Enhancing Accuracy of Metal Target Parameter Estimation Using Neural Networks

Detection technology based on electromagnetic induction is effective in detecting metal targets in an underground environment, and inversion is a common method for solving parameters of these targets. However, in practical applications, the accuracy of detection is often compromised because the metal target is always approximated as a dipole model, leading to the nonlinear overdetermined problem during the parameter estimation. Therefore, this paper proposes a neural network-based algorithm for the inversion of underground metal target parameters to address this issue. Specifically, a neural network model suitable for inversion is first constructed for the metal target inversion problem, thereby transforming the nonlinear overdetermined problem into an optimization problem constrained by the backpropagation of the neural network. Furthermore, the LM algorithm is employed to optimize the iteration process and reduce the time of training of the neural network model. Finally, the numerical experimental results demonstrate that the proposed algorithm completes the optimization in less time and number of iterations. Specifically, it can obtain a horizontal position estimation error of only 2cm and a depth estimation error of only 5cm within 2 s. The results verify the proposed algorithm not only improves accuracy but also meets the requirements of rapid and real-time.