کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
4740413 | 1641166 | 2012 | 9 صفحه PDF | دانلود رایگان |
![عکس صفحه اول مقاله: Regularizing dipole polarizabilities in time-domain electromagnetic inversion Regularizing dipole polarizabilities in time-domain electromagnetic inversion](/preview/png/4740413.png)
Recent advances in time-domain electromagnetic (TEM) sensors have dramatically improved discrimination of buried unexploded ordnance (UXO). In contrast to commercial standard mono-static sensors, the multi-static, multi-component geometries of next generation TEM sensors provide diverse excitations of a detected target. Inversion of observed data using the parametric TEM dipole model typically produces well-constrained estimates that can subsequently be inputted into a discrimination algorithm. In particular, the principal dipole polarizabilities provide information about target size and shape. Shape is represented by two transverse polarizabilities orthogonal to a target's axis of symmetry.Equality of transverse polarizabilities is diagnostic of an axisymmetric body of revolution and so has been proposed as a useful feature to discriminate between axisymmetric UXO and non-axisymmetric metallic clutter. Here we show that estimated transverse polarizabilities can sometimes be poorly constrained in an inversion of multi-static TEM data. This motivates our development of a regularized inversion algorithm that penalizes the deviation between transverse polarizabilities. We then develop an extension of the support vector machine (SVM) classifier that uses all models obtained via regularized inversion to make discrimination decisions. This approach achieves the best performance of all candidate discrimination algorithms applied to a number of real data sets.
► We consider discrimination of unexploded ordnance using multi-static EM sensor data.
► We develop a regularized inversion algorithm for estimation of dipole polarizabilities.
► Models from regularized inversion are inputted into a support vector machine (SVM).
► The kernels of the SVM are weighted by the likelihood of each regularized model.
► SVM with regularized models outperforms a baseline algorithm for all data sets.
Journal: Journal of Applied Geophysics - Volume 85, October 2012, Pages 59–67