A comparative study of GPR reconstruction approaches for landmine detection

This paper compares the performance of three reconstruction techniques frequently applied to process ground penetrating radar (GPR) data in the specific context of landmine detection: 1) Stolt migration, 2) backprojection (BP), and 3) microwave tomographic inversion (MWT). The detection results provided by these algorithms are contrasted with the ones obtained from typically adopted GPR data filtering procedures. To carry out the analysis, we use experimental data collected at a specifically prepared test field, where different targets were buried at shallow depths in inhomogeneous soil. The efficiency of the investigated approaches is quantitatively evaluated in terms of detection accuracies (ROC curves) obtained applying a single pixel-based and an averaged energy detection algorithm. Based on this analysis, we found that the MWT outperforms the other reconstruction algorithms for the smallest mines, which are the most difficult to detect. On the other hand, MWT and BP reconstruction techniques achieve comparable performances against medium mines, and do not improve the outcome in case of big mines with respect to the filtering approaches. Finally, Stolt migration produces the worst results for both medium and small mines.