Identification of metallic mine-like objects using low frequency magnetic fields

Performance Comparison of Automated Induction-Based Algorithms for Landmine Detection in a Blind Field Test

“…Although it has been documented that Eqs. (15) and (20) do not adequately model the EMI response of buried landmines [10,16,22], a simple parametric model that incorporates additional "void" effects has not been proposed. Furthermore, while statistically advanced algorithms such as those used in [6] and [39] may provide improved target response modeling, these algorithms are typically too computationally expensive to be used in real-time humanitarian demining scenarios.…”

“…Some of this work has focused on determining the exact EMI responses from rotationally symmetric bodies in air [9][10][11], as well as simplified phenomenological models that generate similar responses [12]. The characteristic decay rates of time domain signals have also been explored as possible tools for landmine detection [13][14][15][16][17][18]. The wideband EMI spectral responses from different landmines have been shown to be unique [19], and Gao et al have derived the optimal processor for wideband EMI detection for high metal content landmines and have compared its performance to sub-optimal detectors [20].…”

Statistical Classification of Buried Unexploded Ordnance Using Nonparametric Prior Models

“…In theory, the veracity of this model requires illumination of an infinitesimally small target by a uniform field, which is a condition that is not generally met in practice. Moreover, given limited noisy data, at most, a single term in each of the three sums can be estimated reliably [21], leading some investigators to use forms of the model employing only two exponentials [21], [23] on the assumption that targets of interest are spheroidal and, therefore, have two equal eigenvalues. These theoretical and practical limitations of the dipole model impart a position and orientation dependence on the object-specific parameter estimates required for an accurate classification.…”

“…To capture the dependence of the features on the location and orientation of the object, several methods have been proposed in the past. For example, in [23], the authors used Laplace plane poles located on the real s-axis for classification purposes. To account for the fact that the poles differed according to the orientation of the object in space, a single object was regarded as three different targets, one for each of the three orthogonal orientations of the true UXO.…”

“…To account for the fact that the poles differed according to the orientation of the object in space, a single object was regarded as three different targets, one for each of the three orthogonal orientations of the true UXO. Identification for arbitrary orientations and the issue of object localization were not considered in [23]. In [10], instead of a sum of decaying exponentials model, a frequency-domain processing approach was considered.…”