A Novel Classification Method for Subsurface Targets

摘要

In this paper, a novel subsurface targets classification method based on the DE-LM joint inversion technique is presented, which requires few control variables and has strong robustness. It eliminates the inaccuracy of the inversion results when an initial value is far away from the truth value by the Levenberg-Marquardt (LM), and reinforces the weak local search capability of the Differential Evolution (DE) inversion technique. Next, the classification method proposed herein extracts subsurface characteristics of targets and classifies the targets automatically by means of the intrinsic responses of three-dimensional orthogonal magnetic dipole instead of using complex machine learning techniques or setting up the targets library for comparison. The experimental results show that the addition of 15% Gaussian white noise to the synthetic data can still lead to convergence to an optimal solution and classify the subsurface targets quickly and accurately, it demonstrates the effectiveness of the method proposed herein.