UAS Based Mine and UXO Detection in a Model Minefield Environment

Globally, abandoned explosive remnants of war are responsible for thousands of deaths and hindering reconstruction efforts. Traditional methods of demining are dangerous and rely on manual identification and human led exploration. Recent studies have used unmanned aerial systems (UAS) to remotely survey areas potentially affected by landmines and unexploded ordnance (UXO), minimizing the risk to humans in the identification process. These efforts have heavily relied on an operator’s ability to visually identify landmines and UXOs. To minimize the risk of error, parallel studies have been conducted using machine learning (ML) to identify landmines and UXOs. These recent approaches to demining have yielded notable success in locating the abandoned explosives and minimizing human risk. A critical gap in the field of identifying and removing UXOs remains, which is understanding the confounded effects of terrain on UXO identification. This study addresses these challenges through a multispectral approach of UAS-based optical sensors. This study considers various soil conditions including soil type and soil moisture on landmine detection accuracy. Three target objects (simulated UXOs) were placed in a variety of environments (sand, wet sand, top soil, wet top soil, mulch, and wet mulch). All objects placed in this study are meant to replicate common explosives and debris in active and former conflict areas. Data was collected by flying a UAV over a simulated field. Optical sensors were then analyzed on their performance for all soil bed environments based on the rate of detection in raw and binarized form. This study aims to leverage the capabilities of commercially available UAS technology and pre existing data science strategies to assess the efficacy of optical UXO detection methods across diverse environments.