• The Journal of the Acoustical Society of Korea
• /
• v.42 no.4
• /
• pp.329-344
• /
• 2023

In this study, we comprehensively analyze the generalization performance of various deep learning-based active sonar target classifiers when applied to small and imbalanced active sonar datasets. To generate the active sonar datasets, we use data from two different oceanic experiments conducted at different times and ocean. Each sample in the active sonar datasets is a time-frequency domain image, which is extracted from audio signal of contact after the detection process. For the comprehensive analysis, we utilize 22 Convolutional Neural Networks (CNN) models. Two datasets are used as train/validation datasets and test datasets, alternatively. To calculate the variance in the output of the target classifiers, the train/validation/test datasets are repeated 10 times. Hyperparameters for training are optimized using Bayesian optimization. The results demonstrate that shallow CNN models show superior robustness and generalization performance compared to most of deep CNN models. The results from this paper can serve as a valuable reference for future research directions in deep learning-based active sonar target classification.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.5
• /
• pp.1083-1088
• /
• 2013

Detection and classification of undersea mines in shallow waters using active sonar returns is a difficult task due to complexity of underwater environment. Support vector machine(SVM) is a binary classifier that is well known to provide a global optimum solution. In this paper, classification experiments of sonar returns from mine-like objects and non-mine-like objects are carried out using the SVM, and classification performance is analyzed and presented with discussions depending on parameter values of SVM kernel functions.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.10
• /
• pp.2137-2142
• /
• 2012

Many studies in detection and classification of the targets in the underwater environments have been conducted for military purposes, as well as for non-military purpose. Due to the complicated characteristics of underwater acoustic signal reflecting multipath environments and spatio-temporal varying characteristics, active sonar target classification technique has been considered as a difficult technique. And it has a difficult in collecting actual underwater data. In this paper, we implemented the simulator to synthesize the active target signal, to extract feature and to classify the target in the underwater environment. In target signal synthesis, highlight and three-dimensional model are used and multi-aspect based hidden markov model is used for target classification.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.11
• /
• pp.2505-2511
• /
• 2013

Many studies in detection and classification of the targets in the underwater environments have been conducted for military purposes, as well as for non-military purpose. Due to the complicated characteristics of underwater acoustic signal reflecting multipath environments and spatio-temporal varying characteristics, active sonar target classification technique has been considered as a difficult technique. And it has difficulties in collecting actual underwater data. In this paper, we synthesized active target echoes based on ray tracing algorithm using target model having 3-dimensional highlight distribution. Then, Fractional Fourier transform was applied to synthesized target echoes to extract feature vector. Recognition experiment was performed using neural network classifier.

• Journal of IKEEE
• /
• v.22 no.4
• /
• pp.1044-1049
• /
• 2018

CNN (Convolutional Neural Networks) is a neural network that models animal visual information processing. And it shows good performance in various fields. In this paper, we use CNN to classify target and non-target data by analyzing the spectrogram of active sonar signal. The data were divided into 8 classes according to the ratios containing the targets and used for learning CNN. The spectrogram of the signal is divided into frames and used as inputs. As a result, it was possible to classify the target and non-target using the characteristic that the classification results of the seven classes corresponding to the target signal sequentially appear only at the position of the target signal.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.3
• /
• pp.418-424
• /
• 2018

Detection and classification of underwater targets is an important issue for both military and non-military purposes. Recently, many performance improvements are being reported in the field of pattern recognition with the development of deep learning technology. Among the results, DBN showed good performance when used for pre-training of DNN. In this paper, DBN was used for the classification of underwater targets using active sonar, and the results are compared with that of the conventional BPNN. We synthesized active sonar target signals using 3-dimensional highlight model. Then, features were extracted based on FrFT. In the single aspect based experiment, the classification result using DBN was improved about 3.83% compared with the BPNN. In the case of multi-aspect based experiment, a performance of 95% or more is obtained when the number of observation sequence exceeds three.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.4
• /
• pp.330-336
• /
• 2021

In bistatic sonar operation, the scattering strength of a sonar target is characterized by the probe signal frequency, the aspect angle and the bistatic angle. Therefore, the target detection and identification performance of the bistatic sonar may vary depending on how the positions of the target, sound source, and receiver are changed during sonar operation. In this study, it was evaluated which variable is advantageous to change by comparing the target identification performance between the case of changing the aspect angle and the case of changing the bistatic angle during the operation. A scenario of identifying a hollow sphere and a cylinder was assumed, and performance was compared by classifying two targets with a support vector machine and comparing their accuracy using a finite element method-based acoustic scattering simulation. As a result of comparison, using the scattering strength defined by the frequency and the bistatic angle with the aspect angle fixed showed superior average classification accuracy. It means that moving the receiver to change the bistatic angle is more effective than moving the sound source to change the aspect angle for target identification.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.1
• /
• pp.125-130
• /
• 2018

Linear discriminant analysis is a statistical analysis method that is generally used for dimensionality reduction of the feature vectors or for class classification. However, in the case of a data set that cannot be linearly separated, it is possible to make a linear separation by mapping a feature vector into a higher dimensional space using a nonlinear function. This method is called generalized discriminant analysis or kernel discriminant analysis. In this paper, we carried out target classification experiments with active sonar target signals available on the Internet using both liner discriminant and generalized discriminant analysis methods. Experimental results are analyzed and compared with discussions. For 104 test data, LDA method has shown correct recognition rate of 73.08%, however, GDA method achieved 95.19% that is also better than the conventional MLP or kernel-based SVM.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.1
• /
• pp.22-29
• /
• 2016

Many studies in detection and classification of the targets in the underwater environments have been conducted for military purposes, as well as for non-military purpose. Due to the complicated characteristics of underwater acoustic signal reflecting multipath environments and spatio-temporal varying characteristics, active sonar target detection technique has been considered as a difficult technique. In this paper, we describe the basic concept of Fractional Fourier transform and optimal transform order. Then we analyze the relationship between time-frequency characteristics of an LFM signal and its spectrum using Fractional Fourier transform. Based on the analysis results, we present active sonar target detection method. To verify the performance of proposed methods, we compared the results with conventional FFT-based matched filter. The experimental results demonstrate the superiority of the proposed method compared to the conventional method in the aspect of AUC(Area Under the ROC Curve).

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.10
• /
• pp.1909-1916
• /
• 2017

Recently, deep learning algorithms have good performance in various fields, but they are not actively applied to sonar systems. In this study, we carried out experiments to classify active sonar returns into a metal object such as a mine and a rock using a convolutional neural network which is one of the deep learning algorithms. Data augmentation is applied on this paper to avoid overfitting and increase performance. And we analyzed performance variation depending on hyperparameter value and change of the number of training data through data augmentation. The experiments are performed with two training data; an aspect-angle independent and an aspect-angle dependent. As a result, the performances are 88.9% and 94.9% in aspect-angle independent and dependent, respectively. These are up to 4.5% point higher than the performance obtained by applying artificial neural network and support vector machine algorithm in the previous study.