Issue: 11 , Volume: 66 , Уear of publication: 2026

The task of estimating the depth of the roof occurrence, electrical conductivity, and induction mass of conducting objects in a weakly conducting medium using fully connected neural networks has been considered. Previously identified dependencies between the parameters of the EMF signal anomaly distribution from conducting objects and the object parameters made it possible to formulate a regression task for the neural network. To generate the training dataset, numerical simulation of EMF signals was performed using the vector finite element method. After training the neural network on a small dataset of EMF signals from model problems with bed-like objects, the parameters were obtained for objects of the same shape, as well as for objects shaped as a sphere and a hemisphere.