The Application of Convolutional Neural Networks for Extraction of Magnetic Anomaly Field Lineaments

The article focuses on the application of convolutional neural networks (CNNs) for automated extraction of magnetic anomaly field lineaments. In the course of the work, an original CNN U-Net based architecture with pre-trained VGG-16 weights was developed, and its training was conducted on a sample of 500 model examples. The approach presented in this work can be an optimal tool for structural interpretation of magnetic anomaly fields. As a result of testing the proposed CNNs for magnetic field of the Barents Sea local area, the axes of the linear anomalies were identified, largely coinciding with the position of the axes obtained by manual expert interpretation. These fact demonstrates the high efficiency of applying modern artificial neural network technologies.