Author
 |
YOUN, HYOUNG-SUN - OHIO STATE UNIV. |
|
Submitted to: Ohio State University Thesis
Publication Type: Other Publication Acceptance Date: 11/15/2002 Publication Date: 12/20/2002 Citation: YOUN, H. AUTONOMOUS BURIED PIPE DETECTION USING NEURAL NETWORKS. MASTERS THESIS. 2002. THE OHIO STATE UNIVERSITY. 82 P. Interpretive Summary: Technical Abstract: An autonomous pipe detection algorithm using two independent Artificial Neural Networks (ANN) in two dimensional GPR data has been developed. And a pipe orientation estimation method has been discussed. The first neural network, called step-l ANN, was trained with a waveform reflected from a pipe in the time domain. The second neural network, called step-2 ANN, is trained with the horizontal variation of a hyperbolic arc in two-dimensional GPR data. Once the step-l ANN detects a desired waveform, a temporal-spatial region is determined. Then the step-2 ANN detects the spatial variation in the temporal-spatial region determined by the step-l waveform detector according to the desired spatial variation. The detection performance of each step was estimated by Monte Carlo simulation against simulation data with randomly generated clutter magnitudes and locations. The geometrical characteristic of pipe response in G PR data has also been discussed. An equation of a hyperbolic arc from an orientated pipe was derived from the minimal distance point from the pipes to the antenna. A method to estimate a pipe orientation with this equation and width of pipe's hyperbolic arc in 2D GPR data was suggested and demonstrated with actual field data. Estimating linearity and orientation of pipe by fully polarimetric GPR was reviewed and applied to the pipe detection. |
