ShipNet: an Efficient Multi-branch Network for Underwater Acoustic Target Recognition
Time:
As the ocean plays an increasingly crucial role in global ecology and economy, monitoring and protecting the marine environment have become essential tasks. Compared to radio signals, acoustic signals perform better in underwater environments, enabling longer-distance and more stable transmission. Monitoring the acoustic signals of ships is crucial not only for ensuring maritime safety but also for assessing the impact of ship activities on marine ecosystems in real-time. Therefore, monitoring the acoustic signals of ships is vital for marine safety and environmental protection. However, existing studies are limited and often only consider local features, resulting in sub-optimal model performance in complex underwater environments. To address this, we propose ShipNet, the first framework that integrates multi-branch feature fusion, along with the incorporation of dense connectivity and attention mechanisms to enhance the model’s recognition capability and robustness. Extensive experimental results demonstrate that ShipNet achieves state-of-the-art performance across all tested datasets. Additionally, we conduct comprehensive ablation experiments to validate the effectiveness of each branch and module within the system, further proving the superiority of ShipNet.
Recommended citation: