This paper considers the problem of maneuvering target tracking in the clutter environment and proposes a graph attention autoencoder tracking network. Initially, the proposed method constructs the graph by using historical track points and measurement data of the current frame, including the target measurement and clutters, as vertices. Then, the graph is fed into the designed graph attention autoencoder (GAE) network to learn the relationships among the historical trajectory and the measurements of the current frame. Finally, the GAE outputs the tracking result of the current frame. The experimental results highlight that the proposed GAE can outperform the classic maneuvering target tracking algorithms in the clutter environment with low and high maneuverability.

In this letter, we provide the target signal enhancement method based on deep learning for weak target detection. First, the proposed method fully considers the nature characteristic of radar complex echoes and exploits the complex-valued neural networks. Then, the architecture of weak target enhancement complex-valued generative adversarial network (WTE-CGAN) is proposed. More specifically, the generator loss function of generative adversarial network (GAN) is modified, which can be used to reflect the difference between the generated target signal by the generator and the label signal. To keep the training stability of the proposed method, a gradient penalty factor is randomly added to every sample, which embodies the loss function of discriminator. Finally, simulation and measured experiments are given to demonstrate the effectiveness of the proposed method compared with other methods, and it has a significant signal enhancement effect on weak targets.

Human motion recognition based on ultra-wideband through-the-wall radar (UWB TWR) (a radar whose fractional bandwidth of the radar transmitted signal is bigger than 0.25) is faced with the problems of too few samples and the limitation of perspective. In this letter, we propose a multiradar cooperative human motion recognition model based on transfer learning and ensemble learning. Specifically, a ResNeXt network model based on transfer learning is first proposed to deal with the problem of too few samples. The model is pretrained on the public ImageNet database, and then it is transferred to the task of human motion recognition based on multiradar. Compared with a typical convolutional neural network from scratch, the ResNeXt network model based on transfer learning requires shorter epochs and achieves higher accuracy. Then, to solve the problem of model accuracy decline caused by the limitation of perspective, a multiradar human motion recognition model based on ensemble learning is proposed. Experimental results show that compared with the fusion model based on single-view radar, the recognition accuracy of network based on ensemble learning can be higher.

Accurate sensing of the main-lobe active deception jamming is critical for radar anti-jamming and extended target detection in complex electromagnetic environment. This letter therefore deals with the problem of multiple active deception jamming recognition in extended target settings. A residual convolutional neural network with attention mechanism-based radar active deception jamming recognition algorithm is proposed leveraging a hybrid model to capture much rich features through multi-domain feature fusion. The proposed method can outperform state-of-the-art methods in terms of recognition accuracy, model size, and convergence speed. Experimental results demonstrate its effectiveness and robustness.

This letter considers the problem of human activity recognition (HAR) behind the walls using ultra-wideband (UWB) radar. The graph convolutional network (GCN)-enhanced multi-domain fusion network (GMFN) is proposed to improve the recognition performance utilizing the complementarity of the multi-domain features. Specifically, firstly, a multi-branch convolutional neural network (CNN) is proposed to extract the multi-domain features from the range, time-frequency, and range-Doppler domain. Then the multi-domain features are constructed as a graph, and the GCN is employed to fuse the multi-domain features on the graph. Finally, HAR is implemented in the form of the graph classification. The experimental results on the real data show that the proposed GMFN achieves better performance than the state-of-the-art multi-domain fusion HAR methods.

This letter considers the problem of the target detection in the nonhomogeneous sea clutter environment, and proposes the complex-valued U-Net based clutter suppression method. Specifically, firstly, the complex signal features of radar echo sequences are extracted by developing the complex-valued convolutional blocks. Secondly, the complex multi-level features are fused, by employing the up-down sampling structure and skip connections, to suppress nonhomogeneous sea clutter. Further, the false alarm controllable detector is designed to detect the targets. Finally, the performance of the proposed method is evaluated via real data. The results show that it has a higher detection probability compared with the real-valued U-Net.