深度多属性增强水声目标识别方法  

Underwater Acoustic Target Recognition Deep Learning Method Based on Multi-attributes Enhancement

作　　者：李俊豪 杨宏晖[1,2] 盛美萍 LI Junhao;YANG Honghui;SHENG Meiping(School of Marine Science and Technology,Northwestern Polytechnical University,Xi’an 710072,China;Ningbo Institute of Northwestern Polytechnical University,Ningbo 315048,China)

机构地区：[1]西北工业大学航海学院,西安710072 [2]西北工业大学宁波研究院,宁波315048

出　　处：《无人系统技术》2023年第1期43-51,共9页Unmanned Systems Technology

基　　金：国家自然科学基金(52171339)。

摘　　要：被动声纳系统由于其隐蔽性好的特点在军事任务中发挥着重要作用。针对被动水声目标识别问题,开展了水声目标多属性特征提取与识别方法研究。利用深度学习方法从舰船辐射噪声中提取目标多属性特征并识别水声目标。提出了深度多属性增强水声目标识别方法,该方法可以从时域舰船辐射噪声中提取水声目标多属性特征及多属性之间的相关性特征,并用来增强深度模型对水声目标类别属性的表达能力。基于海试实测数据的6类水声目标识别实验结果表明,相比于不考虑多属性的识别方法,提出的深度多属性增强水声目标识别方法的平均正确识别率提高了3.6%~18.2%,并且具有更好的识别稳定性。Passive sonar system plays an important role in military operations because of its good stealthiness.Aiming at the problem of passive underwater acoustic target recognition,the multi-attribute feature extraction and recognition method is studied.Deep learning method is considered to extract multi-attribute features from ship radiated noise and recognize underwater acoustic targets.Underwater acoustic target recognition deep learning method based on multi-attributes enhancement is proposed.The proposed method can extract the multi-attribute features and their correlation features of underwater acoustic target from the time-domain ship radiated noise,and these features are used to enhance the expression ability the category attributes of underwater acoustic target.The experimental results of six kinds of underwater acoustic target recognition based on sea test data show that the average correct recognition rate of the proposed method is improved by 3.6%-18.2%compared with the recognition method without considering multiple attributes,and the proposed method has better recognition stability.

关 键 词：水声目标识别 水声目标多属性 深度学习 机器学习 舰船辐射噪声 被动声纳 

分 类 号：TB56[交通运输工程—水声工程]