Machine Learning Algorithm For Acoustic Inverse Media Scattering Using Fixed Frequency Data

Inverse scattering problems are a class of inverse problems which determines the information of the obstacle or non-uniform medium from the collected far field or the near field of the scattering wave.It is a valuable research topic due to its wide applications in practice,e.g.,the diagnosis of human internal lesions,exploration of oil,natural gas and other mineral deposits,satellite remote sensing detection and so on.With the advent of the era of big data,all kinds of information accumulation eventually transformed into data accumulation.Machine learning has becoming a hot research area nowadays as it is capable of learning nonlinear mapping which automatically summarize the internal principles implied by the big data.So that machine learning is applicable to solve nonlinear problems.The main research content of this paper is to design and implement a machine learning algorithm,which takes the scattered field in a certain region and at small amount of measured points as input data to reconstruct the inhomogeneous media.The algorithm is designed with two neural networks to learn the scattering field and the inhomogeneouos media at the same time.The inputs of two neural networks are the position coordinates corresponding to the measured values of the scattering field.The scattering field and inhomogeneity of the media are the outputs of the neural networks.An effective loss function is proposed.A large number of experimental results show that the proposed machine learning algorithm can efficiently reconstruct the inhomogeneous media.