GPR data preprocessing

In the case of GPR signals, A-scans and B-scans were studied. Useful signals, i.e. containing useful information, can be extracted and the signal to noise ratio of objects detected in A-scans or B-scans has been enhanced. In this context, the VUB has developed [11] a method for signal detection in A-scans using hypotheses test (signal is background/signal is useful) after having removed a mean A-scans from the B-scan.
The RUG has been investigating [12] the applicability of multiresolution decomposition techniques to remove the background of GPR images with buried mines. The second image of fig.(11) shows the result after horizontal filtering, which means simple substraction of the average trace from each row. The third and the fourth images are the reconstructed version by the multiresolution scheme with a non-separable filter and a separable filter respectively.

  

Figure 11: GPR B-scan - original image - background removal experiments

In the same order of idea, the RMA has selected matched filters using the wavelet transform for enhancing the target/background contrast. One sees on fig.(12) that a filter around 1 GHz discriminates very well the air/ground interface (top image) and that a filter around 2.2 GHz discriminates successfully the buried AP mine (bottom image).

  

Figure 12: GPR B-scan - separation of ground-air interface from mine response

The analysis of a sequence of C-scans by means of the Karhunen-Loève transform has been done by the VUB [13] and produces also interesting results. Fig. 13 shows the obtained results. On the left side, sample C-scan images of the sequence are presented. On the right side, the four most significant images (with the largest variances) after the Karhunen-Loève transform are presented.

  

Figure 13: GPR: Principal component analysis of a C-scan sequence

The processing of the absolute value of A-scans using the Hilbert transform has been done by the RMA in order to obtain the A-scan envelops and to enhance the resulting C-scans. The results obtained with this method are shown on fig.(14).

  

Figure 14: GPR: enhancement of a C-scan sequence using the Hilbert transform

RMA ([14]) has decomposed A-scans in a linear combination of wavelets. A limited number of wavelet coefficients (typically 5 coefficients) is sufficient to represent A-scans with a good resolution. The selected wavelet is derived from the emitted radar impulse. The results are shown on fig.(15).

  

Figure 15: GPR: representation of A-scans with a limited number of wavelet coefficients