摘要
实际的海洋环境背景声场通常不服从高斯分布,此情况下多种水声信号处理方法的前提条件得不到满足。针对非高斯海洋背景场下的噪声预处理,研究了一种具有促稀疏性的多层贝叶斯建模方法,使用吉布斯采样对变量的后验概率进行求解。对两组在不同海域采集的水声数据进行建模,分析结果表明非高斯背景场中高阶分量混合系数虽然较小,但对真实噪声分布的描述仍具有较大作用
The real ocean background acoustic field is usually not subjected to the Gaussian distribution,which violates the precondition of many processing methods of underwater acoustic signal. For the problem of noise pretreatment in non-Gaussian background field,a hierarchical Bayesian model which promotes the sparsity is studied and conditional probabilities are derived using Gibbs sampling. Two different sets of underwater acoustic data are modeled using the proposed method and the results indicate that the high order components are important for the accurate description of the real distribution,although they have smaller mixture coefficients.
引文
[1]刘竹青.非高斯混响背景下的信号检测[D].哈尔滨工程大学,2008.
[2]李翠萍.基于高阶统计量的非高斯ARMA模型辨识方法的研究[D].长沙:国防科学技术大学,2002.
[3]MZA’YA V,SCHMIDT G.On approximate approximations using Gaussian kernels[J].Ima Journal of Numerical Analysis,1996,16(1):13-29.
[4]BISHOP.Pattern Recognition and Machine Learning(Information Science and Statistics)[M].Springer-Verlag New York,Inc.2006.
[5]MENG D,DEL T F.Robust Matrix Factorization with Unknown Noise[J].2013:1337-1344.
[6]STERGIOPOULOS,STERGIOS.Advanced signal processing handbook[M].CRC Press,2001.
[7]GILKS W R,WILD P.Adaptive Rejection Sampling for Gibbs Sampling[J].Journal of the Royal Statistical Society,1992,41(2):337-348.
[8]RASMUSSEN C E.The Infinite Gaussian Mixture Model[C]//Advances in Neural Information Processing Systems.DBLP,2000:554-560.
[9]TIPPING M E.Sparse Bayesian learning and the relevance vector machine[J].Journal of machine learning research,2001,1(Jun):211-244.
[10]MORETTI D,DIMARZIO N,MORRISSEY R,et al.Overview of the 3 RD International workshop on the detection and classification of Marine Mammals using passive acoustics[J].Canadian Acoustics,2008,36(1):7-11.