In the field of underwater acoustic recognition, machine learning methods rely on a large number of datasets to achieve high accuracy, while the actual collected signal samples are often very scarce, which has a great impact on the recognition performance. This paper presents a recognition method of an underwater acoustic target by the data augmentation technique and the residual convolutional neural network (CNN) model, which is used to expand training samples to improve recognition performance. As a representative model in residual CNN, the ResNet18 model is used for recognition. The whole process mainly includes mel-frequency cepstral coefficient (MFCC) feature extraction, data augmentation processing, and ResNet18 model recognition. On the base of the traditional data augmentation, this study used the deep convolutional generative adversarial network (DCGAN) model to realize the expansion of underwater acoustic samples and compared the recognition performance of support vector machine (SVM), common CNN, VGG19, and ResNet18. The recognition results of the MFCC, constant Q transform (CQT), and low-frequency analyzer and recorder (LOFAR) spectrum were also analyzed and compared. Experimental results showed that the recognition accuracy of the MFCC feature was better than that of other features at the same method, and using the data augmentation method could obviously improve the recognition performance. Moreover, the recognition performance of ResNet18 using data enhancement technology was better than that of other models, which was due to the combination of the data expansion advantage of data augmentation technology and the deep feature extracting ability of the residual CNN model. In addition, although this method was used for ship recognition in this paper, it is not limited to this. This method is also applicable to other target voice recognition, such as natural sound and underwater voice biometrics.

This paper presents a detailed study of the high-order superdirectivity of circular sensor arrays, which is aimed at completing the authors' recently proposed analytical superdirectivity model. From the limit expression of the maximum directivity factor, it is shown that the circular arrays possess good potential for directivity improvement. It is found that the sensitivity function used as a robustness measurement can also be accurately decomposed into a series of closed-form sensitivity functions of eigenbeams, similar to the optimal beampattern and its corresponding directivity factor. Moreover, the performance of eigenbeams can be regarded as an indicator of error sensitivity, and the robustness constraint parameters can be estimated easily. Two specific approaches are proposed for obtaining robust superdirectivity on the basis of robustness analyses, and their performance is demonstrated experimentally.

This letter proposes a robust superdirective frequency-invariant beamforming method for circular sensor arrays. First, the beampattern is accurately transformed into an equivalent form using the property of circular arrays, and the expressions of directivity factor and white noise gain are correspondingly modified. Second, a closed-form weighting vector is accurately derived based on the minimum mean square error approximation to a desired superdirective beampattern that is expressed in a general form. When the number of elements is sufficiently large, the weighting vector can be further simplified, and the white noise gain is also modified to a concise form that is found to be an increasing function of the number of elements in the low-frequency range. Simulations and experimental results show that this method can provide a robust superdirective frequency-invariant beampattern over a wide frequency band given a suitable number of elements, and it can be readily applied to process broadband signals in practice using small circular sensor arrays.

This paper presents an underwater acoustic target recognition method using feature fusion and residual convolutional neural network (CNN). Mel-frequency cepstrum coefficient (MFCC), Gammatone frequency cepstral coefficient (GFCC), low-frequency analyzer and recorder (LOFAR) spectrum, and constant Q transform (CQT) are extracted and fused firstly. On this basis, their Delta features are calculated and fused secondly. The feature dimension is reduced by neighborhood component analysis (NCA). With the fused features after the dimensionality reduction as input features, the residual CNN based on the ResNet18 model is used as classifier to recognize the underwater acoustic target. The other machine learning models, such as support vector machine (SVM), VGG19, and common CNN, are also compared for inputting different features separately. Experimental results show that, MGCL-Delta-NCA-ResNet18 has the best recognition results among these models, with the recognition accuracy of 97.29%, because this model allows full play to the rich information advantages of feature fusion, advantages of feature dimensionality reduction by the NCA and the ability of ResNet18 to extract abundant characteristics. It can also realize the recognition effectively at low signal-to-noise ratio (SNR). Especially at 0 dB, the recognition accuracy can still reach 86.25%. The proposed method can also recognize multi-target signal effectively in the multiple target scenario. Although this model is used in ship and natural voice recognition, it can also be applied to the recognition of other target sounds, such as marine mammals.

This paper proposes an underwater acoustic target recognition method using the Hilbert–Huang transform (HHT) and data augmentation, with a residual convolutional neural network (CNN) as the classifier. The method involves decomposing the signal via empirical mode decomposition (EMD), denoising based on the principal component analysis (PCA) algorithm, extracting the HHT spectrum features of underwater acoustic targets, performing data augmentation using a deep convolutional generative adversarial network (DCGAN), and employing a ResNet18 model for recognition. The results of different models, including support vector machine (SVM), ordinary CNN, VGG19, and ResNet18 are compared. This paper also analyzes and compares the recognition results of different input features, such as Mel-frequency cepstrum coefficient (MFCC), constant Q transform (CQT), HHT spectrum, and HHT marginal spectrum. The results show that the denoising method based on PCA effectively reduces noise and redundancy. Compared to other features, the HHT spectrum provides better recognition accuracy. Data augmentation can signific ntly improve the accuracy. The residual CNN with the input feature of the HHT spectrum using the data augmentation achieves the best recognition performance, proving effective for recognition at low signal-Bto-noise ratios (SNRs). Especially at 0dB, the recognition accuracy can still reach 90.03%.