Passive Ship Localization in a Shallow Water Using Pre-trained Deep Learning Networks

Subject to the lack of detailed environmental information, the classical matched-field processing (MFP) may not be adapted to the accurate localization of underwater acoustic sources. In this paper, a framework that applies deep learning techniques instead of the MFP method is presented for the localization (direction-finding) of ship acoustic sources in a shallow water environment. The original data is recorded from a 128-element vertical array placed in a shallow water. The acquired array data is first processed by the time-domain conventional beamformer (CBF) in order to obtain the beamformed waveform signals corresponding to each direction-of-arrival (DOA) with a resolution of 1 degree. In the meantime, the GPS and recognized DOA information in the diagram of the target ships are employed to generate the labels for these beamformed signals. Base on the labeled data, a framework is proposed to predict the DOA information of target ships in a deep learning (DL) manner using the pre-trained state-of-the-art convolutional neural networks. Driven directly by the array signal data, the proposed method offers a way for the ship localization to overcome the environmental mismatch problem, which is believed to be better than that of conventional MFP method and some other shallow machine learning methods.