With the increasing application of underwater acoustic target recognition in maritime early warning, seabed exploration, and marine resource protection, accurate recognition of ship acoustic targets has become increasingly important. However, the limited availability of ship noise data often restricts the performance of conventional recognition methods. To address this issue, this paper proposes a ship noise sample construction and recognition method based on multidimensional acoustic features. First, data augmentation and enhancement strategies are adopted to increase the diversity of ship noise samples. Then, Mel spectrogram, MFCC, and MFCC-Mix features are extracted to construct multidimensional acoustic representations. Based on these features, a multi-channel deep residual network (MCDRN) is designed for ship acoustic target recognition. Experimental results show that the proposed method achieves higher recognition accuracy than several baseline models.
Xu et al. (Thu,) studied this question.