A Semi-Supervised Prototype Contrastive Learning Framework for Unknown Specific Emitter Identification

Zhisheng Tan

doi:10.1109/lwc.2026.3740799

A Semi-Supervised Prototype Contrastive Learning Framework for Unknown Specific Emitter Identification

2026-10-06

0

PRE

AI

DOI：10.1109/lwc.2026.3740799

Original

Original request for help

Save

Abstract

En 中文

In open scenarios, unlabeled samples may belong to both known and unknown categories. Therefore, a practical Specific Emitter Identification (SEI) method should be capable of discovering novel categories from such unlabeled data. Toward this end, we propose a Semi-Supervised Prototype Contrastive Learning framework, dubbed SemiPC, which jointly models both known and novel classes within a unified latent space to balance performance across them. Experimental results demonstrate that SemiPC achieves 97.91% overall accuracy and 96.99% novel-class accuracy on the WiSig dataset, outperforming existing methods.

Keywords:

Specific emitter identification

generalized category discovery

semi-supervised learning

deep learning

A Semi-Supervised Prototype Contrastive Learning Framework for Unknown Specific Emitter Identification

Abstract

Journal

Organization

Cited Papers