GPR clutter noise separation by statistical independency promotion

In ground-penetrating radar (GPR) surveys, the GPR data are always contaminated by high-amplitude clutter noise which consists of the director transmitter to receiver signal (also called crosstalk) and antenna ringdown. The crosstalk and antenna ringdown overlay severely the shallow reflection signals. The commonly used approach to filter out the clutter noise is to subtract the mean trace which can give satisfactory results when imaging point reflectors. Unfortunately, the response from an interface may be diminished by using this approach. This paper proposes a novel technique to separate the clutter noise from GPR data which utilizes the statistical independency of clutter noise and reflection signals of plane objects. In this method, the GPR data are modeled as a linear combination of the clutter and reflection signals. With this multichannel signal assumption, a statistical approach based on independent component analysis (ICA) is given to separate the clutter noise and reflection signals. The capacity of the proposed method is illustrated using sand trunk test data and ground track filed data.