Document Type : Original Research Paper
Authors
1 Faculty of Mining, Petroleum & Geophysics Engineering, Shahrood University of Technology, Shahrood, Iran
2 Shahrood University of Technology
3 Faculty of Mining, Petroleum and Geophysics, Shahrood University of Technology, Shahrood, Iran
4 University of Cologne, Post Doctoral Researcher
Abstract
The technique referred to as Complex Resistivity (CR) or Spectral Induced Polarization (SIP) possesses the capability to distinguish between various kinds of minerals or the sources of induced polarization by utilizing the physical characteristics of minerals or polarizable inclusions. The Generalized Effective Medium Theory of Induced Polarization (GEMTip) model is utilized to derive physical characteristics from SIP data. Different inversion methods are applied for this task, though they encounter difficulties such as computational costs, non-linearity, and the intricacy of the inverse issue. To tackle this, a new inversion approach based on Deep Learning (DL) via Convolutional Neural Network (CNN) is proposed for predicting the parameters of polarizable particles from SIP data. The CNN is trained on 20000 synthetic datasets produced using the GEMTip forward model. While DL networks address non-linearities, specific modifications are applied to synthetic datasets to evaluate the influence of non-linearity and correlation on the results. In the Kervian region southwest of Saqqez city, gold mineralization is linked to quartz and pyrite minerals, with two types of pyrite recognized - coarse-grained barren and fine-grained auriferous. The existence of sulfide mineral pyrite, along with variations in pyrite sizes, presents an attractive target for SIP exploration in the investigated area. The trained network is also validated on Gravian data and effectively retrieves parameters as evidenced by the data. The proposed methodology simplifies the inversion process by estimating parameters in one step, enabling a direct and efficient procedure.
Keywords
Main Subjects
)