An Approach to Characterising Ground Probing Radar Target Echoes For Landmine Recognition.
Abstract
This report investigates an approach to characterising Ground Probing Radar (GPR) backscatter echoes from land mines using linear combinations of exponentially damped sinusoids. The GPR signatures of surrogate land mines and PVC cylinders buried in dry sand are measured using an impulse radar system with center frequency of 1.4 GHz and a 90% bandwidth. The GPR signal parameters are represented as sets of complex poles computed from a series of neighbouring signatures recorded over each target type. The algorithm proposed by Kumaresan and Tufts which uses backward linear prediction and the low-rank data matrix approximation based on singular value decomposition is applied to this computation. The performance of the Kumaresan and Tufts (KT) algorithm is compared with that of the Prony method when both techniques are applied to modelling simulated signals. It is concluded that the KT method provides more stable pole estimates. Two approaches to determining the order of the model are examined and compared for simulated and real data. The results show that the poles corresponding to different target types form clusters in the two-dimensional alpha-f space (where a is the pole damping factor and f is the pole frequency). This indicates that these pole clusters can be used for the recognition of land mines.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1998
- Accession Number
- ADA356181
Entities
Organizations
- Defence Science and Technology Group