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.

CN121808399BActive Publication Date: 2026-06-05HEFEI IFLY DIGITAL TECH CO LTD
View PDF 2 Cites 0 Cited by

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

Technical Problem

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.

Method used

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.

Benefits of technology

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.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121808399B_ABST
    Figure CN121808399B_ABST
Patent Text Reader

Abstract

The application provides a flight path analysis model and a training method, a flight path analysis method and equipment and a medium, and belongs to the technical field of artificial intelligence. The model comprises: a flight path coding network, which is used for receiving a flight path sequence of an aircraft and extracting first flight path hidden features; a physical consistency constraint network, which is used for predicting physical state features and performing constraint based on a motion feasible region, and then fusing to generate second flight path hidden features; a multi-modal adaptation network, which is used for cross-modal fusion of the second flight path hidden features and analysis instruction features to obtain multi-modal input features; and a large language model network, which receives the input features to perform semantic reasoning and generates a flight path analysis result text. The application guarantees that the features conform to the physical motion law through the physical constraint network, and realizes deep alignment of the flight path and language through the cross-modal network, thereby breaking through the single-modal analysis bottleneck and significantly improving the high-order analysis capability of flight path situation understanding and reasoning in a complex low-altitude scene.
Need to check novelty before this filing date? Find Prior Art

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