US5373486A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Seismic event classification system
Assignee: United States Department of Energy
•
Filing: 1993-02-03
•
Granted: 1994-12-13
•
Expires: 2013-02-03
Abstract
Abstract In the computer interpretation of seismic data, the critical first step is to identify the general class of an unknown event. For example, the classification might be: teleseismic, regional, local, vehicular, or noise. Self-organizing neural networks (SONNs) can be used for classifying such events. Both Kohonen and Adaptive Resonance Theory (ART) SONNs are useful for this purpose. Given the detection of a seismic event and the corresponding signal, computation is made of: the time-frequency distribution, its binary representation, and finally a shift-invariant representation, which is the magnitude of the two-dimensional Fourier transform (2-D FFT) of the binary time-frequency distribution. This pre-processed input is fed into the SONNs. These neural networks are able to group events that look similar. The ART SONN has an advantage in classifying the event because the types of cluster groups do not need to be pre-defined. The results from the SONNs together with an expert seismologist's classification are then used to derive event classification probabilities.
Inventors
- 👤 Farid U. Dowla
- 👤 Stephen P. Jarpe
- 👤 William Maurer
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-08 16:04
CPC Classifications
G; G01; G01V; G01V1/00; G01V1/003
Claims Analysis
Total Claims: 22
Claim 1
Independent Claim
73 words
A method for automated seismic event classification, said method comprising the steps: detecting a seismic event signal produced from at least one seismic sensor station; pre-processing the detected seismic event signal to produce a pre-processed signal; classifying, using a self organizing neural network (SONN), the pre-processed signal from said seismic sensor station to produce a classified, pre-processed signal; and interpreting, using a knowledge based system, the classified pre-processed signal from said sensor station.
Claim 2
Independent Claim
73 words
A method for automated seismic event classification, said method comprising the steps: detecting a seismic event signal produced from at least one seismic sensor station; pre-processing the detected seismic event signal to produce a pre-processed signal; classifying, using a self organizing neural network (SONN), the pre-processed signal from said seismic sensor station to produce a classified, pre-processed signal; and interpreting, using a knowledge based system, the classified pre-processed signal from said sensor station.
Claim 3
Dependent Claim
27 words
The method of claim 1, additionally including the steps: producing a seismogram from said detected seismic event signal; and computing a translation-invariant binary spectrogram from said seismogram.
Claim 4
Dependent Claim
18 words
The method of claim 2, wherein said translation-invariant binary spectrogram is a two-dimensional fast Fourier transform (2-D FFT).
Claim 5
Dependent Claim
59 words
The method of claim 1, wherein said seismic event is detected by a detection system which continuously computes a ratio of energies by dividing a short term average of said energies by a long term average of said energies and declares an event has occurred when said ratio exceeds a threshold of 2.5 at two or more seismic stations.
Claim 6
Dependent Claim
60 words
The method of claim 1, wherein the pre-processing step is carried out by: producing a seismogram from said detected seismic event signal; computing a time-frequency distribution from said seismogram; transforming said time-frequency distribution into a binary representation; computing a two-dimensional fast Fourier transform (2-D FFT) from said binary representation; and inputting said 2-D FFT into said self organizing neural network.
Claim 7
Dependent Claim
68 words
The method of claim 5, wherein the step of computing a time-frequency distribution from said seismogram is carried out by: computing a power spectrum over many 3 second length time windows; sliding said time windows with 50% overlap for a 400 second segment to form a two-dimensional time-frequency distribution; and formatting said two-dimensional time-frequency distribution into a matrix of 256 by 256 real numbers to form a spectrogram.
Claim 8
Dependent Claim
61 words
The method of claim 6, additionally including the step of noise filtering said two-dimensional time-frequency distribution by the steps of: computing a mean level of a noise signature by averaging a spectral magnitude of an initial 32 of 256 columns of said spectrogram; and subtracting said noise signature from all 256 columns of said spectrogram to form a noise filtered spectrogram.
Claim 9
Dependent Claim
84 words
The method of claim 7, wherein a binary spectrogram is computed from said noise filtered spectrogram by the steps of: computing a mean magnitude value from said noise filtered spectrogram; computing a standard deviation from said mean magnitude value; and performing a threshholding operation wherein all pixel values equal to or greater than said mean magnitude value plus one standard deviation are set to one, and wherein all pixel values less than said mean magnitude value plus one standard deviation are set to zero.
Claim 10
Dependent Claim
64 words
The method of claim 8, wherein a shift-invariant representation is computed from said binary spectrogram by the steps of: representing signal amplitude and phase of said binary spectrogram as a function of frequency to form a one-dimensional Fourier transform; and representing said one-dimensional Fourier transform in two orthogonal directions in terms of its spatial frequency or wavenumber to form a two-dimensional fast Fourier transform.
