【摘要】 针对经验正交函数(Empirical Orthogonal Function, EOF)建模反演得到的声速剖面(Sound Speed Profile, SSP)估计值分辨率比较低的问题,文章采用字典学习方法中的K-奇异值分解(K-Singular Value Decomposition, K-SVD)算法生成声速剖面的非正交原子,研究了该方法生成的学习字典(Learning Dictionary, LD)对声速剖面的重建性能。首先,采用K-SVD算法从获得的数据中训练SSP字典,然后利用正交匹配追踪(Orthogonal Matching Pursuit, OMP)的稀疏方法给出训练信号的稀疏向量,最后通过得到的最优学习字典和稀疏向量反演得到SSP的估计值。结果表明,K-SVD算法比EOF算法使用更少的基函数即可很好地描述SSP的变化,获得更高的反演精度。更多还原

【Abstract】 In view of the problem that the resolution of the estimated value of sound speed profile(SSP) obtained by the empirical orthogonal function(EOF) modeling and inversion is low, the K-singular value decomposition(K-SVD) dictionary learning method is used to generate non-orthogonal atoms of the sound speed profile, and the performance of the learning dictionary(LD) generated by this method in reconstruction of sound speed profile is studied. First, the K-SVD algorithm is used to train the SSP dictionary from the obtained data, then the sparse method of orthogonal matching pursuit(OMP) is used to give the sparse vector of the training signal, and finally the estimated value of SSP is obtained through the obtained optimal learning dictionary and sparse vector inversion. The results show that the K-SVD algorithm uses fewer basis functions than the EOF algorithm to describe the SSP changes well and obtain higher inversion accuracy.更多还原