A 2D processing algorithm for detecting landmines using Ground Penetrating Radar data

Abstract

Ground Penetrating Radar(GPR) is one of a number of technologies that have been used to improve landmine detection efficiency. The clutter environment within the first few cm of the soil where landmines are buried, exhibits strong reflections with highly non-stationary statistics. An antipersonnel mine(AP) can have a diameter as low as 2cm whereas many soils have very high attenuation frequencies above 3GHZ. The landmine detection problem can be solved by carrying out system level analysis of the issues involved to synthesise an image which people can readily understand. The SIMCA ('SIMulated Correlation Algorithm') is a technique that carries out correlation between the actual GPR trace that is recorded at the field and the ideal trace which is obtained by carrying out GPR simulation. The SIMCA algorithm firstly calculates by forward modelling a synthetic point spread function of the GPR by using the design parameters of the radar and soil properties to carry out radar simulation. This allows the derivation of the correlation kernel. The SIMCA algorithm then filters these unwanted components or clutter from the signal to enhance landmine detection. The clutter removed GPR B scan is then correlated with the kernel using the Pearson correlation coefficient. This results in a image which emphasises the target features and allows the detection of the target by looking at the brightest spots. Raising of the image to an odd power >;2 enhances the target/background separation. To validate the algorithm, the length of the target in some cases and the diameter of the target in other cases, along with the burial depth obtained by the SIMCA system are compared with the actual values used during the experiments for the burial depth and those of the dimensions of the actual target. Because, due to the security intelligence involved with landmine detection and most authors work in collaboration with the national government military programs, a database of landmine signatures is not existant and the authors are also not able to publish fully their algorithms. As a result, in this study we have compared some of the cleaned images from other studies with the images obtained by our method, and I am sure the reader would agree that our algorithm produces a much clearer interpretable image.