A Novel AMSS-FFN for Underwater Multisource Localization Using Artificial Lateral Line

The lateral line organs of fish present a promising idea for achieving near-field target awareness. Inspired by the localization of targets by fish, a neural network approach to localize underwater vibration sources is a feasible technical approach. However, previous methods using neural networks are relatively simple, and the robustness of the model is weak. Moreover, the previous studies only focused on single-source localization problems, which were inconsistent with the reality of multiple vibration sources in underwater environments. To address these issues, we develop an artificial lateral system with integrated pressure sensors for acquiring pressure transform signals from underwater multisource vibrations. A novel attention mechanism-based multisensing multisource feature fusion network (AMSS-FFN) is proposed to exploit the information of data fully. Specifically, to reflect the superiority of convolutional neural networks (CNNs) for extracting image features, we transform the 1-D signal into a 2-D grayscale image and a time-frequency image processed by Stockwell transformer. Furthermore, a hybrid attention mechanism that learns channel and location information is introduced, allowing globally important features to be represented. Finally, we use a dynamic learning strategy to fuse features in the time and time-frequency domains. The effectiveness of the method is verified using a laboratory-measured dataset. The results indicate that the prediction accuracy of the proposed method is significantly improved.