复杂的海洋环境会加大水下目标识别的难度，为进一步提高水下目标识别准确率，本文提出基于特征金字塔融合的识别方法。提取了梅尔频率的倒谱系数（Mel Frequency Cepstrum Coefficient，MFCC）、线性预测倒谱系数（Linear Predictive Cepstral Coefficient，LPCC）、色度频谱和短时能量4种频率信息不同的特征，分别基于特征金字塔完成特征的深层信息与浅层信息之间的融合，并将融合特征分别输入迁移学习后的轻量化神经网络shufflenetV2，进行水下目标识别。在Deepship数据集和Shipsear数据集上进行测试，结果表明，本文中4种频率信息不同特征基于特征金字塔融合后水下目标识别准确率均大于98%，相比于原始特征，识别准确率更高。该方法可应用于海洋资源勘探、海洋防御与安全、海洋环境监测等场景，为水下目标识别领域的研究提供了新的思路。
The complex marine environment increases the difficulty of underwater target recognition. In order to further improve the accuracy of underwater target recognition, this paper proposed a recognition method based on feature pyramid fusion. Four different frequency information features, namely Mel frequency cepstral coefficient (Mel Frequency Cepstrum Coefficient, MFCC), linear prediction cepstral coefficient (Linear Predictive Cepstral Coefficient, LPCC), chromaticity spectrum, and short-term energy, were extracted. Based on the feature pyramid, the deep and shallow information of the features were fused, and the fused features were separately input into the lightweight neural network shufflenetV2 after transfer learning for underwater target recognition. The test results on the deepship dataset and shipsear dataset showed that the underwater target recognition accuracy of the four different frequency information features fused based on feature pyramids in this article were all greater than 98%, and compared to the original features, the recognition accuracy was higher. This method can be applied to scenarios such as marine resource exploration, marine defense and safety, and marine environmental monitoring, providing new ideas for research in the field of underwater target recognition.
2025,47(4): 117-123 收稿日期：2024-5-9
DOI：10.3404/j.issn.1672-7649.2025.04.019
分类号：TP391.4
基金项目：国家自然科学基金资助项目（51805154）；湖北省自然科学基金资助项目 （2022CFB473）；湖北省教育厅科学研究计划项目（B2022049）
作者简介：刘梦然(1991-),女,博士,讲师,研究方向为声传感器与信号处理。
参考文献：
[1] 李旻择. 基于深度学习的声目标识别方法研究[D]. 绵阳: 西南科技大学, 2020.
[2] 姜囡, 庞永恒, 高爽. 基于注意力机制语谱图特征提取的语音识别[J]. 吉林大学学报(理学版), 2024, 62(2): 320-330.
JIANG N, PANG Y H, GAO S. Speech recognition based on attention mechanism and spectrogram feature extraction[J]. Journal of Jilin University(Science Edition), 2024, 62(2): 320-330.
[3] 王丽, 涂冰花, 王伟, 等. 长短期记忆循环神经网络在自动语音识别中的应用[J]. 单片机与嵌入式系统应用, 2023, 23(10): 49-52+56.
WANG L, TU B H, WANG W, et al. Application of short-term and long-term memory cyclic neural network in automatic speech recognition[J]. Microcontrollers & Embedded Systems, 2023, 23(10): 49-52+56.
[4] LI D, LIU F, SHEN T, et al. Data augmentation method for underwater acoustic target recognition based on underwater acoustic channel modeling and transfer learning[J]. Applied Acoustics, 2023, 208: 109344.
[5] 张玉皓, 李立钢. 改进的SqueezeNet网络在船舶分类中的应用[J]. 传感器与微系统, 2022, 41(1): 150-152+160.
ZHANG Y H, LI L G. Application of improved SqueezeNet in ship classification[J]. Transducer and Microsystem Technologies, 2022, 41(1): 150-152+160.
[6] 简泽明, 赵旭辉, 胡君豪, 等. 基于改进ShuffleNetV2模型的声目标识别方法研究[J]. 传感器与微系统, 2023, 42(8): 43-45+49.
JIAN Z M, ZHAO X H, HU J H. Research on acoustic target recognition method based on improved ShuffleNetV2 model[J]. Transducer and Microsystem Technologies, 2023, 42(8): 43-45+49.
[7] AL-HATTAB Y A, ZAKI H F, SHAFIE A A. Rethinking environmental sound classification using convolutional neural networks: optimized parameter tuning of single feature extraction[J]. Neural Computing and Applications, 2021, 33(21): 14495-14506.
[8] YASSIR A, ANDALOUSSI S J, OUCHETTO O, et al. Acoustic fish species identification using deep learning and machine learning algorithms: A systematic review[J]. Fisheries Research, 2023, 266: 106790.
[9] HU G, WANG K, LIU L. Underwater acoustic target recognition based on depthwise separable convolution neural networks[J]. Sensors, 2021, 21(4): 14-43.
[10] WANG B, ZHANG W, ZHU Y, et al. An underwater acoustic target recognition method based on AMNet[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 1-5.
[11] XUE L, ZENG X, JIN A. A novel deep-learning method with channel attention mechanism for underwater target recognition[J]. Sensors, 2022, 22(15): 54-63.
[12] IRFAN M, JIANGBIN Z, Ali S, et al. DeepShip: an underwater acoustic benchmark dataset and a separable convolution based autoencoder for classification[J]. Expert Systems with Applications, 2021, 183(5): 115270.
[13] SANTOS-DOMÍNGUEZ D, TORRES-GUIJARRO S, CARDENAL-LÓPEZ A, et al. Shipsear: An underwater vessel noise database[J]. Applied Acoustics, 2016, 113: 64-69.
[14] 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, HAN X C, et al. Underwater acoustic target recognition method based on joint neural network[J]. Ship Science and Technology, 2022, 44(1): 136-141.
[15] KIRILLOV A, GIRSHICK R, He K, et al. Panoptic feature pyramid networks[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019.
[16] 易振通, 吴瑰, 官端正, 等. 轻量化卷积神经网络的研究综述[J]. 工业控制计算机, 2022, 35(10): 109-111+114.
YI Z T, WU G, GUAN D Z, et al. Survey of research on lightweight convolutional neural networks[J]. Industrial Control Computer, 2022, 35(10): 109-111+114.
[17] MA N, ZHANG X, ZHENG H T, et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[C]//Proceedings of the European conference on computer vision (ECCV), 2018.
