Domain-Adaptive UAV Recognition Using IR-UWB Radar and a Lightweight Mamba-Based Network

Impulse radio ultrawideband (IR-UWB) radar shows great promise for uncrewed aerial vehicle (UAV) detection due to its high resolution, strong penetration, and robustness against multipath interference. However, effectively leveraging both spatially static and temporally dynamic information in radar echoes, while overcoming environmental interference, remains challenging. To address this, we propose a lightweight domain-adaptive model, AIR-Mamba. It first employs adaptive gain control and discrete wavelet decomposition to reduce amplitude sensitivity and extract target micro-Doppler features, then utilizes a Mamba backbone based on state-space models (SSMs) to capture long-term motion dynamics. We also introduce a dual-adaptation strategy that combines adversarial learning and correlation alignment (CORAL) to align cross-domain features and enhance generalization. To address real data scarcity, we constructed a multiscenario UAV echo dataset using full-wave electromagnetic simulation, which was validated by measurements in a microwave anechoic chamber. Experimental results show that AIR-Mamba achieves a cross-environment classification accuracy over 96% with only 1.55M parameters, while exhibiting strong noise resistance. This performance demonstrates clear advantages in model size and accuracy, providing a practical solution for real-time UAV detection in resource-constrained environments.