Deep-learning-based feature extraction method for single-vector hydrophone signals

Single-vector hydrophone signal feature extraction is a key link in hydroacoustic target identification. In this paper, a deep learning feature extraction method based on improved ResNet50 is proposed for the problem of insufficient performance of traditional feature extraction methods in low signal-to-noise ratio environments. By designing the feature fusion module and introducing the attention mechanism, the accuracy and anti-interference ability of feature extraction are significantly improved. The method is verified to have good robustness in real sea conditions and provides a new technical way for hydroacoustic target recognition.