Digital Twin for Training Set Generation for Unexploded Ordnance Classification

Abstract

The use of machine learning methods for unexploded ordnance (UXO) detection and classification is very limited. This limitation derives from the lack of representative and enough large training data. To overcome this issue we propose a construction of a digital twin where UXO and non-UXO objects are represented using mathematical models in a simulated Earth magnetic field. The use of digital twins allows for simulating and collecting a large training set which can be used for training machine learning models. In the conducted research we discuss obtained results and point out several of the detected problems.