@techreport{elediasc12715,
            type = {Technical Report},
           title = {An Adaptive Learning-by-Examples Methodology for Accurate Crack Characterization in NDT-NDE Problems},
          author = {M. Salucci and N. Anselmi and G. Oliveri and A. Massa},
       publisher = {University of Trento},
            year = {2016},
             url = {http://www.eledia.org/students-reports/715/},
        keywords = {Eddy current testing, inverse scattering, nondestructive testing and evaluation, statistical learning, learning-by-examples, support vector regression, output space filling, partial least squares, adaptive sampling},
        abstract = {This document presents an adaptive learning-by-examples (LBE) inversion strategy for the accurate and real-time characterization of a defect within a planar conductive structure. More precisely, the developed technique exploits a Partial Least Squares (PLS) linear feature extraction strategy in order to compress the relevant information about the underlying relationship between defect and measurements into a small set of predictive features. Successively, an innovative adaptive sampling strategy is exploited in order to collect a set of N input-output (I/O) training pairs such that an even exploration of the PLS-extracted feature space is obtained.  Such a training database is then used to train a Support Vector Regressor (SVR) for building an accurate and robust estimator of the crack dimensions starting from ECT measurements. Some numerical results are shown in order to validate the proposed approach also when a non-negligible amount of noise is superimposed on testing data.}
}