Unmanned aerial vehicle flight data anomaly detection method based on multi-scale spatio-temporal graph convolution network

CN122286244APending Publication Date: 2026-06-26AIR FORCE UNIV PLA
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
AIR FORCE UNIV PLA
Filing Date
2026-02-28
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing methods for detecting anomalies in UAV flight data are ineffective at capturing complex spatiotemporal correlations in multivariate flight data, and are prone to high false alarm rates in complex environments, making them difficult to adapt to multi-scale and dynamic changes.

Method used

Multi-scale spatiotemporal graph convolutional network (MSTGCNet) is used for anomaly detection in UAV flight data. Normalization and embedding modules are used to enhance data stationarity, graph augmentation hybrid expert module captures cross-scale spatiotemporal relationships, and an adaptive threshold strategy is used to adjust the detection threshold.

Benefits of technology

It achieves comprehensive modeling of spatiotemporal dependencies in UAV flight data, reduces false alarm rate, improves robustness and adaptability of detection, and ensures accurate anomaly detection in dynamic environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122286244A_ABST
    Figure CN122286244A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of UAV detection technology. Addressing the problems of existing methods neglecting inherent scale variations and complex spatiotemporal dependencies in flight data, and relying on fixed thresholds leading to high false alarm rates, this invention proposes a UAV flight data anomaly detection method based on multi-scale spatiotemporal graph convolutional networks. The method includes the following steps: acquiring and preprocessing UAV flight data records; establishing a standardized dataset; constructing a multi-scale spatiotemporal graph convolutional network model; inputting the standardized dataset into the multi-scale spatiotemporal graph convolutional network model; extracting multi-scale features from the input sample sequences to capture spatiotemporal dependencies in the flight data; and employing an adaptive threshold strategy to dynamically adjust the threshold for anomaly detection targets within a sliding window to obtain the final anomaly detection result. This invention effectively achieves multi-scale feature extraction and captures spatiotemporal dependencies, and dynamically adjusts the detection threshold within a sliding window based on local statistical information, significantly reducing the false alarm rate.
Need to check novelty before this filing date? Find Prior Art