Extracting low signal-to-noise ratio events with the Hough transform from sparse array data

Gil Averbuch, Jelle D. Assink, Pieter S.M. Smets, Läslo G. Evers

Research output: Contribution to journalArticleScientificpeer-review

10 Citations (Scopus)
97 Downloads (Pure)


Low-frequency acoustic, i.e., infrasound, waves are measured by sparse arrays of microbarometers. Recorded data are processed by automatic detection algorithms based on array-processing techniques such as time-domain beam forming and f-k analysis. These algorithms use a signal-to-noise ratio (S/N) value as a detection criterion. In the case of high background noise or in the presence of multiple coinciding signals, the event's S/N decreases and can be missed by automatic processing. In seismology, detecting low-S/N events with geophone arrays is a well-known problem. Whether it is in global earthquake monitoring or reservoir microseismic activity characterization, detecting low-S/N events is needed to better understand the sources or the medium of propagation. We use an image-processing technique as a postprocessing step in the automatic detection of low S/N events. In particular, we consider the use of the Hough transform (HT) technique to detect straight lines in beam-forming results, i.e., a back azimuth (BA) time series. The presence of such lines, due to similar BA values, can be indicative of a low-S/N event. A statistical framework is developed for the HT parameterization, which includes defining a threshold value for detection as well as evaluating the false alarm rate. The method is tested on synthetic data and five years of recorded infrasound from glaciers. It is shown that the automatic detection capability is increased by detecting low-S/N events while keeping a low false-alarm rate.

Original languageEnglish
Pages (from-to)WC43-WC51
Number of pages9
Issue number3
Publication statusPublished - 1 May 2018


Dive into the research topics of 'Extracting low signal-to-noise ratio events with the Hough transform from sparse array data'. Together they form a unique fingerprint.

Cite this