Single Vector Hydrophone DOA Estimation: Leveraging Deep Learning with CNN-CBAM

In recent years, single vector hydrophones have attracted widespread attention in target direction estimation due to their compact design and advantages in complex underwater acoustic environments. However, traditional direction of arrival (DOA) estimation algorithms often struggle to maintain high accuracy in nonstationary noise conditions. This study proposes the novel DOA estimation method based on a convolutional neural network (CNN) and the convolutional block attention module (CBAM). By inputting the covariance matrix of the received signal into the neural network and integrating the CBAM module, this method enhances the model’s sensitivity to critical features. The CBAM module leverages channel and spatial attention mechanisms to adaptively focus on essential information, effectively suppressing noise interference and improving directional accuracy. Specifically, CBAM improves the model’s focus on subtle directional cues in noisy environments, suppressing irrelevant interference while amplifying essential signal components, which is crucial for an accurate DOA estimation. Experimental results under various signal-to-noise ratio (SNR) conditions validate the method’s effectiveness, demonstrating superior noise resistance and estimation precision, providing a robust and efficient solution for underwater acoustic target localization.