Radar Inter-Pulse Modulation Mode Recognition Method Based on Multi-Activation Adaptive Combination Networks

To address the degradation of modulation recognition performance caused by radar pulse loss, measurement errors, and spurious pulses in electronic reconnaissance, this paper proposes a multi-task radar inter-pulse modulation pattern recognition method based on the Multi-Activation Adaptive Combination Network (MAAC-NET). The method constructs heterogeneous nodes using multi-scale convolutions and diverse activation functions within the same network layer. Each node independently extracts temporal features, which are then adaptively fused through a learnable weighted combination and re-activated through a secondary activation stage. Finally, the network simultaneously performs recognition of three types of radar inter-pulse modulation patterns—radio frequency (RF), pulse repetition interval (PRI), and pulse width (PW)—through three task-specific classification heads. Experimental results demonstrate that the proposed method maintains over 90% average recognition accuracy across the three tasks under 70% pulse loss or 70% false pulse rate, and achieves 93.13% overall accuracy under a compound degradation scenario (40% pulse loss, 40% false pulses, and RF/TOA measurement errors). These results validate its effectiveness under both single-source and mixed nonideal observation conditions.