Abstract
In deep-sea source localization, some of the existing methods only estimate the source range, while the others produce large errors in distance estimation when estimating both the range and depth. Here, a convolutional neural network-based method with high accuracy is introduced, in which the source localization problem is solved as a regression problem. The proposed neural network is trained by a normalized acoustic matrix and used to predict the source position. Experimental data from the western Pacific indicate that this method performs satisfactorily: the mean absolute percentage error of the range is 2.10%, while that of the depth is 3.08%.
Funding Information
  • Overseas Expertise Introduction Project for Discipline Innovation (B18041)
  • National Natural Science Foundation of China (11974286)
  • Excellent Young Scientists Fund (41906160)

This publication has 12 references indexed in Scilit: