Overview of machine learning methods in underwater source localization

In this paper, an overview of source localization in underwater acoustics based on machine learning was presented. The machine learning methods involved in this paper included multi-layer perception (feed forward neural network), support vector machine, random forest, and deep neural networks composed of convolutional and fully connected layers. According to the recent studies published on international journals and conferences, the key theoretical basis, single hydrophone and array front-end signal preprocessing, and several typical machine learning models in source localization were described in detail. The problems and challenges in real-world applications were also discussed. At last, to the best knowledge, several potential research topics were given.