A Dual Autoencoder Semantic Feature Fusion-Based Method for Open-Set Recognition of Multifunction Radar Working Modes

Multifunction radars are highly adaptive systems that employ phased array antennas and can simultaneously perform multiple working modes, such as search, tracking, recognition, and guidance. The presence of multiple working modes in multifunction radars poses significant challenges in the reconnaissance of noncooperative target radar emitters, as reconnaissance often involves encountering unknown working modes. Existing radar working mode recognition methods generally perform well with complete databases where all categories of working mode samples are fully accessible. However, relatively little research has addressed open-set learning issues involving unknown working modes. A multifunction radar working mode open-set recognition method based on dual autoencoder semantic feature fusion (DASFF) was proposed to address the challenge of effectively distinguishing unknown modes. This method employs a dual-path autoencoder model to extract and fuse features from the raw pulse sequence and the semantic information sequence, which are closely coupled with the radar's working mode. It effectively enhances the distinguishability of different working mode samples in the high-dimensional feature space. Meanwhile, a hybrid loss function is designed to guide model optimization from various measures, enhancing its ability to extract implicit features. Experiments on simulated and measured datasets demonstrate that the proposed algorithm achieves high recognition accuracy and robustness in non-ideal open scenarios with measurement errors, lost pulses, and spurious pulses.