A Convolutional Neural Network Combining Classification and Regression for Source Localization in Shallow Water

Abstract

Abstract Machine learning (ML) has been widely applied to ocean acoustic source localization. The localization problem is often regarded as a classification problem or a regression problem in the previous works. This paper proposes a convolutional neural network (CNN) combining classification and regression (CR-CNN) for source localization in shallow water with vertical array data. The normalized sample covariance matrices (SCMs) of the broadband data received by a vertical line array calculated by an acoustic propagation model are used as the input features of the network in the training process. The proposed method is verified by the observation data in the shallow water area of the East China Sea. The results of simulation and real data show that the performance of proposed method is better than that of the separate classification CNN (C-CNN) and MFP. And the localization results of the experimental data are basically consistent with the geoacoustic parameters sensitivity analysis. The depth estimation of the CR-CNN is much better than MFP, while the range estimation performance of CR-CNN, C-CNN and MFP are equivalent.