利用U-Net网络增强骨密度全波形反演

Enhanced full waveform inversion with U-Net for bone mineral density measurement

该文设计了一种使用U-Net网络解决骨密度全波形反演的初值依赖、多解、病态等问题的方法。首先使用逆时偏移成像,将其结果输入神经网络得到模型的预分布。将该分布作为全波形反演目标函数的约束,可以使反演的结果更接近最优值,还可以减小反演对初始值的依赖。该文进行了一些模拟实验,得出该文的方法可以改进全波形反演对初始值的依赖和容易陷入局部极值的问题。

This paper designs a method to solve the initial value dependence,multiple solutions,and ill-conditions of the full waveform inversion of bone mineral density with U-Net.First,migration imaging is used,and the results are input to the neural network to obtain the pre-distribution of the model.Using this distribution as the constraint of the objective function of the full waveform inversion can make the inversion result closer to the optimal value and reduce the dependence on the initial value.In this paper,some simu-lation experiments have been carried out,and it is concluded that the method in this paper can improve the dependence of the full waveform inversion on the initial value and easily fall into the problem of local extreme value.