Presentation Details
  1. Home
  2. Presentation List
  3. Presentation Details
Anomaly Detection in Satellite Using a New Self-Supervised Contrastive Learning Method based on Interleaved Cross Attention Networks
ID：96 View Protection：ATTENDEE Updated Time：2025-11-10 15:17:47 Hits：70 Poster Presentation

Start Time：Pending（Asia/Shanghai）

Duration：Pending

Session：No Session »

No files

Abstract
    In the domain of satellite anomaly detection using deep learning, traditional reconstruction-based methods often fall short when addressing the challenges posed by small sample problem and high sensitivity requirements. To overcome these limitations, this paper introduces a new method based on a contrastive learning framework. Specifically, we propose the Interleaved Cross Attention Networks (ICAN) as the representation model. The core module, Interleaved Cross Attention (ICA), is designed based on the observation that the temporal dependencies of interleaved down-sampled subsequences of telemetry features are highly similar to those of the original sequences, with minimal loss of critical information. Thus, ICAN captures the flow of information across interleaved time steps, effectively reducing computational complexity while enhancing the extraction of temporal dependencies in telemetry sequences. Furthermore, this paper presents a novel contrastive learning framework centered on ICAN. This framework incorporates a series of efficient modules, including methods for generating positive and negative sample pairs, an adaptive margin contrastive loss function, an anomaly criterion based on Exponentially Weighted Moving Average (EWMA), and an anomaly thresholding method based on Drift-aware Streaming Peaks Over Threshold (DSPOT). These techniques collectively improve the effectiveness of anomaly detection. Finally, experimental results on the AACS dataset, demonstrate the efficacy and superiority of the proposed method.
Keywords
Anomaly detection, satellite telemetry data, contrastive learning, time series.
Speaker
Haotian Zhao
Associate Researcher Harbin Institute of Technology

Submission Author
Haotian Zhao Harbin Institute of Technology
Ming Liu Harbin Institute of Technology
Lixian Zhang Harbin Institute of Technology
Submit Comment
Verify Code Change Another
All Comments
Important Date

  • Conference Date

    Nov 21

    2025

    to

    Nov 23

    2025

  • Oct 20 2025

    Draft paper submission deadline

  • Dec 08 2025

    Registration deadline

Sponsored By
IEEE Instrumentation and Measurement Society
South China University of Technology
Organized By
South China University of Technology
Contact Information
Login to view full contact details
Previous Conferences