Sonar Target Classification Research Articles

Performance improvement of active sonar target classification using feature fusion

In this study, we propose a neural network architecture considering feature fusion to improve active sonar target classification, where training data is scarce. We use various hand-crafted features with different perspectives for the active sonar one- and two-dimention raw data to enhance target classification performance. We use two kinds of features, in 1D feature and 2D features, because different perspectives on the same sample complement effectively by exchanging. For this, we contrived attention based complementary learning module to extract comprehensive features using the two kinds of features. To verify the generalization performance of this model, two real-ocean datasets were used and the classification results are compared with existing DL models.

Temporal-Feature-Based Classification of Active Sonar Targets in a Deep-Water Environment

Temporal-Feature-Based Classification of Active Sonar Targets in a Deep-Water Environment

SONAR target classification with complex-valued neural networks

Popular acoustic signal processing techniques analyze acoustic color, which is a magnitude representation of a frequency spectrum. This paradigm allows for easy visualization of results and simpler models at the cost of throwing out phase information. Analysis and modeling of complex-valued data does have inherent difficulty from the nature of complex numbers. Optimization on the complex field requires alternate partial derivative definitions to circumvent consequences of the Cauchy–Riemann equations regarding holomorphic functions. To make use of phase information, we demonstrate classifier model optimization with complex-valued parameters on data with both magnitude and phase components and show how complex neural networks yield marked improvements over similarly shaped real-valued networks in both classification accuracy and generalization ability. We apply these techniques to small target SONAR with simulated Lamb wave resonance signals for hollow spheres, differentiating between different material classes. [Research funded by the DoD Navy (NEEC) Grant No. N001742010016.]

Bionic study of distance-azimuth discrimination of multi-scattered point objects in bat bio-sonar

This paper presents a novel approach to enhance the discrimination capacity of multi-scattered point objects in bat bio-sonar. A broadband interferometer mathematical model is developed, incorporating both distance and azimuth information, to simulate the transmitted and received signals of bats. The Fourier transform is employed to simulate the preprocessing step of bat information for feature extraction. Furthermore, the bat bio-sonar model based on convolutional neural network (BS-CNN) is constructed to compensate for the limitations of conventional machine learning and CNN networks, including three strategies: Mix-up data enhancement, joint feature and hybrid atrous convolution module. The proposed BS-CNN model emulates the perceptual nerves of the bat brain for distance-azimuth discrimination and compares with four conventional classifiers to assess its discrimination efficacy. Experimental results demonstrate that the overall discrimination accuracy of the BS-CNN model is 93.4%, surpassing conventional CNN networks and machine learning methods by at least 5.9%. This improvement validates the efficacy of the BS-CNN bionic model in enhancing the discrimination accuracy in bat bio-sonar and offers valuable references for radar and sonar target classification.

Simulation-based training of neural networks for low-frequency sonar target classification

Artificial Intelligence (AI) is widely used in various fields, but obtaining sufficient underwater acoustic data for AI training remains challenging. Even if simulation data are used for training, the classification accuracy could be lowered due to changes in the environment around the target. This study presents the research findings on training neural networks for target classification using data generated through simulations. We trained neural networks using target scattering signals obtained through finite element analysis for various environments. Subsequently, we utilized trained neural networks to classify the spherical shells on the seabed acquired in water tank experiments. To solve the problem of lack of data for AI training, we generated training data by applying realistic variations of parameters such as sound speed of water, target material properties, and thickness. Such training data provide more meaningful information than noise-added data. The targets considered were spherical shells with and without internal spaces. Although the target classification accuracy is not yet high enough, it is expected to be used to overcome the lack of data by applying to untrained targets. [Work supported by KIMST funded by the Agency of Korea Coast Guard (KIMST-20210547) and by KRISO funded by Ministry of Oceans and Fisheries (PES4380).]

Training RBF NN Using Sine-Cosine Algorithm for Sonar Target Classification

Radial basis function neural networks (RBF NNs) are one of the most useful tools in the classification of the sonar targets. Despite many abilities of RBF NNs, low accuracy in classification, entrapment in local minima, and slow convergence rate are disadvantages of these networks. In order to overcome these issues, the sine-cosine algorithm (SCA) has been used to train RBF NNs in this work. To evaluate the designed classifier, two benchmark underwater sonar classification problems were used. Also, an experimental underwater target classification was developed to practically evaluate the merits of the RBFbased classifier in dealing with high-dimensional real world problems. In order to have a comprehensive evaluation, the classifier is compared with the gradient descent (GD), gravitational search algorithm (GSA), genetic algorithm (GA), and Kalman filter (KF) algorithms in terms of entrapment in local minima, the accuracy of the classification, and the convergence rate. The results show that the proposed classifier provides a better performance than other compared classifiers as it classifies the sonar datasets 2.72% better than the best benchmark classifier, on average.

