Landmine Detection Using Autoencoders on Multipolarization GPR Volumetric Data

Buried landmines and unexploded remnants of war are a constant threat for the population many countries that have been hit by wars in past years. The huge amount casualties has strong motivation research community toward development safe robust techniques designed landmine clearance. Nonetheless, being able to detect localize buried with high precision an automatic fashion is still considered challenging task due different boundary conditions characterize this problem (e.g., several kinds objects detect, soils meteorological conditions, etc.). In article, we propose novel technique object detection tailored discovery. proposed solution exploits specific kind convolutional neural network (CNN) known as autoencoder analyze volumetric data acquired ground penetrating radar (GPR) using polarizations. This method works anomaly framework, indeed only train on GPR landmine-free areas. system then recognizes dissimilar soil used during training step. Experiments conducted real show requires little no ad hoc preprocessing achieve accuracy higher than 93% sets.