Novel Jamming Pattern Class Discovery via Meta-Metric in Few-Shot Scenarios

The increasing complexity of electromagnetic environments and intelligent jamming poses challenges for discovering novel jamming patterns with limited labeled samples. This letter proposes a meta-metric learning framework for few-shot Novel Jamming Pattern Class Discovery (NJPCD). A Disentangled Meta-Allocator is introduced to construct view-consistent episodic tasks and impose metric constraints, improving class separability and representation stability in jamming pattern recognition. To address the scarcity of novel jamming samples, a prototype-based meta-learning network enables effective cross-task adaptation and reliable novel-class discovery under minimal supervision. By integrating metric regularization into few-shot meta-training, the proposed framework enhances intra-class compactness and inter-class separability of learned embeddings. Simulation results demonstrate superior accuracy and stable performance across known, unknown, and mixed jamming scenarios.