Landmine Detection Using Electromagnetic Spectral Signatures

Abstract

The aim of this thesis was to test and report the applicability of ESS or electromagnetic spectral signature based imaging techniques to detect and identify landmines in both surface deployment and buried deployment. Traditional techniques such as metal detector based listeners while being low cost are being phased out due to the nature of modern constructed landmines. Small IR cameras are cheap and available thanks to the Chinese imports and can be mounted to a small drone to scan vast fields of land in a short period of time. In relative physics, every surface and material has its own electromagnetic signature according to which it reacts with spectral radiation such as light. This results in unique emittance and absorption responses and even color. By applying changing electromagnetic radiation over an object, we can acquire an entire spectral curve which can then be used to decide if a certain object is present at a specific pixel location in the image. For this project, we acquired data on electromagnetic spectral responses for American landmines primarily the claymore and the BLU 43 (butterfly) across various terrains and conditions. A spectral dictionary and bands were defined and the region around 750nm to 1100nm was selected which coincided with LWIR and MWIR bands. The datasets acquired had to undergo several pre-processing techniques and corrections. We rigorously tested all the processes involved from atmospheric corrections to classification, finding out how to reduce computation time to attain real time detection. There were several tradeoffs between detection performance and computing time. We studied several noise factors over the signatures such as wind and rain. We then tested multiple architectures of supervised detection algorithms with our preference to use simpler lighter classifiers to keep computation cost and time minimal