Surface and underwater acoustic target recognition using only two hydrophones based on machine learning.

Area of Science:

• Ocean acoustics
• Signal processing
• Machine learning

Background:

• Distinguishing surface and underwater acoustic targets is challenging due to noise interference.
• Previous single hydrophone methods for passive sonar target recognition suffered from limited accuracy.

Purpose of the Study:

• To develop an accurate method for surface and underwater (S/U) acoustic targets recognition.
• To investigate the effectiveness of using two hydrophones with a Gradient Boosting Decision Tree for improved target identification.

Main Methods:

• A Gradient Boosting Decision Tree model was employed for S/U acoustic targets recognition.
• A large training dataset was generated using the KRAKEN acoustic model.
• Vertical linear array feature extraction was used to assimilate heterogeneous simulation and experimental data.

Main Results:

• The proposed two-hydrophone system achieved 100% accuracy for surface targets and 0.9384 for underwater targets using SACLANT 1993 data.
• Single hydrophone recognition accuracy was significantly lower, at 0.4715 for surface and 0.5620 for underwater targets.
• The model successfully utilized channel information alongside source spectrum information for recognition.

Conclusions:

• The two-hydrophone approach significantly outperforms single hydrophone systems in S/U acoustic target recognition.
• The integration of acoustic modeling, feature extraction, and machine learning provides a robust solution for passive sonar applications.
• This method offers a promising advancement for accurate underwater acoustic target identification.