由于水下环境的复杂性水声目标的识别一直是水声领域研究的热点。目前，基于深度学习的水声目标识别方法大多是基于单一的时域或者频域信号提取水声特征，而忽略了两者之间的时频互补信息，而时频互补信息有助于提高水声目标识别的精度。因此，本文同时从时域和频域角度出发，提出一种基于时频联合和加权决策的水声目标识别方法。该方法首先采用长短时记忆网络(LSTM)提取水声信号的时域特征进行识别，然后采用二维卷积神经网络(2D-CNN)提取水声信号的频域特征进行识别，最后将二者的识别结果进行加权决策融合。该方法的有效性在ShipEar数据集上进行验证，其识别精度达94.13%，高于其他现有方法。该方法为基于深度学习的水声目标识别方法的发展提供了新思路。
Recognition of hydroacoustic targets has been a hot issue in the field of hydroacoustics due to the complexity of the underwater environment. At present, most of the deep learning-based hydroacoustic target recognition methods are based on a single time-domain or frequency-domain signal to extract hydroacoustic features, while ignoring the time-frequency complementary information between the two, which can help improve the accuracy of hydroacoustic target recognition. Therefore, this paper proposes a hydroacoustic target recognition method based on joint time-frequency features and weighted decision from both the perspective of time and frequency domains. The method firstly adopts long short term memory network (LSTM) to extract time domain features of hydroacoustic signals for recognition; then adopts two-dimensional convolutional neural network (2D-CNN) to extract frequency domain features of hydroacoustic signals for recognition; finally, the recognition results of both are weighted for decision fusion. The effectiveness of the method was verified on the ShipEar dataset, and its recognition accuracy reached 94.13%, which is higher than other existing methods. The method provides a new idea for the development of deep learning-based hydroacoustic target recognition methods.
2024,46(1): 137-142 收稿日期：2022-10-24
DOI：10.3404/j.issn.1672-7649.2024.01.023
分类号：TN912.34
基金项目：海洋防务技术创新中心创新基金资助项目（No.JJ202170503)
作者简介：潘晓英(1981-),女,博士,教授,研究方向为数据增强
参考文献：
[1] 黄海宁, 李宇. 水声目标探测技术研究现状与展望[J]. 中国科学院院刊, 2019, 34(3): 264-271.
HUANG H N, LI Y, et al. Underwater acoustic detection: current status and future trends[J]. Bulletin of Chinese Academy of Sciences, 2019, 34(3): 264-271.
[2] 李兰瑞, 李鹏, 刘天宇, 等. 水声信号检测与识别技术研究现状[J]. 通信技术, 2020, 53(12): 2904-2907.
LI L R, LI P, LIU T Y, et al. Research status of hydroacoustic signal detection and identification technology[J]. Communications Technology, 2020, 53(12): 2904-2907.
[3] 岳成海, 王旭, 宫俊玲, 等. 基于深度学习的水下目标声学识别与定位技术研究[J]. 数字海洋与水下攻防, 2021, 4(6): 492-497.
YUE C H, WANG X, GONG J L, et al. Research on underwater target recognition and location technique based on deep learning[J]. Digital Ocean & Underwater Warfare, 2021, 4(6): 492-497.
[4] 张少康, 王超, 田德艳, 等. 长短时记忆网络水下目标噪声智能识别方法[J]. 舰船科学技术, 2019, 41(12): 181-185.
ZHANG S K, WANG C, TIAN D Y, et al. Intelligent recognition of underwater target noise based on long short-term memory networks[J]. Ship Science and Technology, 2019, 41(12): 181-185.
[5] 王升贵, 胡桥, 陈迎亮, 等. 基于深度学习的水下目标识别方法研究[J]. 舰船科学技术, 2020, 42(12): 141-145.
WANG S G, HU Q, CHEN Y L, et al. Research on underwater target recognition method based on deep learning[J]. Ship Science and Technology, 2020, 42(12): 141-145.
[6] TIWARI V. MFCC and its applications in speaker recognition[J]. International Journal on Emerging Technologies, 2010, 1(1): 19-22.
[7] YU Y, SI X, HU C, et al. A review of recurrent neural networks: LSTM cells and network architectures[J]. Neural Computation, 2019, 31(7): 1235-1270.
[8] O'SHEA K, Nash R. An introduction to convolutional neural networks[J]. arXiv preprint arXiv: 1511.08458, 2015.
[9] SANTOS-DOMÍNGUEZ D, TORRES-GUIJARRO S, CARDENAL-LÓPEZ A, et al. ShipsEar: An underwater vessel noise database[J]. Applied Acoustics, 2016, 113: 64-69.
[10] 任晨曦, 王黎明, 韩星程, 等. 基于联合神经网络的水声目标识别方法[J]. 舰船科学技术, 2022, 44(1): 136-141.
REN C X, WANG L M, HANG X C, et al. Underwater acoustic target recognition method based on joint neural network[J]. Ship Science and Technology, 2022, 44(1): 136-141.
