In this article, we propose an active target detector by using a shallow neural network (NN) with novel features under small sonar data, where deep learning (DL) models are restricted. The features are contrived by considering differences between target and clutter signals in the temporal variation of amplitude and frequency when using an active ping; these are extracted from preprocessed spectrograms of beam time series. The classification ability of the features is demonstrated through comparison with that of prevalent timbral features in the training data set that reflect the human auditory system. Finally, the detection performance of the shallow NN trained using the proposed features is compared with those from the fine-tuned DL models. During the generalization test, the DL models miss a significant number of targets in the test data set—including target signals with patterns unseen during training—and the VGG16 used for detailed comparison has a detection probability of 0.69. The probability increases to 0.86 via the shallow NN, with remarkably enhanced target detection. This demonstrates the competitive performance and generalization ability of the shallow NN, using features that consider active sonar returns under a lack of data.

This paper presents a conceptual model design for entity-level combat simulation to enhance model composability and reusability. For conceptual modeling, we first describe the following three problem situations: (1) joint design of logical/physical modeling; (2) flexible model modification during scenario extension; and (3) cooperative modeling with different domain experts. To this end, we propose a two-dimensional model partition method for a combat entity, which partitions the combat entity depending on the functional aspect horizontally and the abstract level vertically. Thus, the proposed method guarantees transparent simplification of the combat entity and facilitates flexible model composition when simulating in the integration and the interoperation environments. Based on the proposed conceptual modeling, empirical measurements demonstrate the enhancement of model composability and reusability during scenario extension for anti-submarine warfare from static decoy to mobile decoy and from pattern-running torpedo to wire-guided torpedo simulations.

In this paper, a compressive sensing-based data fitting direction-of-departure/direction-of-arrival (DOD/DOA) estimation algorithm is proposed to apply the superior performance of compressive sensing method to the bistatic MIMO sonar systems. The algorithm proposed in this paper optimizes the output data via convex optimization-based sparse recovery, so that it is possible to estimate the DOD and the DOA for each target accurately. In order to minimize the amount of computation, the cost function with constraint condition is implemented in this paper. Furthermore, the constraint condition parameter of the cost function is analytically derived. Through various simulations, it is shown that the superior DOD and DOA estimation performance of the proposed algorithm and that the analytical derivation of the constraint condition parameter is useful for determination of regularization parameter.

Accurate estimation of the frequency component is an important issue to identify and track marine objects (e.g., surface ship, submarine, etc.). In general, a passive sonar system consists of a sensor array, and each sensor receives data that have common information of the target signal. In this paper, we consider multiple-measurement sparse Bayesian learning (MM-SBL), which reconstructs sparse solutions in a linear system using Bayesian frameworks, to detect the common frequency components received by each sensor. In addition, the direction of arrival estimation was performed on each detected common frequency component using the MM-SBL based on beamforming. The azimuth for each common frequency component was confirmed in the frequency-azimuth plot, through which we identified the target. In addition, we perform target tracking using the target detection results along time, which are derived from the sum of the signal spectrum at the azimuth angle. The performance of the MM-SBL and the conventional target detection method based on energy detection were compared using in-situ data measured near the Korean peninsula, where MM-SBL displays superior detection performance and high-resolution results.

We propose an active sonar target classifier using attention-based complementary learning (ABCL) to mitigate poor generalization with scarce active sonar data. ABCL utilizes the information of different features (spectrograms of short-time Fourier transform and constant-Q transform in the present study), which provide distinct perspectives for the same raw data. The complementary learning module of ABCL first summarizes the input features using a transformer encoder layer, including self-attention. Subsequently, the attention mechanism is applied to the refined input features from the transformer encoder layer to generate a comprehensive feature with better generalization using limited active sonar data. The comprehensive feature from two different features enables a shallow network to discriminate between active targets and nontargets and allows ABCL to be light. Two active sonar datasets from two oceanic environments were used for training and testing. The generalization of ABCL was examined using a receiver operating characteristic curve. ABCL showed a comparable performance to state-of-the-art deep learning (DL) models in vision in a test scenario and achieved the best performance with low variance in the other test scenario, where other DL models showed a poor probability of detection at a low probability of false alarms.

Passive sonar systems are used to detect the acoustic signals that are radiated from marine objects (e.g., surface ships, submarines, etc.), and an accurate estimation of the frequency components is crucial to the target detection. In this paper, we introduce sparse Bayesian learning (SBL) for the frequency analysis after the corresponding linear system is established. Many algorithms, such as fast Fourier transform (FFT), estimate signal parameters via rotational invariance techniques (ESPRIT), and multiple signal classification (RMUSIC) has been proposed for frequency detection. However, these algorithms have limitations of low estimation resolution by insufficient signal length (FFT), required knowledge of the signal frequency component number, and performance degradation at low signal to noise ratio (ESPRIT and RMUSIC). The SBL, which reconstructs a sparse solution from the linear system using the Bayesian framework, has an advantage in frequency detection owing to high resolution from the solution sparsity. Furthermore, in order to improve the robustness of the SBL-based frequency analysis, we exploit multiple measurements over time and space domains that share common frequency components. We compare the estimation results from FFT, ESPRIT, RMUSIC, and SBL using synthetic data, which displays the superior performance of the SBL that has lower estimation errors with a higher recovery ratio. We also apply the SBL to the in-situ data with other schemes and the frequency components from the SBL are revealed as the most effective. In particular, the SBL estimation is remarkably enhanced by the multiple measurements from both space and time domains owing to remaining consistent signal frequency components while diminishing random noise frequency components.