Resolution and robustness to noise of the sensitivity-based method for microwave imaging with data acquired on cylindrical surfaces

Abstract

The spatial resolution limit of a Jacobian-based microwave imaging algorithm and its robustness to noise are evaluated. The focus here is on tomographic systems where the wideband data are acquired with a vertically scanned circular sensor array and at each scanning step a 2D image is reconstructed in the plane of the sensor array. The theoretical resolution is obtained as one-half of the maximum-frequency wavelength with far-zone data and about two-thirds of the array radius with near-zone data. Validation examples are given using analytical electromagnetic models. The algorithm is shown to be robust to noise when the response data are corrupted by Gaussian white noise.

Fast quantitative microwave imaging with resolvent kernel extracted from measurements

“…The power map bears similarity to the 'sensitivity map' in sensitivity-based imaging [21,22]-it can be viewed as the sensitivity map scaled by the permittivity contrast of the small scatterer in the CO. In [21,22], however, the sensitivity maps are built using incidentfield information acquired from simulations or semi-analytical models, which are not likely to possess sufficient fidelity for near-field measurements.…”

“…To this end, the power maps yield only qualitative imaging results, i.e., there is no property estimation involved. With wideband data, the OUT power maps at all frequencies are usually combined (often after proper normalization [21,22]) to obtain a single qualitative image of the OUT.…”

“…It has been attempted to reconstruct the OUT by using a simulated resolvent kernel, computed from the incident field distributions [21,22] associated with each of the antennas in the simulation model of the experiment. This attempt failed, i.e., the obtained power maps did not show the targets.…”

“…The qualitative reconstruction methods of active microwave imaging include holographic [13][14][15][16][17], confocal [18][19][20], sensitivity-based [21,22], and time-reversal [24][25][26] methods. Since these methods employ linearized scattering models, typically based on the Born approximation [27], they do not allow for quantitative reconstruction.…”

“…This is rarely valid in the near zone of the transmitting and receiving antennas. In [15][16][17] and [21,22], this limitation is overcome by making use of simulation models of the incident field and the Green function distributions. Unfortunately, the fidelity of the simulation models, although better than the analytical approximations, may also be questionable for the same reasons as the simulations used in the optimization-based quantitative reconstruction methods.…”