Multi-modal network traffic classification method and system based on multi-granularity traffic semantic representation

Abstract

The application relates to a multi-modal network traffic classification method and system based on multi-granularity flow semantic representation, which comprises the following steps: dividing original traffic into session granularity according to bidirectional flow, extracting flow-level statistical features and constructing one-dimensional byte sequence; constructing multi-level topological association representation by using inter-flow association modeling branch, mapping flow-level enhanced representation into nodes through an inter-flow association graph, and explicitly modeling the cooperative mode between flows based on space-time consistency; constructing two-layer hierarchical Mamba architecture of cooperative packet-level semantic analysis and flow-level time sequence dynamics, explicitly injecting the time interval between packets into the state evolution process, and mining the coupling features of load content and time sequence dynamics; using an adaptive gating fusion module to dynamically adjust the contribution proportion of intra-flow sequence features and inter-flow association topological features, and outputting a classification result. The application realizes the cooperative representation of intra-flow microscopic semantics and inter-flow macroscopic association, can adaptively fuse multi-modal features, and significantly improves the accuracy and robustness of network traffic classification.

Images