Trajectory analysis model and training method, trajectory analysis method and device, and medium
By combining track coding, physical constraints, and multimodal adaptation networks with a large language model, this approach addresses the issues of track analysis features deviating from physical reality and insufficient semantic interpretation in existing technologies, achieving efficient, reliable, and deep understanding of track analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI IFLY DIGITAL TECH CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-05
AI Technical Summary
Existing trajectory analysis techniques rely on data fitting, which causes latent features to deviate from the objective physical motion boundary. They cannot be combined with external business instructions for in-depth situational reasoning and semantic interpretation, and lack cross-modal adaptation mechanisms, resulting in low reliability of feature representation and unreliable analysis results.
Latent features are extracted using a track coding network, combined with a physical consistency constraint network and a multimodal adaptation network, and semantic reasoning is performed through a large language model to ensure that features are constrained within the physical feasible domain and to achieve deep alignment between tracks and language.
It enhances the physical rationality and reliability of flight path analysis, stimulates a deeper understanding of complex low-altitude flight behavior and higher-order logical reasoning ability, and generates intuitive comprehensive situational semantic analysis results.
Smart Images

Figure CN121808399B_ABST
Abstract
Citation Information
Patent Citations
Cross-modal large model construction method and system based on track spatio-temporal characteristics
CN121392489A
Low-altitude aircraft intelligent identification system based on edge network
CN121436148A