Improving classification performance of sonar targets by applying general regression neural network with PCA

doi:10.1016/j.eswa.2007.07.021

Expert Systems with Applications

Volume 35, Issues 1–2, July–August 2008, Pages 472-475

https://doi.org/10.1016/j.eswa.2007.07.021 Get rights and content

Abstract

The remote detection of undersea mines in shallow waters using active sonar is a crucial subject required to maintain the security of important harbors and cost line areas. The discrimination sonar returns from mines and returns from rocks on the sea floor by human experts is usually difficult and very heavy workload. Neural network classifiers have been widely used in classification of complex sonar signals due to its adaptive and parallel processing ability. In this paper, due to the advantages on fast learning and convergence to the optimal regression surface as the number of samples becomes very large, general regression neural network (GRNN) has been used to solve the problem of classification underwater targets. Principal component analysis (PCA) has been established as a feature extraction method to improve classification performance. Receiver operating characteristic (ROC) analysis has been applied to the neural classifier to evaluate the sensitivity and specificity of diagnostic procedures.

Introduction

Automatic identification and classification of underwater signals on the basis of sonar signals are challenging problem due to the complexity of the ocean environment. Identification by human experts is usually not objective and a very heavy workload. Neural networks with their adaptive and computational advantages appear to be ideally suited to active sonar classification. The pioneer paper by Gorman and Sejnowski, 1988a, Gorman and Sejnowski, 1988b were perhaps the first papers which reported the application of neural networks to this area. They used simple spectral features as the input to the neural network classifier in order to distinguish a mine from cylindrical shaped rock positioned on a sandy ocean floor. After these pioneer papers, there has been growing interest in the use of neural networks for the automatic recognition of sonar targets. Among the several neural classifiers, multi-layer perceptron (MLP) classifier has been used in several applications about sonar target classification (Gorman and Sejnowski, 1988a, Gorman and Sejnowski, 1988b, Jing and El-Hawary, 1994, Shazeer and Bello, 1991). Radial basis function networks (RBFN) (Chen, 1992, Yegnanarayana et al., 1992), probabilistic neural networks (PNN) (Chen, 1992), general regression neural networks (Kapanoğlu & Yıldırım, 2004), and conic section function neural networks (CSFNN) (Erkmen & Yıldırım, 2006) have been also the efficient feed-forward neural networks widely used to classify sonar signal in literature.

Feature extraction is the important preprocessing step to classification of complex signal such as vision, speech identification or the problem mentioned here of detecting objects in sonar returns. Input dimensionality of these problems becomes a serious drawback to classification. Feature extraction techniques such as PCA, and neural discriminating analysis (NDA) reduce a high dimensional signal to a lower dimensional feature set, which preserves the most useful and relevant information on the feature space. In (Larkin, 1997, Soares-Filho et al., 2001), several feature extraction techniques were used to preprocess given sonar signals.

Recently, general regression neural network (GRNN) is successfully used in pattern recognition and image processing due to the advantages on fast learning and convergence to the optimal regression surface as the number of samples becomes very large (Polat & Yildirim, 2006). In this paper, GRNN has been used to classify sonar returns from two different targets on the sandy ocean bottom a mine and cylindrical shaped rock. The emphasis here is not the classifier itself, but also the process of the feature extraction technique, PCA, improves classifier performance. In this paper, the receiver operation characteristics (ROC) analysis (Woods & Bowyer, 1994) has been also applied to the neural classifier to show the accuracy of classification. ROC analysis is an established method of measuring diagnostic performance in sonar studies.

This paper organized as follows. Section 2 briefly describes the GRNN structure. PCA technique is examined in Section 3. Section 4 presents some concepts related to ROC analysis. In Section 5, the neural classifier designs, simulation results and performance evaluation with ROC analysis are presented. Finally, Section 6 outlines some conclusions.

Section snippets

General regression neural networks (GRNN)

General regression neural networks (GRNN) are memory-based network that provides estimates of continuous variables and converges to the underlying (linear or nonlinear) regression surface. GRNN is a one pass learning algorithm with highly parallel structure. It does not require an iterative training procedure as in MLP. The principal advantages of GRNN are fast learning and convergence to the optimal regression surface as the number of samples becomes very large. GRNNs have been shown to

Traditional principal component analysis

Principal component analysis is a multivariable statistical analysis technique of data compression and feature extraction. Envisaging practical applications that require online operations for the classifiers, an attractive approach for the classifier design is to reduce the dimensionality of the N-dimensional input space by projecting input data onto a reduced a number of M directions (M ≪ N) that can facilitate the classification task. The PCA describes the original data space in a base of

Receiver operating characteristic (ROC) analysis

ROC analysis is an established method of measuring diagnostic performance for the analysis of radar images. The ROC curve is a good measure when performance of different schemes needs to be compared. The evaluation criteria are based on the ROC curve, used in sonar target classification system to indicate trade-off of the conditional probability of correct classification versus conditional probability of false-alarm responses. Equivalently, the ROC curve is a graphical representation of the

