基于深度学习的船舶辐射噪声识别研究  

Ship-radiated noise recognition research based deep learning

作　　者：朱可卿 田杰[1] 黄海宁[1] 张扬帆[1] 

机构地区：[1]中国科学院声学研究所,北京100190 [2]中国科学院大学,北京100049

出　　处：《应用声学》2018年第2期238-245,共8页Journal of Applied Acoustics

摘　　要：为了改善船舶辐射噪声识别系统的性能,进一步提高船舶辐射噪声识别的正确率,该文提出采用一种基于深度学习的船舶辐射噪声识别方法。该方法首先提取了船舶辐射噪声的频谱、梅尔倒谱系数等特征,将提取特征后的图像样本分别用于训练卷积神经网络和深度置信网络,再对船舶辐射噪声进行识别。通过文中所给实例,将深度学习和支持向量机两种识别方法的性能进行比较,得出深度学习方法可以有效地提高船舶辐射噪声识别正确率的初步结论。In order to improve the accuracy and the performance of ship-radiated noise recognition system,this paper introduces a method for ship-radiated noise recognition based deep learning. First, we extract the features of ship-radiated noise, such as spectrum feature, Mel-frequency cepstral coefficients(MFCC), etc.Then we train convolutional neural network(CNN) and deep belief network(DBN) with these feature-extracted noise image samples and recognize ship-radiated noise. After that we make a contrast with the performance of classification of support vector machine(SVM). The result shows that deep learning-based method in this paper can improve the accuracy of ship-radiated noise recognition effectively.

关 键 词：深度学习 卷积神经网络 深度置信网络 船舶辐射噪声识别 

分 类 号：TB566[交通运输工程—水声工程]