Feature Extraction and Classification of Simulated Monostatic Acoustic Echoes from Spherical Targets of Various Materials Using Convolutional Neural Networks

Active sonar target classification remains an ongoing area of research due to the unique challenges associated with the problem (unknown target parameters, dynamic oceanic environment, different scattering mechanisms, etc.). Many feature extraction and classification techniques have been proposed, but there remains a need to relate and explain the classifier results in the physical domain. This work examines convolutional neural networks trained on simulated data with a known ground truth projected onto two time-frequency representations (spectrograms and scalograms). The classifiers were trained to discriminate the target material type, geometry, and internal fluid filling, while the hyperparameters were tuned to the classification task using Bayesian optimization. The trained networks were examined using an explainable artificial intelligence technique, gradient-weighted class activation mapping, to uncover the informative features used in discrimination. This analysis resulted in visual representations that allowed the CNN choices to be related to the physical domain. It was found that the scalogram representation provided a negligible classification accuracy increase compared with the spectrograms. Networks trained to discriminate between target geometries resulted in the highest accuracy, and the networks trained to discriminate the internal fluid of the target resulted in the lowest accuracy.

A multi-order Fractional Fourier domain feature union method for active sonar target classification

Different order domains in Fractional Fourier Transform (FrFT) reflect the features of different time–frequency planes. They contain multi-domain information about the target. The optimal order domain feature is used widely because of its anti-reverberation interference performance, but other order domains also have anti-reverberation ability, and even have more separable performance. It is of great significance to use effectively the multi-order domain feature for target classification and recognition. Therefore, a multi-order Fractional Fourier domain feature union method based on the information entropy is proposed in this paper. This method aims to find more distinguishable and separable target features than that of the optimal order domain by calculating the information entropies of the amplitude features and union them by determining the weights according to the value of the information entropies. The performance of the presented method is verified using the field data of four types of similar targets’ echo signals when the signal-to-reverberation ratios (SRRs) are respectively 4 dB, 0 dB, −3 dB, and −6 dB. Experimental results show that the method proposed in this paper can reflect the overall characteristics of the signals and has a more distinguishable ability and a good anti-reverberation ability.

Canonical correlation analysis as a feature extraction method to classify active sonar targets with shallow neural networks.

Sonar target recognition remains an active area of research due to the complex entanglement of features from various acoustic scatterers, background clutter, and distortion by waveguide propagation effects. An equally challenging issue is due to different acoustic echoes returned from the target (including different target elements) itself. This work investigates the sonar target classification problem from a statistical perspective and aims to extract salient target feature vectors. Specifically, a multivariate statistical method is employed, canonical correlation analysis (CCA), as a feature extraction technique prior to multi-class classification of active sonar field data. The intuition behind using CCA is that persistent features slowly morph over time due to the changing aspect angles and platform positions and can be represented by maximally correlated projections of consecutive pings. CCA is applied using a sliding window, and the projections are used as feature vectors to train a neural network classifier. The smallest increase in classification accuracy when comparing the projection feature vectors to unprocessed feature vectors was 10%. The largest increase was 34%. The results are further examined through the use of confusion matrices and layer-wise relevance propagation, which distributes the trained networks output score to the input layer.

Spike-Based Approximate Backpropagation Algorithm of Brain-Inspired Deep SNN for Sonar Target Classification.

With the development of neuromorphic computing, more and more attention has been paid to a brain-inspired spiking neural network (SNN) because of its ultralow energy consumption and high-performance spatiotemporal information processing. Due to the discontinuity of the spiking neuronal activation function, it is still a difficult problem to train brain-inspired deep SNN directly, so SNN has not yet shown performance comparable to that of an artificial neural network. For this reason, the spike-based approximate backpropagation (SABP) algorithm and a general brain-inspired SNN framework are proposed in this paper. The combination of the two can be used for end-to-end direct training of brain-inspired deep SNN. Experiments show that compared with other spike-based methods of directly training SNN, the classification accuracy of this method is close to the best results on MNIST and CIFAR-10 datasets and achieves the best classification accuracy on sonar image target classification (SITC) of small sample datasets. Further analysis shows that compared with artificial neural networks, our brain-inspired SNN has great advantages in computational complexity and energy consumption in sonar target classification.

Active target classification using a shallow neural network with dimension reduction

Techniques for automatically identifying sonar targets using machine learning algorithms are being actively developed, and modern algorithms based on deep learning are often considered first. However, deep learning based algorithms require a lot of data to train a neural network, and if there is not enough data, overfitting easily occurs and the performance on real data can be degraded. In fact, in the case of active target detection where there is not enough data, it has been reported that using a shallow neural network after extracting appropriately designed features from the raw data showed better detection performance than applying deep learning directly to the raw data. With regard to this, we investigate the performance of a shallow neural network combined with dimensionality reduction techniques for active sonar target classification. In particular, several linear and nonlinear dimension reduction techniques are compared in terms of target classification performance, and the effects of the characteristics of background noise and the presence of reverberation on the target classification performance are discussed. [This work was supported by the research project PES4380 funded by KRISO.]

