In cooperation with Lawrence Livermore National Laboratory (LLNL) in the United States we are investigating a new promising technique, named empirical matched field processing, to process mining explosions.
The benefit of this technique is that the steering vector derived from a measured ensemble covariance matrix can have any structure. It needs not correspond to a plane wave, and specifically can capture the effects of refraction and scattering that are the sources of signal incoherence over an array aperture under plane-wave processing assumptions. This observation leads to coherent processing across an array aperture at frequencies higher than possible in conventional procedures.
Detailed map showing individual mines in the Khibiny/Olenegorsk subregion.
We have used the new technique to classify mining explosions in individual mines in this region. The results are very impressive, and exceed by far the classification results using conventional techniques.
It is instructive to view the classification results in terms of a scatter diagram illustrating the marginal problem of classification between just two mines. The figure below shows the matched field processing classification statistics for compact underground explosions in the Kirovsk and Rasvumchorr mines under the two alternative hypotheses that the events are in the Kirovsk mine or the Rasvumchorr mine. The statistics are computed in the high frequency band (7.8125 - 12.5 Hz). They show good separation between the two populations despite the fact that the mines lie on the same backazimuth viewed from ARCES (117.99 versus 118.15 degrees).
We have examined whether the points that appear to be misclassified represent incorrect ground truth event assignments to the two mines. To this end, we requested from the Kola Regional Seismological Centre detailed local seismic recordings for 12 of the events that were most obvious outliers. Using the post-processing algorithm developed by Kværna and Ringdal (1994), we were able to conclude that the ground truth identity of all of these 12 events was indeed incorrect. Thus the technique is reliable enough to be used for quality control of ground truth information.
Populations of Rasvumchorr compact (red crosses) and Kirovsk compact
(black circles) explosions separate well using matched field classification in
the high frequency band shown in this figure (7.8125 - 12.5 Hz). Relatively
few points fall between the two clouds. This observation suggests that the
Kirovsk events that fall solidly within the main Rasvumchorr population
cloud (and vice versa) may have incorrect ground truth identities, and
this has in fact been shown to be the case (see text for details).
|
 |
|
|
|
