基于深度自编码网络的舰船辐射噪声分类识别
|
摘要
针对水下被动声呐目标分类识别问题,借签深度学习网络在图像、语音等领域的成功运用,提出一种基于深度自编码网络的舰船辐射噪声分类识别方法。首先使用Welch功率谱估计方法获得舰船辐射噪声的功率谱特征,然后对原始训练样本集结构优化得到新训练样本集,并构建训练深度自编码网络。依据总体正确识别概率和各类目标正确识别概率对网络参数进行优化设置,实现对舰船辐射噪声的分类识别。经过大量海上实录舰船辐射噪声的分类识别实验,验证了该方法的可行性和实用性。对比BP神经网络分类器,具有更高的正确分类识别概率。
In order to solve the problem of the classification and recognition of the targets in the underwater passive sonar, a method of the ship radiation noise based on the deep auto-encoding networks is proposed. The deep learning networks has been successfully applied in the fields of the image and speech. Firstly, Welch power spectrum estimation method are used to obtain the power spectral characteristics of the ship radiated noise. Secondly the original training sample set structure is optimized to obtain a new training sample set to construct and train a deep auto-encoding networks.Based on the correct recognition probability of all kinds of targets, the networks parameters are optimized to classify and identify the kind of ships. Using a large number of marine records, the ship radiated noises to experiment, the deep learning networks is a feasible and practicable method. Finally, it can have a higher correct classification recognition probability than the BP neural networks classifier.
In order to solve the problem of the classification and recognition of the targets in the underwater passive sonar, a method of the ship radiation noise based on the deep auto-encoding networks is proposed. The deep learning networks has been successfully applied in the fields of the image and speech. Firstly, Welch power spectrum estimation method are used to obtain the power spectral characteristics of the ship radiated noise. Secondly the original training sample set structure is optimized to obtain a new training sample set to construct and train a deep auto-encoding networks.Based on the correct recognition probability of all kinds of targets, the networks parameters are optimized to classify and identify the kind of ships. Using a large number of marine records, the ship radiated noises to experiment, the deep learning networks is a feasible and practicable method. Finally, it can have a higher correct classification recognition probability than the BP neural networks classifier.
引文
[1]曾庆军,王菲,黄国建.基于连续谱特征提取的被动声纳目标识别技术[J].上海交通大学学报,2002,36(3):382-386.ZENG Qing-jun,WANG Fei,HUANG Guo-jian.Technique of passive sonar target recognition based on continuous spectrum feature extraction[J].Journal of Shanghai Jiaotong University,2002,36(3):382-386.
[2]MEISTER J.A neural network harmonic family classifier[J].J.Acoustic.Soc.Am,1993,9(3):1485-1495.
[3]XIAO Tang.Multiple competitive learning network fusion for object classification[J].IEEE Trans on Systems,Man and Cybernetics part B:Cybernetics,1998,28(4):532-543.
[4]陈兆基,杨宏晖,戴健.自适应免疫算法的SVME用于水下目标识别[J].声学技术,2012,31(6):587-590.
[5]MURTHN MANOHAR,BHASKAR D,RAO.All-polo modeling of speech based on the minimum variance disortionless response spectrum[J].IEEE Transactions on Speech and Audio Processing,2000,8(3):221-239.
[6]赵亚楠,李钢虎,曾渊.基于最小均方无失真响应和支持向量机的被动声纳目标识别[J].声学技术,2011,30(3):223-226.
[7]HINTON G,E,SALAKHUTDINOV R R.Reducing the dimensionality of data with neural networks[J].Science,2006,31(3):504-507.
[8]HINTON G E.Training products of experts by minimizing contrastive divergence[J].Neural Computation,2000,1(14):1771-1800.
[9]HINTON G E,TEH Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[10]RAINA R,BATTLE A,LEE H.Self-taught learning:transfer learning from unlabeled data[C]//International Conference on Machine Learning Corvallis,USA:International Conference press,2007:759-766.
[11]张建华.基于深度学习的语音识别应用研究[D].北京:北京邮电大学,2015.
[12]陈硕.深度学习神经网络在语音识别中的应用研究[D].广州:华南理工大学,2013.
[13]龚丁禧,曹长荣.基于卷积神经网络的植物叶片分类[J].计算机与现代化,2014,2(4):12-19.
[14]刘超颖,杨健,李俊.基于深度自编码网络的异质人脸识别[J].计算机应用与软件,2016,33(10):176-180.
[15]LI J,CHANG H,YANG J.Sparse deep stacking network for image classification[C]//9th AAAI Conference on Artificial Intelligence(AAAI 2015),Austin,Texas,USA:2015.Austin:AAAI Press,2015:1-7.
[16]王强,曾向阳.深度学习方法及其在水下目标识别中的应用[J].声学技术,2015,34(2):138-140.WANG Qiang,ZENG Xiang-yang.Deep learning methods and their applications in underwater targets recognition[J].Technical Acoustics,2015,34(2):138-140.
