Estimation of false alarms rates for GPR detection using the euler characteristic of Gaussian random fields

Thresholds for GPR anomaly detection are often chosen offline and experimentally after data processing. For a real-time operation, especially for mine and IED detection, it is desirable to choose a threshold according to a certain false alarm rate. Because of spatial correlations caused for example by the detection algorithm and heterogeneous soils it is not enough to use only statistics of individual pixels in the decision map. In this work we present a method for estimating the number of false alarms. For this we apply techniques based on the Euler-Poincare characteristic of stationary Gaussian random fields to an exemplary detection scheme. The necessary parameters are estimated from only a small target free area in the data.