Thermal tomography is a nondestructive method for detecting inhomogene
ities in a material by localizing variations in its thermal conductivi
ty. Based on a finite element discretization of the heat conduction eq
uation, we obtain a set of equations that relate the conductivity of a
medium to temperature measurements on the surface of the medium. We i
nvestigate the use of both a linearization and regularization techniqu
e and a randomized search procedure based on a genetic algorithm to in
vert this set of equations. We found a tradeoff exists between the acc
uracy of the conductivity mapping and the resolution of the conductivi
ty mapping. To increase the resolution of the mapping, we propose a zo
oming method in which the finite elements are grouped into blocks and
a low-resolution mapping of the conductivity is obtained. Improved map
pings are then obtained by increasing the number of blocks in regions
where inhomogeneities appear to be present and repeating the inversion
process.