An Underwater Acoustic Target Recognition Method Based on Combined Feature With Automatic Coding and Reconstruction

Abstract

Underwater acoustic target recognition is one of the main functions of the SONAR systems. In this paper, a target recognition method based on combined features with automatic coding and reconstruction is proposed to classify ship radiated noise signals. In the existing underwater acoustic target recognition systems, the target category features are mostly constructed based on the power spectrum according to a certain presupposed model, and some useful information in the data is discarded artificially. In the proposed recognition method, a feature extractor based on auto-encoding is designed. The feature extractor uses the restricted Boltzmann machine (RBM) to automatically encode the combined data of the power spectrum and demodulation spectrum of ship radiated noise without supervision and extracts the deep data structure layer by layer to obtain the signal feature vector. The extracted feature vector is sent to a Back Propagation (BP) neural network to realize target recognition. Due to the high cost of ship radiated noise acquisition, the sample size of ship radiated noise signals is often hard to meet the needs of neural network training. A method of data augmentation is designed by RBM auto-encoder to construct the expanded sample set, which improves the performance of the recognition system. The experimental results based on the actual ship's radiated noise show that the proposed method has better performance than the traditional methods.

A Survey of Underwater Acoustic Data Classification Methods Using Deep Learning for Shoreline Surveillance

“…A data-augmentation technique was used in [ 123 ] to double the size of the dataset used to create the ShipsEar dataset ( Section 3.4 ). In this augmentation procedure, the original data were divided into five categories, with 600 samples each, then an RBM autoencoder was applied to reconstruct the original signals.…”

“…The scarcity of underwater acoustic data for training ML classification algorithms is a common statement in most related publications. In order to address this issue, Boltzmann Machine (RBM) autoencoders have been proposed in [ 123 ], aiming at reconstructing the original data to be used in the classification of vessels in passive sonar data. NN models that include RBM as a denoising element have been shown to greatly outperform the traditional Gaussian Mixture Model (GMM) with a Gammatone Frequency Cepstral Coefficient feature-extraction layer.…”

Present status and challenges of underwater acoustic target recognition technology: A review

“…Luo et al [37][38][39] used the normalized spectrum of the signal as input and completed the space-frequency joint detection of the line spectrum of underwater acoustic signals using a restricted Boltzmann machine. Yang et al [40] proposed a new cooperative deep learning method for underwater acoustic target recognition by combining deep long-and short-term memory networks and deep self-coding neural networks.…”

An Underwater Acoustic Target Recognition Method Based on Spectrograms with Different Resolutions

“…In Refs. [4,5], the power spectrum is used as the input of the classifier to achieve a good classification of ship targets. Auditory-based models are also used for feature extraction in underwater acoustic signals.…”

“…The method proposed in Ref. [5] is based on a DNN classifier and uses the combined features of the power spectrum and the DEMON spectrum, achieving a classification accuracy of 92.6%. The classifier proposed in Ref.…”