Radar signal recognition plays a vital role in electronic warfare. For the multifunction radars (MFRs) with complex dynamical modes, the signal recognition needs to identify not only the emitter but also its current functional state. Existing research on MFR recognition mainly focuses on hierarchical modelling approaches. Inspired by recent progress of deep neural networks, the authors propose to further develop radar signal modelling with recurrent neural networks. Here, the authors propose a more efficient method for functional state recognition of MFRs based on a gated recurrent unit (GRU). The method makes full use of the ability of GRUs to automatically learn the characteristics of radar signal sequences, and the nonlinear modelling of GRUs can deal with the corrupted data well. Simulation results on the Mercury radar show that the proposed method achieves a great recognition accuracy with little prior information, and the designed GRU model outperforms the predictive state representation model under the same condition.

In regards to electromagnetic warfare application, this work proposes a radar intra-pulse recognition algorithm based on bicubic interpolation Wigner-Ville distribution (WVD). The proposed method is aimed at overcoming the complexity of traditional feature extraction and enhancing the robustness of intelligent intra-pulse recognition systems. Firstly, we convert the WVD matrices of radar intra-pulse signals into square matrices using bicubic interpolation. This enables the intra-pulse recognition algorithm to handle radar signals with varying lengths. Secondly, we use the square matrices directly to train a convolutional neural network (CNN), rather than first saving them as images and then training the network with those images. This is because automatic storage and retrieval of images can sometimes alter the numerical values in the matrices. We enhance the performance of the CNN by utilizing batch normalization and one-hot encoding techniques, and continuously evaluate the CNN during training to save the best performing model. Finally, we use the trained CNN to identify the intercepted radar intra-pulses, and verify the reliability of the proposed method based on radar signals generated by hardware devices. Experimental results demonstrate that the proposed algorithm can recognize radar intra-pulses with varying pulse widths. Furthermore, the proposed algorithm exhibits strong recognition performance even at low signal-to-noise ratio, and has lower time complexity than existing algorithms.

Recent studies have proven that synthetic aperture radar (SAR) automatic target recognition (ATR) models based on deep neural networks (DNN) are vulnerable to adversarial examples. However, existing attacks easily fail in the case where adversarial perturbations cannot be fully fed to victim models. We call this situation perturbation offset. Moreover, since background clutter takes up most of the area in SAR images and has low relevance to recognition results, fooling models with global perturbations is quite inefficient. This paper proposes a semi-white-box attack network called Universal Local Adversarial Network (ULAN) to generate universal adversarial perturbations (UAP) for the target regions of SAR images. In the proposed method, we calculate the model’s attention heatmaps through layer-wise relevance propagation (LRP), which is used to locate the target regions of SAR images that have high relevance to recognition results. In particular, we utilize a generator based on U-Net to learn the mapping from noise to UAPs and craft adversarial examples by adding the generated local perturbations to target regions. Experiments indicate that the proposed method effectively prevents perturbation offset and achieves comparable attack performance to conventional global UAPs by perturbing only a quarter or less of SAR image areas.

With the widespread use of multifunction radars (MFRs), it is hard for the traditional radar signal recognition technology to meet the needs of current electronic intelligence systems. For signal recognition of an MFR, it is necessary to identify not only the type or individual of the emitter but also its current state. Existing methods identify MFR states through hierarchical modeling, but most of them rely heavily on prior information. In the paper, we focus on the MFR state recognition with actual intercepted MFR signals and develop it by introducing recurrent neural networks (RNNs) of deep learning into the modeling of MFR signals. According to the layered MFR signal architecture, we propose a novel end-to-end state recognition approach with two RNNs’ connections. This approach makes full use of RNNs’ ability to directly tackle corrupted data and automatically learn the features from input data. So, it is practical and less dependent on prior information. In addition, the hierarchical modeling method applied to the end-to-end network effectively restricts the scale of the end-to-end model so that the model can be trained with a small amount of data. Simulation results on a real MFR show the excellent recognition performance of our end-to-end approach with little prior information.

Recent studies have proven that synthetic aperture radar (SAR) automatic target recognition (ATR) models based on deep neural networks (DNN) are vulnerable to adversarial examples. However, existing attacks are easily failed in the case where adversarial perturbations cannot be fully fed to victim models. We call this situation perturbation offset. Moreover, since background clutter takes up most of the areas in SAR images and has low relevance to recognition results, fooling models with global perturbations is quite inefficient. This paper proposes a semi-whitebox attack network, called Universal Local Adversarial Network (ULAN), to generate universal adversarial perturbations (UAP) for the target regions of SAR images. In the proposed network, we calculate the model’s attention heatmaps through layer-wise relevance propagation (LRP), which is used to locate the target regions of SAR images that have high relevance to recognition results. In particular, we utilize a generator based on the U-Net to learn the mapping from noise to UAPs and craft adversarial examples by adding the generated local perturbations to target regions. Experiments indicate that the proposed method fundamentally prevents perturbation offset and achieves comparable attack performance to conventional global UAPs by perturbing only a quarter or less of SAR image areas.