Claim 11
Dependent Claim
97 words
The method of claim 1, wherein said classifying step is carried out by: forming cluster nodes or groups from said pre-processed signals; topologically or spatially ordering of said clusters or groups; representing each cluster node by a set of weights that is evolved during learning in a training phase of the system to minimize the sum of the distances of the input patterns from their corresponding cluster nodes as represented by the weights; and optimizing learning parameters of said SONN by using 7 clusters with a neighborhood size of 3 and an initial learning rate of 0.3.
Claim 12
Dependent Claim
70 words
The method of claim 10, wherein said classifying step is carried out by: inputting said pre-processed signal into a Kohonen Network; monitoring said Kohonen Network with a matching algorithm; inputting said pre-processed signal into an Adaptive Kohonen Network or an Adaptive Resonance Theory Network if said matching algorithm determines that said Kohonen Network is unable to place said pre-processed signal into one of said cluster groups of said Kohonen Network.
Claim 13
Dependent Claim
117 words
The method of claim 1, wherein said classifying step is carried out by: operating a Seismic Event Classification System (SECS) in a training mode wherein probability estimates for classification categories for a given cluster are found by dividing the number of events in a particular class in a cluster by a total number of events in that cluster while using said SECS with many hundreds of events for which a ground truth is known; and operating the Seismic Event Classification System (SECS) in a classification mode when a new event is detected, wherein said neural network associates said new event into a certain cluster group by matching an input with templates (or weights) of each cluster node.
Claim 14
Dependent Claim
56 words
The method of claim 1, wherein said interpreting step is carried out by: inputting said classified pre-processed signals into a rule-based expert system that identifies seismic phases, phase arrival time, and event phase association thereby producing identities; and applying said identities to a set of if-then rules which use seismology data to interpret a seismic event.
Claim 15
Independent Claim
81 words
An apparatus for automated seismic event classification, said apparatus comprising: means for detecting a seismic event signal produced from at least one seismic sensor station, said detecting means to produce a detected signal; means for pre-processing said detected signal to produce a pre-processed signal; means for classifying using a self organizing neural network (SONN), said pre-processed signal to produce a classified, pre-processed signal; and means for interpreting, using a knowledge based system, said classified, pre-processed signal to produce an interpreted signal.
Claim 16
Independent Claim
81 words
An apparatus for automated seismic event classification, said apparatus comprising: means for detecting a seismic event signal produced from at least one seismic sensor station, said detecting means to produce a detected signal; means for pre-processing said detected signal to produce a pre-processed signal; means for classifying using a self organizing neural network (SONN), said pre-processed signal to produce a classified, pre-processed signal; and means for interpreting, using a knowledge based system, said classified, pre-processed signal to produce an interpreted signal.
Claim 17
Dependent Claim
28 words
The apparatus of claim 14, further comprising an interactive system which receives said detected signal and said interpreted signal and wherein said interactive system produces an event bulletin.
Claim 18
Dependent Claim
42 words
The apparatus of claim 15, further comprising: a local area network (LAN) connected to said detecting means, said interpreting means, and said interactive system; a system monitor connected to said LAN; and an archiver connected to said LAN and said detecting means.
Claim 19
Dependent Claim
12 words
The apparatus of claim 14, wherein said SONN is a Kohonen Network.
Claim 20
Dependent Claim
14 words
The apparatus of claim 14, wherein said SONN is an Adaptive Resonance Theory Network.
Claim 21
Dependent Claim
37 words
The apparatus of claim 14, wherein said SONN is a hybrid comprised of a Kohonen network in combination with a network selected from a group consisting of an Adaptive Resonance Theory Network and an Adaptive Kohonen Network.
Claim 22
Dependent Claim
51 words
The apparatus of claim 14, wherein said SONN is selected from a group consisting of a Kohonen Network, an Adaptive Resonance Theory Network (ART), an Adaptive Kohonen Network (AK) and a hybrid comprised of a Kohonen Network and a network selected from a group consisting of an ART and an AK.
References Cited (Backward Citations)
US5054006A
Cited
US4236233A
Cited
US4783754A
Cited
US4713775A
Cited
US4916615A
Cited
US4811308A
Cited
US4939648A
Cited
US4979124A
Cited
US5146541A
Cited
US5056066A
Cited
US5181171A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1997-12-12 | 1998-06-13 | 1998-12-13 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2001-12-12 | 2002-06-13 | 2002-12-13 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2005-12-12 | 2006-06-13 | 2006-12-13 | $7700.00 | Window Closed / Lapsed |