Classification Of Passive Sonar Signals Research Articles

Feature Extraction in Fractional Fourier Domain for Classification of Passive Sonar Signals

The acoustic signals radiated from the marine vessels contain information about their machinery characteristics that can be useful for the detection and classification purposes. To achieve reliable accuracy in classification task, informative discriminant features should be extracted from received signals. STFT (Short Time Fourier Transform) is the most basic and popular signal processing method in the classification process of passive sonar signals, but it suffers from some fatal shortages. In this work, we present an improved spectrogram based on the windowed fractional-Fourier transform of the acoustic signal with the optimal FrFT order, which accounts for signals with multiple non-stationary components. The discriminating capability of two groups of features extracted from the processed signal, using PCA and LDA techniques, have been compared. The achieved results declare the significant improvement in the classification accuracy by the new proposed method using the LDA feature extraction technique and the proper order of FrFT. However, in order to eliminate the constraints of searching for the proper FrFT-order and increasing the reliability and stability of practical sonar classification systems, a parallel combination of the proposed method is introduced.

Class‐modular multi‐layer perceptron networks for supporting passive sonar signal classification

The automatic classification of passive sonar signals in real time is an extremely useful tool for supporting sonar operators decision making during submarine missions. An appealing architecture for the development of accurate, flexible, modular and real-time classification systems is the class-modular multi-layer perceptron (CM-MLP). However, for this task, several design parameters have to be tuned and a more extensive guideline about how to conduct this process is missing in the literature. This work aims at discussing the main design phases related to the development of CM-MLP systems for passive sonar signal classification. For each phase, several approaches are discussed and their cost-effectiveness is experimentally evaluated using statistical significance tests. This analysis is based on real signals produced by 8 vessel classes and acquired by an omnidirectional hydrophone in a shallow water environment, during 263 experimental runs of 34 ships. Results show that some design parameters significantly affect the accuracy of the classification system, especially the complexity of each module. An accuracy of 84.4% is achieved by using tuned design parameters.

Source depth classification of passive sonar signals using amplitude statistics.

A number of at-sea measurements have shown that fluctuations in the amplitude of passive sonar signals are affected by source depth, range, and the propagation path from the source to the receiver. However, to date, received signal statistics have not been incorporated into passive sonar classification algorithms because the field results have not been accompanied by development of a statistics-based classifier architecture whose performance can be assessed using accepted signal processing metrics, e.g., a receiver operating characteristic curve. The performance of a statistics-based classifier depends critically upon statistical knowledge of the received signal, which in turn depends upon statistical knowledge of relevant environment parameters (sound speed in the water column and sediment, for example). Thus the classifier architecture must make use of the environmental information which in reality is only approximately known and must be described in statistical terms. Given a statistical description of the environment, an ocean acoustic propagation model and Monte Carlo simulation can be used to predict received signal statistics. Once received signal statistics are available for the different classes, a classifier can be designed. Here we examine a likelihood ratio binary source depth classifier for passive sonar. [Work sponsored by ONR Undersea Signal Processing.]

Auditory-motivated techniques for detection and classification of passive sonar signals

Hydrophone arrays used to make passive sonar recordings are becoming increasingly complex, thus placing great demands on sonar operators who are required to identify transient and tonal events. Machine hearing systems can mitigate this problem, by performing initial classification and detection of passive sonar signals based on perceptually relevant principles. We describe two such systems. In the first system, transient sounds are characterised using measures of timbre and acoustic pattern. Acoustic features relating to timbre were identified by a multidimensional scaling study, in which human listeners judged the similarity of transient sonar sounds. The perceptual features give improved classification performance compared to commonly-used statistical measures derived from the power spectrum. The second system is an auditory-motivated approach to narrowband detection, which is based on information in zero-crossing intervals rather than signal power. This is motivated by the observation that the fine time structure of an acoustic stimulus is preserved by the phase locking of auditory nerve fibres. We show that a power detector outperforms the interval detector when the signal is centred on the analysis filter. However, when the signal frequency deviates from the centre of the analysis band, interval-based detection improves and power-based detection worsens.