Fast UAV Classification Based on Radar Trajectory Sequences

Accurate classification of unmanned aerial vehicles and birds based on radar trajectory sequences is essential for airspace security. To enable timely recognition under short-sequence conditions, we propose a Temporal Decay-Evidence Fusion (TDEF) model. The method employs multi-layer Long Short-Term Memory (LSTM) networks to extract temporal features and exploits Dirichlet distributions to model frame-level uncertainty. A learnable temporal decay module adaptively adjusts the influence of past frames, and Dempster-Shafer theory is applied for progressive evidence fusion. Experiments on real radar data show that TDEF achieves superior accuracy with lower false alarm and miss rates, demonstrating its effectiveness for fast and reliable target recognition.