protocols, reducing labeling costs by 85%.

Technology provider:Nanjing University of Posts and Telecommunications