Self-supervised representation learning approach for trajectory of guided missile using auto encoder

Abstract

and (3) importance of anomaly information lies in its ability to enhance the performance of learning good representations. Evaluation on various features and guidance phases shows that different types of features and guidance phases have varying influences on classification performance, with acceleration variable and initial guidance phase being the most informative for anomaly detection. The proposed method in this paper has the potential to significantly improve the effectiveness and efficiency of anomaly detection in guided missile trajectories and can be applied in other types of trajectories and anomaly detection tasks.

(2) comparing the influence of each guidance phase on classification performance

This paper proposes Trajectory Auto-Encoder (TrAE), a self-supervised representation learning approach based on guided missile trajectory data. TrAE is designed to identify anomalies by learning compact representations from partial trajectory states without labels, which can effectively improve the performance of anomaly detection. Furthermore, TrAE can learn good representations of Normal set data even with limited labeled anomaly data, indicating its potential for use in scenarios where labeled anomaly data is scarce. By applying a self-supervised approach, which is widely used in many machine learning tasks, this paper demonstrates that (1) selecting one feature variable group among position, velocity, acceleration, angular velocity, Euler angle, and command input can lead to good representations, implying the tendency of the trajectory