JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Underwater vessel sound recognition based on multi-layer feature and attention mechanism

  • 0Beijing Institute of Petrochemical Technology, Beijing, 102617, China.
Scientific Reports +

|

April 2, 2025

|

April 2, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces a novel Emphasized Dimension Attention and Future Fusion-Time Delay Neural Network (EDAFF-TDNN) for accurate vessel recognition using hydroacoustic signals. The model achieves high accuracy in complex marine environments, improving underwater target identification.

Area Of Science

  • Marine acoustics
  • Signal processing
  • Artificial intelligence

Background

  • Vessel recognition using hydroacoustic signals is crucial but challenging due to complex marine environments and noise.
  • Underwater acoustic signal transmission is affected by environmental variability and interference from sources like waves and marine life.
  • Accurate identification of vessel targets is hindered by signal degradation and noise.

Purpose Of The Study

  • To propose an advanced deep learning model for robust vessel recognition from hydroacoustic signals.
  • To enhance the model's ability to capture contextual information and focus on critical features in noisy conditions.
  • To improve the feature representation capability for better performance in complex underwater scenarios.

Main Methods

  • Development of the Emphasized Dimension Attention and Future Fusion-Time Delay Neural Network (EDAFF-TDNN).
  • Integration of Squeeze and Excitation Block (SE-Block) for dynamic feature map weight adjustment and contextual information capture.
  • Implementation of feature fusion for multi-layer feature extraction and an attention mechanism for focusing on key feature dimensions.

Main Results

  • The EDAFF-TDNN model demonstrated improved performance in capturing contextual information and feature representations.
  • The attention mechanism enabled the model to focus on crucial information, enhancing performance in complex underwater scenarios.
  • Experiments on the ShipsEar dataset yielded a high recognition accuracy of 98.2% for vessel targets.

Conclusions

  • The proposed EDAFF-TDNN model effectively addresses the challenges of vessel recognition in complex marine environments.
  • The combination of emphasized dimension attention and future fusion significantly improves the accuracy and robustness of hydroacoustic signal analysis.
  • This research contributes a powerful tool for underwater target identification, with potential applications in maritime surveillance and safety.

Keywords: