Few-Shot Multitask Network for Joint Detection and Modulation Recognition of LPI Radar Signals

Abstract

Low probability of intercept (LPI) radar signals present significant challenges for modern electronic support systems due to characteristics such as complex modulations and time-frequency energy distributions. Worse still, small sample conditions deteriorate the challenges in complex electromagnetic environments. To address these challenges, we propose a few-shot multitask network FM-Net for joint detection and modulation recognition of LPI radar signals. FM-Net incorporates multitask learning for accurate joint signal detection and modulation recognition by leveraging the shared information between tasks. Concurrently, few-shot learning is incorporated to adapt to scenarios with limited annotated data. Furthermore, FM-Net introduces two key improvements to metric learning: designing class-covariance metric to mitigate biases caused by feature correlations, and employing label smoothing loss to enhance generalization capabilities. Simulation results demonstrate FM-Net achieves superior performance in few-shot joint detection and recognition of LPI radar signals.