Sonar data classification by using few-shot learning and concept extraction

Given the heterogeneity and difficulty of classifying sonar targets, classifying these targets is a challenging problem in practice. In this paper, we propose a novel classification of sonar targets based on few-shot learning (FSL). The FSL can classify the data with only a few labeled training data. We also use the wavelet transform to denoise sonar signals and apply the short-time Fourier transform (STFT) to denoised signals. We propose a concept extraction method based on the STFT and find concept points to improve the performance of FSL classification. We evaluate the impact of concept numbers on accuracy. The main distinguishing feature of our proposed technique is the low sonar data requirement compared to other classification techniques. Our numerical experiments show that the proposed classification technique using the concept extraction method significantly improves the system performance in terms of accuracy.

A novel sonar target detection and classification algorithm

A novel sonar target detection and classification algorithm

Active Sonar Target Classification Method Based on Fisher’s Dictionary Learning

The marine environment is complex and changeable, and the interference of noise and reverberation seriously affects the classification performance of active sonar equipment. In particular, when the targets to be measured have similar characteristics, underwater classification becomes more complex. Therefore, a strong, recognizable algorithm needs to be developed that can handle similar feature targets in a reverberation environment. This paper combines Fisher’s discriminant criterion and a dictionary-learning-based sparse representation classification algorithm, and proposes an active sonar target classification method based on Fisher discriminant dictionary learning (FDDL). Based on the learning dictionaries, the proposed method introduces the Fisher restriction criterion to limit the sparse coefficients, thereby obtaining a more discriminating dictionary; finally, it distinguishes the category according to the reconstruction errors of the reconstructed signal and the signal to be measured. The classification performance is compared with the existing methods, such as SVM (Support Vector Machine), SRC (Sparse Representation Based Classification), D-KSVD (Discriminative K-Singular Value Decomposition), and LC-KSVD (label-consistent K-SVD), and the experimental results show that FDDL has a better classification performance than the existing classification methods.

Active sonar target classification using a physics-cognizant feature representation

Active sonar target classification is challenging due to the non-linear overlap of changing oceanic and target parameters, creating entangled acoustic color spectra that should be disentangled prior to classification. A physics-cognizant feature extraction algorithm, used before interfacing with three machine learning techniques for active sonar target classification of experimental field data, is presented. The feature extraction algorithm convolves a two-dimensional Gabor wavelet across acoustic color spectra prior to threshold-based binarization, feature culling, and dimensional reduction. The optimal two-dimensional Gabor wavelet parameters are chosen through sensitivity analysis by a support vector machine (SVM) on a disjoint subset of data. Classification is performed on the second subset of data with an SVM, random forest tree, and neural network on the Gabor filtered spectra and unfiltered spectra to show the increased classification accuracy of the application of the geometric wavelet. Classification results are presented as confusion matrices for four targets of two public domain experiments. Ongoing and future work will include extending this feature extraction technique using various geometric feature representations to capture and describe far-field large scattering mechanisms from targets such as oil rigs, tankers, and shipwrecks. [This research is funded by the Office of Naval Research under Grant No. N00014-19-1-2436.]

Sparse Representation Classification Method for Active Sonar Target Based on Multi-order Fractional Fourier Domain Feature Fusion

Marine environment noise and reverberation interference are serious, and the poor target separability is the bottleneck problem in active sonar target classification and recognition. In order to solve this problem, based on the echo signal model of active sonar target and the principle of FRactional Fourier Transform (FRFT), this paper deduces the multi-order FRFT domain feature representation form, establishes the FRFT domain sparse representation model, and proposes a method to classify the sparse representation of active sonar targets with multi-order FRFT domain feature fusion. The method achieves the purpose of suppressing noise and reverberation interference through the energy aggregation of FRFT and removing the residual of sparse decomposition; Through the fusion of multi-order FRFT domain features, the separability of targets is improved, and the active sonar target classification with low SNR is realized. Experimental results show that the classification accuracy of the proposed method can reach more than 90% when the SNR is about 0 dB.

