Discrimination of Blasts
Significant portion of the recorded seismic data may represent seismic signals from development or production blasts. Such artificial events must be identified and excluded from certain types of analysis (e.g. assessment of seismic hazard).
Multivariate maximum-likelihood Gaussian discrimination technique makes it possible to classify the recorded and processed seismic events. Initially a set of discriminating characteristics is calibrated using a training dataset (collection of blast and collection of normal events). Four of such characteristics were found to be effective in mines:
- time of day;
- waveform similarity with the neighbouring events;
- ratio between high- and low-frequency seismic radiation;
- radiation pattern.
Application of the calibrated technique to an unclassified event results in a number expressing the probability that the event belongs to a population of blasts (or normal events). Updating of the training dataset and re-calibration of the discriminating parameters may be required in the course of routine application of the technique (e.g. discovery of the misclassification cases).