Sparsity-based approach for ocean acoustic tomography using learned dictionaries

Abstract

Ocean acoustic tomography (OAT) is commonly used to infer the ocean environmental changes from acoustic measurements. Prior information of sound speed, if used judiciously in OAT inversion, can make a significant contribution to the accuracy improvement. In this paper, a sparsity-based OAT approach is proposed to invert sound speed with effective use of the prior information. By learning a compact dictionary from prior information, the sound speed of interest can be represented sparsely, and the OAT inverse problem can be solved more efficiently by minimizing the cost function with an additional constraint. Simulations of OAT inverse problem using the approach proposed both in horizontal slice and vertical slice demonstrate the advantages of the developed method: it can make good use of prior information of sound speed, such as the sound speed distribution measured by CTD, to enhance the accuracy of inversion; the weights of travel time measured in-situ and prior information can be readily adjusted by changing the values of relevant parameters, which enhances the flexibility of the proposed algorithm.