TRUST-Net: Robust and Uncertainty-Aware Work Mode Recognition of Multifunction Radar Under Non-Ideal Pulse Sequences

The robust and accurate recognition of multifunction radar (MFR) work modes is important for behaviour awareness. However, measurement errors, missing pulses and false pulses can distort the inherent temporal structure of intercepted pulse sequences in nonideal reconnaissance scenarios, degrading recognition performance. To address this issue, this paper proposes a temporal robust and uncertainty-aware mode recognition network (TRUST-Net) for mode recognition with nonideal pulse sequences. TRUST-Net reconstructs informative pulse description word (PDW) features by jointly modelling pulse repetition interval (PRI), radio frequency (RF) and pulse width (PW), enhancing discriminative representation. To capture multiscale temporal dependencies under distorted sequences, we integrate a bidirectional gated recurrent unit with a temporal convolutional network (BiGRU–TCN), combining global context modelling with efficient causal dilated convolutions. To improve reliability under ambiguous samples, an uncertainty-aware recognition module based on Dempster–Shafer (D–S) evidence theory is introduced, where the network estimates class-wise evidence and models the output using a Dirichlet distribution to represent classification confidence and recognition uncertainty. Simulations show that TRUST-Net achieves recognition accuracies of 93.71%, 94.87% and 89.51% under missing pulse, false pulse and composite interference scenarios, with performance improvements exceeding 2.22%, 1.73% and 1.43% over baseline methods. The method demonstrates robustness under complex nonideal reconnaissance scenarios.