Identification of Mine Explosions Using Manifold Learning Techniques

Itay Niv, Yuri Bregman, Neta Rabin*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Scopus citations


In many regions of the world, mine activity accounts for the bulk of the total seismicity. Therefore, automatic identification of repeating mining events would significantly reduce the analyst's workload at many seismological organizations. This work focuses on the problem of identifying mine blasts that originate from a specific mine, given a stream of seismic data. The mine pattern to identify is complex, as the mine spans several kilometers and the recorded blasts may be of a different type, magnitude, and location. Identification of such single blasts from ongoing recordings that contains diverse seismic events and noise is a challenging task, which is typically not executed in a fully automatic manner but is rather carried out by manual analysis. In this work, manifold learning techniques, which faithfully model high-dimensional data into a compact, low-dimensional space, are utilized. In particular, a modification of diffusion maps is applied for creating a compact representation of a large set of mine patterns, bypassing the need to select a representative subset of patterns. This representation is stable with respect to the changing time streams that hold new seismic data. In addition, manifold-based data fusion techniques are implemented for processing the three-channel seismic recordings. The capabilities of the model are presented on several different datasets and compared with deep learning techniques. The results highlight the method's computational and explainable proprieties.

Original languageEnglish
Article number5912313
JournalIEEE Transactions on Geoscience and Remote Sensing
StatePublished - 2022


  • Diffusion maps (DM)
  • manifold learning
  • mine explosions
  • pattern recognition
  • seismic signal processing


Dive into the research topics of 'Identification of Mine Explosions Using Manifold Learning Techniques'. Together they form a unique fingerprint.

Cite this