Deep learning method with auditory passive attention for underwater acoustic target recognition under the condition of ship interference

Recognizing the interested target in the presence of interfering ship is a kind of an unavoidable recognition scenario in passive underwater acoustic target recognition. This puts forward higher requirements for underwater acoustic target recognition methods in extracting weak but important features. Inspired by auditory passive attention mechanism of human auditory system, an underwater acoustic target recognition deep learning method based on auditory passive attention is proposed. By applying passive attention loss in a series of deep layers, the proposed method realizes the deep feature grouping according to different categories in any deep layer, which could increase the separation of the deep features and help to extract diverse nuances among targets. Further, three kinds of auditory passive attention loss are considered to achieve passive attention from different perspectives. The recognition experiment and visualization experiment are carried out to verify the performance. The results show that the proposed method achieve best recognition performance, and the average F1-scores are 67.43% (SNR is −18dB), 77.01% (SNR is −12dB), 84.62% (SNR is -6dB), 89.21% (SNR is 0 dB) and 90.79% (SNR is 6 dB). The visualization results verify the proposed method could simulates the auditory passive attention mechanism of human auditory system to some extent.