Simulation results

The aim of this study is to employ general regression neural networks to classify sonar returns. To improve classification performance and simplify network complexity, PCA is used as feature extraction method. The data set, which is the original sonar data used by Gorman and Sejnowski, 1988a, Gorman and Sejnowski, 1988b was taken from the University of California collection of machine-learning-databases. Although this data set is an old one, it is a well-known benchmark for sonar problems. This

Conclusion

In this paper, a GRNN has been used to classify underwater target collected from two sources: a metal cylinder and similarly shaped rock. The emphasis in here is not the classifier itself, but also the process of the feature extraction technique, PCA. When the performance of GRNN using with PCA are compared with the performance of GRNN using without PCA in terms of successful classification rates, it is realized that PCA improves the neural classifier performance significantly. The low

References (15)

R.P. Gorman et al.
Analysis of hidden units in a layered network trained to classify sonar targets
Neural Networks
(1988)
Chen, C. H. (1992). Neural networks for active sonar classification. Pattern recognition, vol. II. conference B:...
B. Erkmen et al.
Conic section function neural networks for sonar target classification and performance evaluation using ROC analysis
R.P. Gorman et al.
Learned classification of sonar targets using a massively parallel network
IEEE Transactions on Acoustics, Speech, and Signal Processing
(1988)
Jing, Y., & El-Hawary, F. (1994). A multilayered ANN architecture for underwater target tracking. Electrical and...
Kapanoğlu, B., & Yıldırım T., (2004). Generalized regression neural networks for underwater target classification....
Kusy, M., & Zajdel, R. (2005). Comparison of two radial function based models: support vector machine and neural...

There are more references available in the full text version of this article.

Cited by (74)

