Showing 1–1 of 1 results for author: Landrø, M

Feature Selection based on Principal Component Analysis for Underwater Source Localization by Deep Learning

Authors: Xiaoyu Zhu, Hefeng Dong, Pierluigi Salvo Rossi, Martin Landrø

Abstract: In this paper, we propose an interpretable feature selection method based on principal component analysis (PCA) and principal component regression (PCR), which can extract important features for underwater source localization by only introducing the source location without other prior information. This feature selection method is combined with a two-step framework for underwater source localizatio… ▽ More In this paper, we propose an interpretable feature selection method based on principal component analysis (PCA) and principal component regression (PCR), which can extract important features for underwater source localization by only introducing the source location without other prior information. This feature selection method is combined with a two-step framework for underwater source localization based on the semi-supervised learning scheme. In the framework, the first step utilizes a convolutional autoencoder to extract the latent features from the whole available dataset. The second step performs source localization via an encoder multi-layer perceptron (MLP) trained on a limited labeled portion of the dataset. The proposed approach has been validated on the public dataset SwllEx-96 Event S5. The result shows the framework has appealing accuracy and robustness on the unseen data, especially when the number of data used to train gradually decreases. After feature selection, not only the training stage has a 95\% acceleration but the performance of the framework becomes more robust on the depth and more accurate when the number of labeled data used to train is extremely limited. △ Less

Submitted 25 November, 2020; originally announced November 2020.