The eigenvalues of the sample covariance matrix can capture signal correlations and noise characteristics well, which are widely used for spectrum sensing in cognitive radio networks. Some totally blind spectrum sensing algorithms were proposed, such as maximum eigenvalue-to-arithmetic mean (ME–AM), maximum eigenvalue-to-geometric mean (ME–GM), and arithmetic-to-geometric mean of eigenvalues (AGM) methods. This article makes full use of the advantages of these algorithms and proposes a universal spectrum sensing algorithm based on maximum eigenvalue, arithmetic mean, and geometric mean of eigenvalues. The proposed algorithm takes the weighted geometric mean of the test statistics of the ME–AM and AGM algorithms as new test statistic and includes the ME–AM, ME–GM, and AGM algorithms as special cases. Following the random matrix theory framework, in a similar vein to the ME–AM and ME–GM algorithms, we derive the analytical expressions of false alarm probability, threshold, and detection probability of the proposed method using the Tracy–Widom distribution of maximum eigenvalue. In addition, we discuss the problem of designing the optimal weighting factor through theoretical analysis and simulation experiments. Finally, simulation results verify that the universal algorithm can effectively improve the detection performance.

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.

Proper prior knowledge of target scattering centers (SCs) can help to obtain better detection performance of range-spread targets. However, target SCs are sensitive to the target’s attitude relative to the radar and vary significantly among different targets. The existing approaches that employ pre-determined prior knowledge may suffer performance degradation when the prior information does not match the practical scenarios. A possible way to circumvent this drawback is to estimate the SCs of different targets adaptively and check the presence of a target utilizing the range cells occupied by the most likely target SCs. For this reason, this paper develops a generalized likelihood ratio test based on adaptive SCs estimation (ASCE-GLRT) for range-spread target detection in compound-Gaussian clutter. Under the assumption that the target SCs are sparse, we model the problem of SCs estimation as a sparse signal representation. Moreover, since the sparse assumption may not always be satisfied in practice, a modified sparsity regularization method is proposed to enhance the robustness of the estimation performance of targets with different scattering characteristics. Theoretical analysis shows that the proposed detector can achieve the constant false-alarm rate property. The performance assessments conducted by numerical simulation and field tests confirm the effectiveness and robustness of the proposed detector.

This paper deals with the problem of marine target detection utilizing underlying spatial-temporal properties of the radar echoes through the deep learning technology. A spatial-temporal graph neural network (ST-GNN) detector is proposed for efficient detection, which directly extracts the spatial-temporal features from the echoes without the data preprocessing operation and detects the target using the extracted features. Specifically, the ST-GNN detector first extracts high-dimensional feature from the echoes using convolutional neural network (CNN). Then, it constructs the high-dimensional features as a spatial-temporal graph, from which the spatial-temporal features is jointly extracted. Finally, it utilizes the nonlinear feature dimension reduction and maps the features into the binary probability space to obtain the detection results. Experimental results on two public radar databases for marine target detection validate the effectiveness and superiority of the proposed detector against several popular marine target detection methods from the aspects of detection performance and real-time test efficiency.

This paper focuses on the design of the optimal or near-optimal detector resorting to extreme eigenvalues. A general framework for detector design involving model-driven and data-driven approaches is introduced. Specifically, the extreme eigenvalues based likelihood ratio test (LRT) is derived via the model-driven approach. Merging the model-driven and data-driven approaches, the Naive Bayesian detector is proposed based on the extreme eigenvalues, which converts the design of test statistic into a two-class decision boundary construction problem, and a solution is provided by the Naive Bayesian classifier. To render the detectors more practical, two near-optimal detectors called $\alpha $ -sum and $\alpha $ -product of maximum and minimum eigenvalues ( $\alpha $ -SMME, $\alpha $ -PMME) are further designed, in which $\alpha $ is a weight coefficient. Furthermore, the theoretical performance analysis of the $\alpha $ -SMME and $\alpha $ -PMME algorithms is provided, and the optimal weight selection is further obtained by solving an optimization problem under the Neyman-Pearson criterion. Finally, simulation experiments demonstrate that the proposed detectors achieve performance improvements over the state-of-the-art detectors using extreme eigenvalues, and almost coincide with the detection performance of the LRT detector.

In this letter, the correlation structure inherent in the received data is exploited to improve detection performance for sea-surface small floating targets in short observation time. Three detectors are devised resorting to the eigenvalues of covariance matrix that are versatile statistics reflecting the signal correlation. Specifically, the proposed detectors respectively exploit the maximum eigenvalue to arithmetic mean (MAM) of all eigenvalues, the maximum eigenvalue to geometric mean (MGM) of all eigenvalues, and the maximum eigenvalue to minimum eigenvalue (MME) to form test statistics. At the analysis stage, the three-parameter Burr function is exploited to approximate the statistical distributions of the test statistics under null hypothesis and alternative hypothesis. Besides, the analytic expressions of false alarm probability, detection probability, and thresholds of the proposed detectors are derived. Finally, simulation results on real sea clutter show that the proposed detectors provide effective solutions to the problem of target detection in sea clutter.