Infrasonic Signal Detection Usin

Document Sample
Infrasonic Signal Detection Usin Powered By Docstoc
					Infrasonic Signal Detection Using
        Hough Transform

D. J. Brown, B.L.N. Kennett, C. Tarlowski
        Research School of Earth Sciences,
          Australian National University,
                 Canberra, 0200,

                                             October, 2002
 •   Motivation
 •   The Hough Transform
 •   Three-Dimensional array processing
 •   ‘INFER’ detector design
 •   Examples
 •   Summary

                                          October, 2002
•   Based on the premise that signal duration may be a significant
     – a signal with an extended duration is at least indicative that ‘something’
       is happening ‘somewhere’

•   Infrasonic detection algorithms with an ‘instantaneous’ detection
    philosophy will likely never operate at sufficiently low thresholds to
    detect all significant signals
     – e.g., a pure Fstat-detector cannot reasonably operate with the detection
       threshold set much below 6.
     – traditionally a threshold is set and signal duration is defined by how long
       the signal stays above threshold.

•   It remains to obtain a correct measure of signal duration

                                                                 October, 2002
                                           •   Acoustic signal from Etna volcano
Motivation                                     recorded at I26DE
Etna Volcano                                    – a significant signal
                                           •   Distinct lack of
                                                – correlation
                                                – amplitude
                                           •   Average Fstat around 4.3
                                           •   Only two integration intervals with
                                               Fstats > 6
                                                – Using a pure Fstat detector,
                                                  measuring signal duration based
                                                  on above-threshold-time-durations
                                                  will yield very short signal durations
                                           •   Signal presence is obvious to the
                                               human eye by the dominant
                                               azimuth and trace velocity

                                                                  October, 2002
               [I26DE, Freyung, Germany]
                                      •   Acoustic signal from shuttle launch
   Motivation                             STS-96 recorded at DLIAR, Los
                                           – a significant signal
                                      •   Distinct lack of
                                           – correlation
                                           – amplitude
                                      •   Average Fstat around 4.3
                                      •   Maximum Fstat ~ 6
                                           – Using a pure Fstat detector,
                                             determining signal duration based
                                             on above threshold time durations
                                             will yield very short signal durations
                                      •   Signal presence is obvious to the
                                          human eye due to the dominant
                                          azimuth and trace velocity

                [DLIAR, Los Alamos]

                                                             October, 2002
 •   Need to design a detection strategy based on the persistence
     of the measured backazimuth during the passage of the signal
      – Certain tools in Pattern Recognition theory may be useful

 •   Envisage a detection algorithm that
      – has two parts:
          • parameter extraction over short time intervals.
          • seeks regions of persistent backazimuth, regardless of any threshold
      – Thresholding
          • since the short duration industrial-type signals are more prevalent than
            the longer duration signals of interest, and generally not as important,
            define a set of signal duration thresholds that may be frequency band-
      – Uses 3D array geometry

                                                                    October, 2002
The Hough Transform
 •   method for doing pattern recognition
 •   extracts parametric curve information from binary pixelated data in the
     presence of noise
 •   consider binary image data:           S  xi , yi  where i  1,, N
 •   apply the transformation:             m  xit  yi
      – maps points in S into lines in parameter space, P
      – all points on the same line in S have the same intersection point in P
 •   Pairs of points (xiyi), (xj,yj) in S are mapped to the point (m,t)in P where
                           y j  yi 
              m  yi  xi 
                           x j  xi 
                        y j  yi 
              t                 
                       x x 
                        j       i 
 •   Problem of detecting spatially extended lines becomes one of finding
     a local maximum
                                                         October, 2002
The Hough Transform

Image Space, S                     Parameter Space,    P

                 Hough Transform

                   dy / dx  2

                                           October, 2002
     The Hough Transform
     Image Space, S                                    Parameter Space,     P


                                     Hough Transform

               [DLIAR, Los Alamos]

                                                            October, 2002
 INFER Detector Design (I)

• Detection philosophy:
      measure accurately:
       • signal duration
       • arrival time
       • backazimuth
      one single detection per phase
       • relieves the burden on down-stream processing
       • does away with the infrasonic ‘coda’ phase

                                                         October, 2002
 INFER Detector Design (I)
• Two stage detection process
      signal parameter estimation in 3D
       • delay-and-sum correlation (in use)
       • global minimization of the mis-match between theoretical and stacked trace
         beam powers (presently testing -see Poster)
       • maximum beam-power by contracting grid (presently testing -see Poster)
      Signal detection via the Hough Transform
       • to prevent spurious associations between pairs of points, define a maximum
         time separation that can exist between any pair of points.
       • define an accumulator mesh whose granularity is commensurate with the
         uncertainty in the measured azimuth.
       • keep track of the features of all integration intervals that contributed to any
         one accumulator cell..

                                                                          October, 2002
  INFER Detector Design (II)
• Phase Identification
      may need to decide phase during source location
      Basic nomenclature:
            phase                     description                           |s| limits
          Iw         Tropospheric-wind ducted phase                           < 340
          Is         Stratospheric ducted phase                             340-380
          It         Thermospheric ducted phase                                > 380

• Thresholding for detection
      signal duration dependent thresholds
       • Fstat
       • STA/LTA
            – take the STA interval to be the entire integration interval
            – take the LTA interval to be the entire time-block requested
            – various norms: L1, L2 integrated power
       • signal duration                                                         October, 2002
 INFER Detector Design (III)
• Thresholds
   – signal duration dependent
   – processing frequency-band dependent

                                           October, 2002
    3D Array Processing
•   Jin Wang (1999) shows that significant
    vertical sensor separations can strongly
     – measured backazimuth
     – magnitude slowness
     – beam power                                 e
•   We extend slowness plane to 3rd
    dimension: (sx,sy,sz)
•   Create a mesh in the 3D half-space
                                                      |s| = 1/Cs
•   Consider only those points that lie in an
    e thick dome centered on the acoustic
•   [Recently] Use e = 0 !
     – Use Cs determined by
         • Value of Temperature recorded at the
         • climatological model                                    October, 2002
         • Constant = 330 m/s
Shuttle launch signal -DLIAR

                               October, 2002
 Gas pipeline explosion -DLIAR

                                 October, 2002

          October, 2002

          October, 2002

          October, 2002

Seattle Earthquake -I10CA

                            October, 2002
Acapulco Bolide -I59US

                         October, 2002
Etna volcano -I26DE

                      October, 2002
• Detecting infrasound signals from IMS array data by seeking
  regions of constant backazimuth seems to be a robust
   – backazimuth can be accurately determined
   – signal duration can be accurately inferred
• The Hough Transform can be used quite effectively to look for
  regions of constant backazimuth
   – signals with moderate to small Fstat can be detected
• An automatic signal detection algorithm called INFER based on
  the Hough Transform has been created
   – uses 3D array geometry
   – uses correlator to do feature extraction
       • testing other procedures
   – incorporates basic phase id
   – interfaces dynamically with the CSS database tables    October, 2002