Pattern Analysis and Classification with the New ACDA Seismic Signature Data Base.

Abstract

The new and expanded ACDA seismic data base makes it possible for meaningful comparison of different seismic recognition techniques based on the same data set. There are 157 earthquake and 157 explosion records in the data base. Pattern analysis in frequency domain as well as two-dimensional space is performed to seek for classification clues. Although useful structure of the seismic records is not available, the mathematical features provided by the autocorrelation function have 86.36% correct recognition on testing set by using 3 features (autocorrelation coefficients), 80 selected good quality training samples per class and the nearest-neighbor decision rule. All samples in the training set are identified correctly and thus the overall recognition rate of 93.00% is achieved. This result is better than the 89.2% recognition using dynamic spectral ratios (Table 7). The autocorrelation coefficients which are simple to calculate also perform better than the linear predictor (Markel) coefficients and other discriminants.

Document Details

Document Type
Technical Report
Publication Date
Aug 12, 1975
Accession Number
ADA015925

Entities

People

  • Chia‐Hung Chen
  • I. Chang Lin

Organizations

  • University of Massachusetts Dartmouth

Tags

DTIC Thesaurus Topics

  • Autocorrelation
  • Classification
  • Coefficients
  • Data Sets
  • Databases
  • Frequency
  • Frequency Domain
  • Recognition
  • Seismic Signatures
  • Training
  • Two Dimensional

Fields of Study

  • Engineering

Readers

  • Regression Analysis.
  • Seismology
  • Sensor Fusion and Tracking Systems.

Technology Areas

  • Space
  • Space - Space Objects