To address the issue associated with poor accuracy rates for specific emitter identification (SEI) under low signal-to-noise ratio (SNR) conditions, where the single-dimension radar signal characteristics are severely affected by noise, we propose an attention-enhanced dual-branch residual network structure based on the adaptive large-margin Softmax (ALS). Initially, we designed a dual-branch network structure to extract features from one-dimensional intermediate frequency data and two-dimensional time–frequency images, respectively. By assigning different attention weights according to their importance, these features are fused into an enhanced joint feature for further training. This approach enables the model to extract distinctive features across multiple dimensions and achieve good recognition performance even when the signal is affected by noise. In addition, we have introduced L-Softmax to replace the original Softmax and propose the ALS. This approach adaptively calculates the classification margin decision parameter based on the angle between samples and the classification boundary and adjusts the margin values of the sample classification boundaries; it reduces the intra-class distance for the same class while increasing the inter-class distance between different classes without the need for cumbersome experiments to determine the optimal value of decision parameters. Our experimental findings revealed that, in comparison to alternative methods, our proposed approach markedly enhances the model’s capability to extract features from signals and classify them in low-SNR environments, thereby effectively diminishing the influence of noise. Notably, it achieves the highest recognition rate across a range of low-SNR conditions, registering an average increase in recognition rate of 4.8%.

In the real world, radar specific emitter identification (SEI) faces complex open set recognition (OSR) scenarios, where traditional closed-set classifiers cannot distinguish between known and unknown categories similarly distributed in the feature space. Furthermore, feature overlap in the open space challenges the design of judgment thresholds. Different from most current approaches, we propose a dynamic boundary adversarial model based on convolutional prototype learning (CPL-DBA). It assigns magnetization weights of various directions and magnitudes to samples during training, depending on the features' distances from the prototype center. As a result, the model clusters samples that are closer to the prototype center while repelling extreme samples farther away. Additionally, we enhance the detection performance of unknown classes by introducing dynamic compact constraint terms. We design trainable closed dynamic boundaries for each known class prototype. This allows for adaptive expansion and contraction of the closed classification boundaries. Consequently, the dynamic boundaries improve the compactness of known features, which can also serve as detection thresholds for unknown classes. Extensive experimental results on the three simulated datasets, one simulator acquisition dataset, and one actual collected dataset indicate that the proposed dynamic framework achieves the best performance compared to other OSR methods.

To address the challenge of effectively sorting radar signals with identical modulation types and overlapping parameter ranges in complex electromagnetic environments using traditional methods, this paper proposes a one-dimensional convolutional neural network, U1DADM, based on semantic segmentation. This network extracts deep semantic information from preprocessed digitized signals and introduces a masking strategy for overlapping pulse regions. The proposed digitized data processing method achieves joint modeling of intra-pulse and inter-pulse features. The core module of the U1DADM network, adaptive dilated convolution, achieves multi-scale feature fusion and long-range feature dependency modeling through dynamic receptive field adjustment. The masking strategy mitigates data defects and assists the network in focusing on repetitive pulse regions. Experimental results indicate that the proposed method surpasses similar deep learning methods and traditional sorting methods under various data conditions. Furthermore, it exhibits strong robustness and generalization ability even under harsh conditions with spurious and missed pulses, achieving high-precision radar signal sorting.