Threshold estimation using wavelets and curvelets on ground penetrating radar data for noise and clutter suppresion

This thesis provides an evaluation of the Redundant Discrete Wavelet Transform with application to the removal of additive white or colored Gaussian noise on a synthetic GPR signal. Special attention is given to the parameter that controls the number of decomposition levels. Evaluation is performed using a level-dependent threshold to estimate and remove noise. Results are presented using noisy synthetic Ground Penetrating Radar pulses to compare Wiener filtering and thresholding the Redundant and Non-redundant Discrete Wavelet transform. Additional results are presented on the effects of choosing a number of decomposition levels using signal-to-noise ratio measurements, which suggest the importance of choosing this parameter. Recommendations are made and supported which determine the order of thresholding before or after the practice of trace averaging. Using GPR images, an application of a novel 2D threshold model in the newly discovered curvelet domain is compared to average trace subtraction. Promising results are presented on both synthetic and actual landmine data, which shows thresholding as a viable method of clutter suppression.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat