Beschreibung:
As a classification problem in machine learning, source localization is solved by training a feed-forward neural network (FNN) on ocean acoustic data. The FNN is fed with normalized sample covariance matrices (SCMs). The output of network, which represents the probability for range, is used to determine the source ranges. Since it is a data-driven method, no acoustic propagation models are needed. As shipping noise has a broad frequency band, an approach of statistical inference for source localization is presented by taking advantage of multi-frequency information. It is demonstrated by the vertical array data from Noise09 experiment.