Determining the effectiveness of using three-dimensional printing to train computer vision systems for landmine detection

The object of this study is the effectiveness of using three-dimensional printing to train computer vision models for landmine detection. The ongoing war in Ukraine has resulted in significant landmine contamination, particularly after russia’s full-scale invasion in 2022. Given the enormous amount of potentially landmine-contaminated land, fast and efficient demining techniques are required, as human probing and metal detectors are labor-intensive and slow-moving. Machine learning offers promising solutions to speed up the landmine detection process by deploying recognition models on robots and unmanned aerial vehicles. However, training such systems faces certain challenges. Firstly, the number of annotated data available for training is limited, which can hinder the model’s ability to generalize to real-world scenarios. Secondly, the use of real or even defused landmines is dangerous due to the potential for accidental detonation. This study aims to overcome the problem of limited data and the risk of using real landmines. Three-dimensional printing makes it possible to create safe and diverse training data, which is essential for model performance. The model trained on replicas, achieved 98 % and 91 % precision on printed and actual landmines, respectively. This high precision is attributed to the realism of copies and the use of advanced machine learning algorithms. This approach successfully addressed the research problem due to the safety, accessibility and diversity of copies. The models trained on copies of landmines could be used in humanitarian demining operations. These operations often employ unmanned aerial vehicles or robots to identify landmines that are thrown remotely, exposed on the surface, or partially hidden