[17]严韶光,康春玉.基于功率谱特征的CNN被动声纳目标分类方法[J].声学技术,2017,36(6):441-442.YAN Shao-guang,KANG Chun-yu,LI Jun.A method of CNNpassive sonar target classification based on power spectrum feature[J].Technical Acoustics,2017,36(6):441-442.
[18]黄丽霞,王亚楠,张雪英,等.基于深度自编码网络语音识别噪声鲁棒性研究[J].计算机工程与应用,2017,53(13):49-54.
[19]BENGIO Y,LAMBLIN P,POPOVICI D.Greedy layer-wise training of deep networks[C]//Twenty-First Annual Conference on Neural Information Processing Systems(NIPS 2007),Vancouver,British Columbia,Canada:2007.Cambridge:MITPress,2015:153-160.
[20]康春玉,章新华,张安清.一种基于谱估计的被动声呐目标识别方法[J].哈尔滨工程大学学报,2003,24(6):627-631.KANG Chun-yu,ZHANG Xin-hua,ZHANG An-qing.Amethod of passive sonar target recognition based on spectrum evaluation[J].Journal of Harbin Engineering University,2003,24(6):627-631.
[21]ERHAN D,BENGIO Y,COURVILLE A.Why does unsupervised pre-training help deep learning[J].Journal of Machine Learning Research,2010,11(3):625-660.
[2]MEISTER J.A neural network harmonic family classifier[J].J.Acoustic.Soc.Am,1993,9(3):1485-1495.
[3]XIAO Tang.Multiple competitive learning network fusion for object classification[J].IEEE Trans on Systems,Man and Cybernetics part B:Cybernetics,1998,28(4):532-543.
[4]陈兆基,杨宏晖,戴健.自适应免疫算法的SVME用于水下目标识别[J].声学技术,2012,31(6):587-590.
[5]MURTHN MANOHAR,BHASKAR D,RAO.All-polo modeling of speech based on the minimum variance disortionless response spectrum[J].IEEE Transactions on Speech and Audio Processing,2000,8(3):221-239.
[6]赵亚楠,李钢虎,曾渊.基于最小均方无失真响应和支持向量机的被动声纳目标识别[J].声学技术,2011,30(3):223-226.
[7]HINTON G,E,SALAKHUTDINOV R R.Reducing the dimensionality of data with neural networks[J].Science,2006,31(3):504-507.
[8]HINTON G E.Training products of experts by minimizing contrastive divergence[J].Neural Computation,2000,1(14):1771-1800.
[9]HINTON G E,TEH Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[10]RAINA R,BATTLE A,LEE H.Self-taught learning:transfer learning from unlabeled data[C]//International Conference on Machine Learning Corvallis,USA:International Conference press,2007:759-766.
[11]张建华.基于深度学习的语音识别应用研究[D].北京:北京邮电大学,2015.
[12]陈硕.深度学习神经网络在语音识别中的应用研究[D].广州:华南理工大学,2013.
[13]龚丁禧,曹长荣.基于卷积神经网络的植物叶片分类[J].计算机与现代化,2014,2(4):12-19.
[14]刘超颖,杨健,李俊.基于深度自编码网络的异质人脸识别[J].计算机应用与软件,2016,33(10):176-180.
[15]LI J,CHANG H,YANG J.Sparse deep stacking network for image classification[C]//9th AAAI Conference on Artificial Intelligence(AAAI 2015),Austin,Texas,USA:2015.Austin:AAAI Press,2015:1-7.
[16]王强,曾向阳.深度学习方法及其在水下目标识别中的应用[J].声学技术,2015,34(2):138-140.WANG Qiang,ZENG Xiang-yang.Deep learning methods and their applications in underwater targets recognition[J].Technical Acoustics,2015,34(2):138-140.
[17]严韶光,康春玉.基于功率谱特征的CNN被动声纳目标分类方法[J].声学技术,2017,36(6):441-442.YAN Shao-guang,KANG Chun-yu,LI Jun.A method of CNNpassive sonar target classification based on power spectrum feature[J].Technical Acoustics,2017,36(6):441-442.
[18]黄丽霞,王亚楠,张雪英,等.基于深度自编码网络语音识别噪声鲁棒性研究[J].计算机工程与应用,2017,53(13):49-54.
[19]BENGIO Y,LAMBLIN P,POPOVICI D.Greedy layer-wise training of deep networks[C]//Twenty-First Annual Conference on Neural Information Processing Systems(NIPS 2007),Vancouver,British Columbia,Canada:2007.Cambridge:MITPress,2015:153-160.
[20]康春玉,章新华,张安清.一种基于谱估计的被动声呐目标识别方法[J].哈尔滨工程大学学报,2003,24(6):627-631.KANG Chun-yu,ZHANG Xin-hua,ZHANG An-qing.Amethod of passive sonar target recognition based on spectrum evaluation[J].Journal of Harbin Engineering University,2003,24(6):627-631.
[21]ERHAN D,BENGIO Y,COURVILLE A.Why does unsupervised pre-training help deep learning[J].Journal of Machine Learning Research,2010,11(3):625-660.