Surface and Underwater Acoustic Source Discrimination Based on Machine Learning Using a Single Hydrophone

In shallow water, passive sonar usually has great difficulty in discriminating a surface acoustic source from an underwater one. To solve this problem, supervised machine learning method using only one hydrophone is implemented paper. Firstly, simulated training data are generated by normal mode model KRAKEN with the same environment setup as that SACLANT 1993 experiment. Secondly, k-nearest neighbor (kNN) classifiers trained and evaluated scores of precision, recall, F1 accuracy. Thirdly, random subspace kNN finely on three hyperparameters (the number nearest neighbors, predictors selected at learners ensemble) to obtain best model. Fourthly, deep called ResNet-18 also applied, it reaches balance between precision while accuracies both simulation experimental all 1.0. Further, 48 hydrophones vertical linear array (VLA) analyzed kinds methods (kNN, ResNet-18) separately, results compared. It concluded performance best. Both suggest feasibility discrimination even single hydrophone.