Maximum Entropy Particle Filter For Ocean Acoustic Tomography

Ocean current is one of the important components in the Ocean state monitoring,which can be measured by ocean acoustic tomography(OAT).It aims to infer the state of ocean from the acoustic propagation.Tomography is an image formation.In reality,many random,unpredictable physical factors affect the image structure,which causes an image formation of stochastic process.And the inference of current field also turns into a stochastic process inference.In this paper,we will infer the original image from observation data and research the ocean acoustic tomography under the framework of information theory.Traditional current acoustic tomography is called perturbation method,which is based on ray theory.It projects ray paths to a certain plane,and introduces the basis function to linearize the integral formula of travel time difference.Then we can infer the current field.However,the traditional method is not suitable for deep ocean environments,because it often does not consider the refraction of ray paths.In addition,the traditional method may lead to the inaccuracy and instability of the undetermined inverse problem,because the amount of unknown parameters is larger than observed ones.This paper presents a new method of current acoustic tomography,which constructs a three-dimensional observation equation by layering the propagation of ray paths in vertical direction.Moreover,current acoustic tomography is an image formation,and it is necessary to describe the image with a stochastic process.In this paper,the basis function coefficient representing the current field is modeled as the maximum entropy rate stochastic process,and the state space model is constructed by combining the maximum entropy rate stochastic process with the three-dimensional observation equation.In this paper,the maximum entropy method and particle filter are combined in the Bayesian filtering framework,and the maximum entropy particle filter method is proposed to solve the state space model.Through processing and analyzing the experimental data,we verified that the maximum entropy particle filter method has better accuracy and stability.