Hierarchical aggregation perceptual pipeline for tactical intention recognition

Ying Li, Junsheng Wu, Weigang Li, Wei Dong, Aiqing Fang

Abstract: Tactical intention recognition involves the analysis of target information to interpret and accurately determine hostile intentions, which is crucial for auxiliary decision-making. The recognition challenges are posed by hierarchical and long-term dependencies in tactical intention. To this end, we propose a pipeline that enhances intention recognition performance by perceiving and aggregating the hierarchy information within the tactical context, termed Hierarchical Aggregation Perceptual Pipeline (HAGP). Specifically, the HAGP comprises two pipelines: maneuver features perceive (MFP), and intention features aggregate (IFA). The MFP captures the maneuver features, which are sub-intentioned with hierarchical information, and the IFA aggregates long-term dependencies in each intention. Then, combining these representations to facilitate precise tactical intention recognition. Extensive experimental results on the tactical dataset demonstrate the superiority of our pipeline compared with the state-of-the-art methods.