针对传统机器学习在水下目标识别方面严重依赖先验知识、识别精度低的难题，提出基于深度学习的水下目标辨识方法。该方法通过短时傅里叶变换进行时频分析获取水下目标信号的LOFAR谱图，将目标从一维序列空间映射至类别可分性更高的二维矢量空间。利用深度卷积神经网络自适应实现对目标LOFAR图特征提取，最后采用全连接层将特征变换至类别空间，用Softmax函数实现水下目标智能辨识。结合7类不同水下目标的实测舰船辐射噪声数据从网络模型结构参数、激活函数、池化方法以及数据片段长度等方面对深度学习分类精度进行验证。结果表明，利用二维时频谱图变换和卷积神经网络相结合的方法可有效降低噪声的影响，分类精度可达98.57%。验证了基于深度学习的水下目标辨识方法的有效性，为海洋装备智能目标探测与识别提供了一种新的研究思路与方法。
Aiming at the problem that traditional machine learning relies heavily on prior knowledge and low recognition accuracy in underwater target recognition, an underwater target recognition method based on deep learning is proposed. This method uses short-time Fourier transform to perform time-frequency analysis to obtain the LOFAR spectrogram of the underwater target signal, and maps the target from a one-dimensional sequence space to a two-dimensional vector space with higher class separation. Then the deep convolutional neural network is used to adaptively extract the features of the target LOFAR map, and finally the fully connected layer is used to transform the features into the category space, and the softmax function is used to realize the intelligent identification of underwater targets. Combining the actual measured ship radiated noise data of 7 different underwater targets, the deep learning classification accuracy is verified from the network model structure parameters, activation function, pooling method, and data segment length. The results show that the two-dimensional time-spectrogram transform and The combined method of convolutional neural network can effectively reduce the impact of noise, and the classification accuracy can reach 98.57%, which verifies the effectiveness of the underwater target identification method based on deep learning, and provides a new approach for intelligent target detection and recognition of marine equipment Research ideas and methods.
2020,42(12): 141-145 收稿日期：2020-08-20
DOI：10.3404/j.issn.1672-7649.2020.12.028
分类号：TP391
作者简介：王升贵(1978-),男,高级工程师,主要从事机电一体化研究
参考文献：
[1] HINTON G, SALAKHUTDINOV R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(504)
[2] 杨宏晖, 申昇, 姚晓辉, 等. 用于水声目标特征学习与识别的混合正则化深度置信网络[J]. 西北工业大学学报, 2017(2)
[3] 陈越超, 徐晓男. 基于降噪自编码器的水中目标识别方法[J]. 声学与电子工程, 2018(1): 30-33
[4] 吕海涛, 巩健文, 孔晓鹏. 基于卷积神经网络的水声目标分类技术[J]. 舰船电子工程, 2019, 39(2): 158-162
[5] LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324
[6] 胡桥, 郝保安, 吕林夏, 等. 基于组合支持向量机的水声目标智能识别研究[J]. 应用声学, 2009, 28(6): 421-430
[7] HU Q, HAO B, LV L, et al. Hybrid intelligent detection for underwater acoustic target using emd, feature distance evaluation technique and FSVDD[C]//2008 International Congress on Image and Signal Processing, IEEE, Sanya, China, 2008, 4: 54-5.
[8] WANG, Y., WANG, C., ZHANG, H. Ship classification in high-resolution SAR images using deep learning of small datasets[J]. Sensors 2018, 18, 2929.
[9] MARTIN S B, POPPER A N. Short- and long-term monitoring of underwater sound levels in the Hudson River (New York, USA).[J]. The Journal of the Acoustical Society of America, 2016, 139(4): 1886-1897
[10] Murat KÜÇÜKBAYRAK, Özhan GÜNEŞ, Nafiz ARICA: Underwater Acoustic Signal Recognition Methods[J]. Journal of Naval Science and Engineering 2009, 5(3): 64-78.
