Machine Learning Based Method for Acoustic Recognition of Anthropogenic Underground Voids

Underground tunnel identification remains difficult in spite of the development of recognition methods. There is often no direct access to the Earth's surface above bunkers; above-ground structures may be located there; the soil structure may contain dense layers opaque to radio waves. Therefore, it is promising to use geophysics acoustic recognition methods. The study aims to develop an information system schematic diagram for identifying man-made underground cavities using machine learning methods based on acoustic reconnaissance data. In contrast to known methods for monitoring vibration waves during earthquakes, the initiation of vibration waves can be carried out by delivering precise strikes of known power to specified surface points. However, solving the inverse problem for the propagation of acoustic waves is problematic due to the small relative sizes of these structures. The scientific novelty lies in the fact that we find a solution to this problem using machine learning methods based on model calculations for a known geological structure of the soil. Combining it with satellite observation data on the above-ground structures makes it possible to build a neural network for analyzing the vibrations of sensors located in controlled territory.