The processed Deepships dataset and ShipsEar dataset.

Underwater acoustic target classification (UATC) aims to identify the type of unknown acoustic sources using passive sonar in oceanic remote sensing scenarios. However, the variability of underwater acoustic environment and the presence of complex background noises create significant obstacles to improving accuracy of UATC. To address these challenges, we develop an innovative deep neural network (DNN) algorithm integrated by  multiscale feature extractor and efficient channel attention mechanism. Firstly, auditory fusion features, including MFCC and GFCC, along with their differential values, are concatenated to represent the amplitude and phase structure information of underwater acoustic signals in time-frequency (TF) domain. Secondly, the integration of multi-scale convolution with an efficient channel attention (ECA) mechanism is introduced to learn and select crucial information from the auditory fusion features. The proposed algorithm efficiently manages and refine the importance of coarse-to-fine representations of acoustic signals, thereby improving the adaptability and reliability in various UATC tasks. Experimental results using the provided datasets have demonstrated that the proposed algorithm significantly outperforms state-of-the-art methods in classification accuracy.

水下声目标分类（Underwater acoustic target classification, UATC）旨在海洋遥感场景中，利用被动声呐识别未知声源的类型。然而，水下声学环境的多变性与复杂背景噪声的存在，给提升UATC的分类精度带来了显著阻碍。为应对上述挑战，我们提出了一种融合多尺度特征提取器与高效通道注意力机制的创新深度神经网络（Deep Neural Network, DNN）算法。首先，将梅尔频率倒谱系数（Mel-Frequency Cepstral Coefficients, MFCC）、伽马通频率倒谱系数（Gammatone Frequency Cepstral Coefficients, GFCC）及其差分参数进行拼接，以表征水下声信号在时频（Time-Frequency, TF）域的幅相结构信息。其次，引入多尺度卷积与高效通道注意力（Efficient Channel Attention, ECA）机制的融合结构，从上述听觉融合特征中学习并筛选关键信息。所提算法可有效管理并细化声信号从粗到细的表征权重，进而提升各类UATC任务中的适应性与可靠性。基于所提供的数据集开展的实验结果表明，所提算法的分类精度显著优于当前最优方法。