Mining sea turtle nests: An amplitude independent feature extraction method for GPR data

We use a Ground Penetrating Radar (GPR) to localize eggs of sea turtles laid in sand. GPR technology has been developed to detect subsurface structures, and successfully applied in archeology, civil engineering, and demining. Typical uses rely on relatively strong signals due to high contrast in dielectric properties of the buried manmade objects and the soil. Signal to noise ratios in our task are substantially lower, as the variances in humidity and granularity of layers of salty sand, and the presence of nuisance artifacts such as rocks, clogs of seashells, air pockets, etc., contribute to making turtle nest detection a challenging task. We present a combination of signal processing, pattern recognition, and feature selection techniques that stand up to these challenges. Our approach is evaluated using ground truth data collected in the field. We believe that this method can be useful in a range of non-standard GPR applications, especially when the signals to noise ratios are low.