使用深度学习的多通道水下目标识别

为解决低信噪比条件下水下目标识别率低的问题,提出一种适用于多通道水听器阵列的深度学习水下目标识别方法。首先是采用子通道特征级联的方法利用多通道信息;在特征提取方面,采用对信号的不同频率区间进行加权的特征提取器,并对提取的特征进行正则规整;最后采用深度神经网络(Deep Neural Network,DNN)实现目标识别。实验首先在仿真条件下对所提出方法的有效性进行验证,结果表明在-15 dB信噪比条件下的五目标识别任务中,使用多通道级联特征的深度神经网络的识别正确率达到96.7%,显著高于基于支持向量机(Support Vector Machine,SVM)的方法。在后续的湖上试验中,深度神经网络的平均正确率达到96.0%,进一步验证了所提出方法的有效性。 

Multi-channel underwater target recognition using deep learning

In order to solve the problem of low recognition rate under low SNR conditions in underwater target recognition,a deep learning based underwater target recognition framework for multi-channel hydrophone arrays is proposed.Firstly,sub-channel feature splicing is used for utilizing multi-channel information.In feature extraction,a feature extractor that weighted different frequency range of the signal is used and then regularizes the extracted features.Finally,a deep neural network was used for target recognition.The effectiveness of the proposed method was verified in the simulation data.The results showed that the recognition accuracy rate of the deep neural network using cascaded features of multichannel signals reached 96.7%in the five-target recognition task under-15 dB SNR,which was significantly higher than the SVM based method.In the experiments on the lake,the average accuracy rate of the deep neural network reached 96.0%,which further illustrated the effectiveness of the method we proposed.