Discriminability-Driven Spatial-Channel Selection with Gradient Norm for Drone Signal OOD Detection

TL;DR

This work tackles drone signal OOD detection in interference-rich environments by introducing discriminability-driven spatial-channel selection (DDSCS) and a gradient-norm perturbation sensitivity measure. The method adaptively weights time-frequency features across spatial and channel dimensions based on inter-class dispersion, while the gradient norm captures prediction instability under perturbations; these cues are fused with an energy-based score to produce joint inferences. Empirical results on a drone signal dataset show DDSCS outperforms baselines in accuracy, recall, F1, and AUROC, with strong resilience across SNR levels and drone types. The approach advances practical OOD detection for radar-free, robust drone operation by leveraging protocol-specific TF patterns and dynamic stability signals.

Abstract

We propose a drone signal out-of-distribution (OOD) detection algorithm based on discriminability-driven spatial-channel selection with a gradient norm. Time-frequency image features are adaptively weighted along both spatial and channel dimensions by quantifying inter-class similarity and variance based on protocol-specific time-frequency characteristics. Subsequently, a gradient-norm metric is introduced to measure perturbation sensitivity for capturing the inherent instability of OOD samples, which is then fused with energy-based scores for joint inference. Simulation results demonstrate that the proposed algorithm provides superior discriminative power and robust performance via SNR and various drone types.

Figures (4)

• Figure 1: WEM performance of various algorithms via SNR.
• Figure 2: WEM of the proposed algorithm versus various drone types.
• Figure 3: Sensitivity analysis of selection parameters $\alpha$ and $\beta$.
• Figure 4: Impact of fusion weight $\lambda$ on detection performance.