Following a geometric paradigm, the estimation of a Toeplitz structured covariance matrix is considered. The estimator minimizes the distance from the Sample Covariance Matrix (SCM) while complying with some specific constraints modeling the covariance structure. The resulting constrained optimization problem is solved globally resorting to the Dykstra' projection framework. Each step of the procedure involves the solution of two convex sub-problems, whose minimizers are available in closed form. Simulation results related to typical radar environments highlight the effectiveness of the devised method.

Traditional detection and tracking methods struggle with the complex and dynamic maritime environment due to their poor generalization capabilities. To address this, this paper improves the YOLOv5 network by integrating Transformer and a Convolutional Block Attention Module (CBAM) with the multi-frame image information obtained from radar scans. It proposes a detection and tracking method based on the Detection Tracking Network (DTNet), which leverages transfer learning and the DeepSORT tracking algorithm, enhancing the detection capabilities of the model across various maritime environments. First, radar echoes are preprocessed to create a dataset of Plan Position Indicator (PPI) images for different marine conditions. An integrated network for detecting and tracking maritime targets is then designed, utilizing the feature differences between moving targets and sea clutter, along with the coherence of inter-frame information for moving targets, to achieve multi-target detection and tracking. The proposed method was validated on real maritime targets, achieving a precision of 99.06%, which is a 7.36 percentage point improvement over the original YOLOv5, demonstrating superior detection and tracking performance. Additionally, the impact of maritime regions and weather conditions is discussed, showing that, when transferring from Region I to Regions II and III, the precision reached 92.2% and 89%, respectively, and, when facing rainy weather, although there was interference from the sea clutter and rain clutter, the precision was still able to reach 82.4%, indicating strong generalization capabilities compared to the original YOLOv5 network.

Micro-Doppler (m-D) signals are susceptible to interference from a large number of Doppler signals and ambient noise, and the single use of micro-Doppler signatures for the classification of small, slow, and low-speed targets poses certain limitations. In this paper, a dynamic multi-feature data fusion neural network(DMFFNNet) classification method is proposed. First, K-band frequency-modulated continuous-wave(FMCW) radar is used to collect echo data from five types of rotorcraft UAVs and Bionic bird. After preprocessing the data, 2D range period graphic and 2D time-frequency spectrograms are obtained. we investigate the construction of new data representations in the range-periodic domain, designing networks to extract dynamic time-varying features of the data. In order to be able to obtain accurate localized features, a local feature extraction module is proposed to extract local features from the range period graph, while a global feature extraction module is used to extract global features from the time-frequency spectrograms. In order to be able to extract dynamic information about the data, a 3D network is employed to capture dynamic change feature in the 3D range period data. Finally, a feature fusion module is designed to integrate the extracted features, and in order to be able to better extract the features of the target, an attention mechanism is added to the fusion network to extract the temporal and spatial features in the spectrogram and fuse them to further improve the overall performance of the model. Experimental results show that, compared to single-channel CNN classification methods, incorporating dynamic feature data enables the network to achieve better classification accuracy.

Excellent performance in supervised learning-based radar signal modulation recognition (RSMR) techniques rely on the quantity and quality of labeled datasets, while the high cost and difficulty involved in analyzing and labeling radar signal samples limits its development. A RSMR algorithm using self-supervised contrastive learning (SSCL) methodology is proposed to address this issue. Specifically, within the classical contrastive learning framework MoCo V2, a customized data augmentation method is devised to capture time-frequency features of the radar signal. Additionally, the feature extraction network ResNet50 is improved by decoupling spatial and channel filters, resulting in greater sensitivity to the time-frequency features. To enhance the recognition accuracy, two loss functions, alignment and uniformity, are employed in place of the info noise contrastive estimation (InfoNCE) loss, and both loss functions are optimized directly. The recognition accuracy of the proposed method can reach 97.66% at a signal-to-noise ratio (SNR) of 4 dB.