Finding Anti-Personnel Mines
in a GPR created image

A. Perrig - Student CS
B. Gros - Assistant LAMI
C. Bruschini - Assistant LAMI
Ph. Mächler - Assistant LAMI


Project of 7th Semester

LAMI, EPFL Lausanne

February, 8th 1996

Abstract:

In this report, we will review methods to treat GPR created images in order to extract relevant image feature information. First, we will discuss image enhancement and filtering techniques. The problem of displaying the data is addressed. Given the filtered image, we will then evaluate different techniques for finding the relevant information about the mines in the treated image.

• Introduction
• Problem Statement
• Requirements Specification
• Introduction to the GPR
• Taking GPR Images
• Image Enhancement and Filtering
  • Requirements
  • Pretreatments
  • Filtering
  • Image Representation and Display
• Finding and extracting information
  • Requirements
  • Fourier Transforms
  • Neural Networks
• Next steps
• Conclusion
• Acknowledgements
• References
• Matlab functions
  • sirload256.m: Reads 8 bit SIR10 DZT files
  • sirload16.m: Reads 16 bit SIR10 DZT files
  • normalize.m: Normalizes a matrix
  • dspim.m: Display images
  • reduire_colonnes.m: Reduce the image size and reduce noise
  • background_removal.m: Removes the background of the image
  • hor_remove.m: Generating the horizontal line removal mask
  • hist_stretch.m: Histogram stretching function
  • frequences.m: Computes the depth of the mine
  • fftshift_1d: Changes the matrix representation of the result of a Fourier Transform
  • hyperbola.m: Generates hyperbolae
  • adjust.m: Computes the linear symetry and adjusts the result to ideally fit our needs
  • gradient_direction.m: Uses 8 different gradient masks to calculate the approximate gradient direction
  • lin_sym.m: Computes the linear symmetry, written by Stéphane Michel
  • neural.m: Generates the training data, trains and tests the neural network
• rap_images.m: Generates all the Matlab pictures shown in this report
• About this document ...