This paper presents a novel data acquisition for ground penetrating radars (GPR) based on compressive sensing (CS) by exploiting sparseness in two dimensional FFT of B-scan data.Linear projections of the returned signal with random vectors generated by Mersenne Twister algorithm are taken as measurements.Half-quadric is used to solve

l

1

minimization.Using experimental GPR data

it is shown that the number of acquisition data is only the 1/15 of the number usually considered necessary.