Active Sonar Target Classification with Power-Normalized Cepstral Coefficients and Convolutional Neural Network

Detection and classification of unidentified underwater targets maneuvering in complex underwater environments are critical for active sonar systems. In previous studies, many detection methods were applied to separate targets from the clutter using signals that exceed a preset threshold determined by the sonar console operator. This is because the high signal-to-noise ratio target has enough feature vector components to separate. However, in a real environment, the signal-to-noise ratio of the received target does not always exceed the threshold. Therefore, a target detection algorithm for various target signal-to-noise ratio environments is required; strong clutter energy can lead to false detection, while weak target signals reduce the probability of detection. It also uses long pulse repetition intervals for long-range detection and high ambient noise, requiring classification processing for each ping without accumulating pings. In this study, a target classification algorithm is proposed that can be applied to signals in real underwater environments above the noise level without a threshold set by the sonar console operator, and the classification performance of the algorithm is verified. The active sonar for long-range target detection has low-resolution data; thus, feature vector extraction algorithms are required. Feature vectors are extracted from the experimental data using Power-Normalized Cepstral Coefficients for target classification. Feature vectors are also extracted with Mel-Frequency Cepstral Coefficients and compared with the proposed algorithm. A convolutional neural network was employed as the classifier. In addition, the proposed algorithm is to be compared with the result of target classification using a spectrogram and convolutional neural network. Experimental data were obtained using a hull-mounted active sonar system operating on a Korean naval ship in the East Sea of South Korea and a real maneuvering underwater target. From the experimental data with 29 pings, we extracted 361 target and 3351 clutter data. It is difficult to collect real underwater target data from the real sea environment. Therefore, the number of target data was increased using the data augmentation technique. Eighty percent of the data was used for training and the rest was used for testing. Accuracy value curves and classification rate tables are presented for performance analysis and discussion. Results showed that the proposed algorithm has a higher classification rate than Mel-Frequency Cepstral Coefficients without affecting the target classification by the signal level. Additionally, the obtained results showed that target classification is possible within one ping data without any ping accumulation.

RS-HeRR: a rough set-based Hebbian rule reduction neuro-fuzzy system

Interpretabilty is one of the desired characteristics in various classification task. Rule-based system and fuzzy logic can be used for interpretation in classification. The main drawback of rule-based system is that it may contain large complex rules for classification and sometimes it becomes very difficult in interpretation. Rule reduction is also difficult for various reasons. Removing important rules may effect in classification accuracy. This paper proposes a hybrid fuzzy-rough set approach named RS-HeRR for the generation of effective, interpretable and compact rule set. It combines a powerful rule generation and reduction fuzzy system, called Hebbian-based rule reduction algorithm (HeRR) and a novel rough-set-based attribute selection algorithm for rule reduction. The proposed hybridization leverages upon rule reduction through reduction in partial dependency as well as improvement in system performance to significantly reduce the problem of redundancy in HeRR, even while providing similar or better accuracy. RS-HeRR demonstrates these characteristics repeatedly over four diverse practical classification problems, such as diabetes identification, urban water treatment monitoring, sonar target classification, and detection of ovarian cancer. It also demonstrates excellent performance for highly biased datasets. In addition, it competes very well with established non-fuzzy classifiers and outperforms state-of-the-art methods that use rough sets for rule reduction in fuzzy systems.

Classification of Sonar Targets Using an MLP Neural Network Trained by Dragonfly Algorithm

Due to the compatibility of the designed classifiers with MLP Neural Networks (MLP NNs), in this article, MLP NNs have been used to identify and classify active and passive sonar targets. On the one hand, the great importance of precise and immediate classification of sonar targets, and on the other hand, being trapped in local minimums and the low convergence speed in classic MLP NNs have led the newly proposed Dragonfly Algorithm (DA) to be offered for training MLP NNs. In order to assess the performance of the designed classifier, this algorithm have been compared with BBO, GWO, ALO, ACO, GSA and MVO algorithms in terms of precision of classification, convergence speed and the ability to avoid local optimum. To have a comprehensive comparison, the three sets of active and passive data were used. Simulation results indicate that DA-based classification have better results in all three datasets compared to benchmark algorithms.

Passive sonar target classification using multi-layer perceptron trained by salp swarm algorithm

Due to the variability of the radiated signal of the sonar targets, passive sonar target classification is a challenging problem in the real world application. Adaptive Mel-frequency cepstral coefficients (MFCCs) and multi-layer perceptron (MLP) approaches are proposed, including cepstral features to alleviate dataset's dimension and MLP network to adapt the variability in changing condition. In spite of the capabilities of MLP networks, low classification accuracy, and getting stuck in local minima are the main shortcomings of MLP networks. To overcome these shortcomings, this paper proposes the use of the newly introduced salp swarm algorithm (SSA) for training MLP network. In order to investigate the efficiency of the proposed classifier, four high-dimensional benchmark functions, as well as an experimental passive sonar data set, are employed. The designed classifier is compared to gray wolf optimizer (GWO), biogeography-based optimization (BBO), interior search algorithm (ISA), and group method of data handling (GMDH) in terms of classification accuracy, entrapment in local minima, and convergence speed. The results showed that the proposed classifier is more efficient than the other benchmark algorithms; therefore, the SSA classifies sonar data set as much as 0.9017 percent better than GMDH being the best results among the other classifiers.