FD-TD synthesis and resonance based identification of GPR returns from buried land-mines

Van der Leij, Willem Rienk (1999-10)

Thesis (Msc.Ing) -- University of Stellenbosch, 1999.

Thesis

ENGLISH SUMMARY: Numerical electro-magnetic methods are quickly gaining popularity in the design and optimisation of real world sensors that require complex and expensive testing. Most stepped-frequency GPRs synthesise a defined pulse in the time domain per frequency step. The FDTD method is well suited to modelling pulse propagation and computing scattering information from objects over a wide bandwidth. The Body-of-Revolution expansion to the FDTD method reduces the dimensionality of the algorithm and significantly reduces the computation time and resource load of the simulations. The rotational symmetry requirement still allows the inclusion of a wide variety of land-mine target shapes and stratified media. The BOR FDTD simulator is used to provide time domain data from buried scattering targets modelled to resemble real world conditions as closely as possible. Natural resonance information is extracted from this data using Linear Predictive Coding. Different land-mine targets will have different natural resonance signatures associated with them and these are used as feature vectors for the training and testing of artificial neural networks. Identification is then performed on a set of land-mines in the presence of ground cutter.

AFRIKAANSE OPSOMMING: Numeriese elektro-magnetiese metodes raak toenemendgewild in die ontwerp en optimering van praktiese sensors wat komplekse en duur toetsing vereis. Meeste trap-frekwensie GPR sintetiseer 'n gedefinieerde puis in die tydgebied by elke frekwensiestap. Die FDTD metode is geskik vir die modulering van pulspropagasie en komputasie van veldverspreiding van objekte oor 'n wye bandwydte. 'n Liggaam-van-Omwenteling uitbreiding van die FDTD metode verminder die dimensionaliteit van die algoritme en dus beduidend die komputasietyd en geheue gebruik van die simulasie. Die rotasionele simmetrie vereiste laat nog die modulering van 'n wye stel landmyn teikens en gelaagde mediums toe. Die BOR FDTD simulator word gebruik om tyd gebiedgestrooide data vanaf landmyne te skep wat wereld kondisies so akkuraat as moontlik naboots. Liniere voorspellende kodering word gebruik om natuurlike resonansie informasie uit die data te onttrek. Verskillende landmyn teikens het elkeen hul eie natuurlike resonansie beskrywing en die kenmerkvektore word gebruik vir die ontwerp en validasie van neurale netwerke. Identifikasie op 'n stel landmyne in verskeie grond kondisies is gedoen.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/51271
This item appears in the following collections: