基于融合式神经网络的目标多磁性参数反演研究

Study on multi-magnetic parameters inversion of target based on fusion neural network

针对由不同磁性材料构成的复杂目标体多个磁导率参数反演困难的问题,基于深度学习提出了一种高效的融合式神经网络反演方法.首先,通过融合式神经网络中的分类器将大尺度范围目标磁场信号进行粗化分类,以映射到小尺度局部范围;然后,再通过融合式神经网络中的解算器进行多磁性参数的精细化回归预测.将所提出的融合式神经网络与传统的全连接神经网络进行了仿真实验对比,实验结果表明:融合式神经网络对多磁导率参数的推算精度高达97.5%,比传统全连接神经网络具有更高的预测精度.

Aiming at the problem of difficult inversion for multiple magnetic permeability parameters of complex targets composed of different magnetic materials,an efficient fusion neural network inversion method based on deep learning was proposed.First,the large-scale target magnetic field signal was coarsely classified by the classifier in the fusion neural network,so as to be mapped to the small-scale local range.Then,the refined regression prediction of multi-magnetic parameters was carried out by solvers in the fused neural network.Compared the proposed fusion neural network with the traditional fully connected neural network,simulation experiment results show that the prediction accuracy of multiple magnetic permeability parameters by the fusion neural network is as high as 97.5%,which is higher than that of the traditional fully connected neural network.