Unraveling complex data diversity in underwater acoustic target recognition through convolution-based mixture of experts
2024, Expert Systems with Applications
Underwater acoustic target recognition is a difficult task owing to the intricate nature of underwater acoustic signals. The complex underwater environments, unpredictable transmission channels, and dynamic motion states greatly impact the real-world underwater acoustic signals, and may even obscure the intrinsic characteristics related to targets. Consequently, the data distribution of underwater acoustic signals exhibits high intra-class diversity, thereby compromising the accuracy and robustness of recognition systems. To address these issues, this work proposes a convolution-based mixture of experts (CMoE) that recognizes underwater targets in a fine-grained manner. The proposed technique introduces multiple expert layers as independent learners, along with a routing layer that determines the assignment of experts according to the characteristics of inputs. This design allows the model to utilize independent parameter spaces, facilitating the learning of complex underwater signals with high intra-class diversity. Furthermore, this work optimizes the CMoE structure by balancing regularization and an optional residual module. To validate the efficacy of our proposed techniques, we conducted detailed experiments and visualization analyses on three underwater acoustic databases across several acoustic features. The experimental results demonstrate that our CMoE consistently achieves significant performance improvements, delivering superior recognition accuracy when compared to existing advanced methods.
Open-set recognition with long-tail sonar images
2024, Expert Systems with Applications
Current sonar image recognition methods excel in closed-set and balanced scenarios, but real underwater data often follow an open-set and long-tailed distribution, leading to misclassifications, especially among tail classes. Although open-set long-tail recognition (OLTR) tasks have received attention in natural images in recent years, there has been a lack of systematic research in sonar images. To address this gap, we present the first comprehensive study and analysis of open-set long-tail recognition in sonar images (Sonar-OLTR). In this paper, we establish a Sonar-OLTR benchmark by introducing the Nankai Sonar Image Dataset (NKSID), a new collection of 2617 real-world forward-looking sonar images. We investigate the challenges posed by long-tail distributions in existing open-set recognition (OSR) evaluation metrics for sonar images and propose two improved evaluation metrics. Using this benchmark, we conduct a thorough examination of state-of-the-art OSR, long-tail recognition, OLTR, and out-of-distribution detection algorithms. Additionally, we propose a straightforward yet effective integrated Sonar-OLTR approach as a new baseline. This method introduces a Push the right Logit Up and the wrong logit Down (PLUD) loss to increase feature space margins between known and unknown classes, as well as head and tail classes within known classes. Extensive experimental evaluation based on the benchmark demonstrates the performance and speed advantages of PLUD, providing insights for future Sonar-OLTR research. The code and dataset are publicly available at https://github.com/Jorwnpay/Sonar-OLTR.
Fault detection and isolation of gas turbine using series–parallel NARX model
2022, ISA Transactions
This paper describes the design and implementation of intelligent dynamic models for fault detection and isolation of V94.2(5)/MGT-70(2) single-axis heavy-duty gas turbine system. The series–parallel structure of nonlinear autoregressive exogenous (NARX) models are used for fault detection, which initiate greater robustness and stability against uncertainties and perturbations. Moreover, to improve the fault detection robustness against uncertainties, the Monte Carlo technique is used in the proposed fault detection structure to select the best threshold. The analysis of fault detectability and fault detection sensitivity are accomplished to analyze the performance of the suggested technique. The fault isolation process is also achieved by using the residual classification approach. The results show the feasibly, robustness, and performance of the presented approach for fault diagnosis of nonlinear systems in the presence of uncertainties.
Estimation of heavy metals using deep neural network with visible and infrared spectroscopy of soil
2020, Science of the Total Environment
Heavy metal contamination in soil disturbs the chemical, biological, and physical soil conditions and adversely affects the health of living organisms. Visible and near-infrared spectroscopy (VNIRS) shows a potential feasibility for estimating heavy metal elements in soil. Moreover, deep learning models have been shown to successfully deal with complex multi-dimensional and multivariate nonlinear data. Thus, this study implemented a deep learning method on reflectance spectra of soil samples to estimate heavy metal concentrations. A convolutional neural network (CNN) was adopted to estimate arsenic (As), copper (Cu), and lead (Pb) concentrations using measured soil reflectance. In addition, a convolutional autoencoder was utilized as a joint method with the CNN for dimensionality reduction of the reflectance spectra. Furthermore, artificial neural network (ANN) and random forest regression (RFR) models were built for heavy metal estimation. Principal component analysis was utilized for dimensionality reduction of the ANN and RFR models. Among these models, the CNN model with convolutional autoencoder showed the highest accuracies for As, Cu, and Pb estimates, having R² values of 0.86, 0.74, and 0.82, respectively. The convolutional autoencoder disentangled the relevant features of multiple heavy metal elements and delivered robust features to the CNN model, thereby generating relatively accurate estimates.
Underwater sonar image classification using adaptive weights convolutional neural network
2019, Applied Acoustics
As an important part of oceanographic surveying, underwater sonar image classification has attracted much attention. Most of the existing classification methods cannot be widely used in underwater sonar image classification. However, deep learning models can automatically extract underwater sonar image features to improve the classification accuracy through an internal network structure. In the present study, a novel deep learning model with adaptive weights convolutional neural network (AW-CNN) was proposed to classify underwater sonar images. To solve the random initialization of filter weights in a convolutional neural network (CNN), the generated weights of the deep belief network (DBN) were applied to adaptively replace the randomly trained filter weights of the CNN in the implementation process of the AW-CNN for underwater sonar image classification. Specifically, first, dimension conversion was accomplished by using the increment-dimension function to unify the inputs of the CNN and the DBN. Then, according to the dimension conversion, the internal fusion of the two models is realized, and replacement of the randomly trained filter weights of the CNN was completed. Finally, in order to further improve the classification accuracy, the local response normalization (LRN) function is proposed to normalize the adaptive weights in the network initialization. Compared with other models, the classification results demonstrate that the proposed AW-CNN approach has the capability to effectively and successfully divide sonar images into their relevant seabed classes, which is beneficial to finding mines and shoals and detecting the integrated degree of the dam bottom.
A Novel Subset Feature Selection Framework for Increasing the Classification Performance of SONAR Targets
2018, Procedia Computer Science
Proposing a strong subset of feature for a classifier to detect the SONAR (sound navigation and ranging) target is the most valuable subject in the sonar data analysis for the safety of naval vessels. In this research article, we introduced a novel generalized feature selection framework to determine the best subset of features for classifying sonar data set based on Symmetrical Uncertainty(SU). Using the proposed framework, we tried to form ‘M’ candidate subsets of features. Each set comprises of a finite number of features without any ingeminate. The resultant subset is analyzed using various Tree, Lazy, Bayesian, and Rule based classifiers. As each subset formed by proposed method consists of limited number of features in it, an equal number of top features extracted by existing filter based feature selection methods( Chi Squared -Chi, Information Gain(IG), Gain Ratio(GR), and ReliefF(Rel)) are taken into consideration, and analyzed with the same classifiers. After careful investigation of obtained results by existing and proposed methods, it has been noticed that minimum one of the candidate subset of features outperforming than some of the existing methods. For this experiment, a real time SONAR dataset found at UCI machine learning repository is considered. This framework also applied on other benchmark datasets to make it generalized.

View all citing articles on Scopus

View full text

Improving classification performance of sonar targets by applying general regression neural network with PCA

Abstract

Introduction

Section snippets

General regression neural networks (GRNN)

Traditional principal component analysis

Receiver operating characteristic (ROC) analysis

Simulation results

Conclusion

Neural Networks

Conic section function neural networks for sonar target classification and performance evaluation using ROC analysis

Learned classification of sonar targets using a massively parallel network

IEEE Transactions on Acoustics, Speech, and Signal Processing