Full text loading...
-
Neural Georadar Probing of Stratified Media
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 5th International Congress of the Brazilian Geophysical Society, Nov 1997, cp-299-00162
Abstract
A problem of much interest is how to use remote probing to characterise the electrical structure of layered media. The characterisation amounts to finding the electrical constitutive parameters along with the thicknesses of the layers which make up the media. A robust method for solving this problem is proposed. The data set is composed of ratios between vertical and horizontal magnetic fields at four frequencies. The transverse electric polarisation is employed. The reason for using the TE polarisation is that the TE surface wave sends more power into the ground, per unit length of the antenna, than does the transverse magnetic surface wave. Providing all the other conditions are equal, there is more subsurface power to be reflected when the TE polarisation is employed, and thus the signal-to-noise ratio of magnetic wave-tilt data tends to be higher than that of electric wave-tilt data. The wave-tilt method has four important qualities. Firstly, the fact that measurements can be made from aircraft makes rapid and efficient reconnaissance mapping of large geographic areas possible. Secondly, since the wave tilt is a ratio of field quantities, it presents increased immunity to noise and model limitations. Therefore, as long as the field intensities at the logging points are above the noise level, they are independent of the fields at the transmitting antenna. Thirdly, the wave-tilt method can be used at low frequencies, where the depth of penetration is of the order of tens of metres. Finally, the magnetic wave tilt is highly sensitive to lateral anomalies. This fact makes the wave-tilt georadar able to detect geologic-fault zones or ore bodies. A neural network solves the inverse problem of stratified media from magnetic wave-tilt data. The network hinders the calculated parameters from being grievously degraded, owing to noise in the data and model limitations. The neural network automates the resolution of the inverse problem. This permits the characteristic parameters of the probed medium to be known in situ, so that a corrective action can be immediately taken for the survey to succeed. The neural inversion bestows the properties of generalisation and abstraction on the subsurface radar. Therefore, interpreting the results becomes easier.