Enhanced Event Detection with Seismic Listening Sensors

Physical Sciences Inc. (PSI) proposes to develop innovative statistical data processing and analysis methods for detecting anomalous urban seismic activities. A five-step process is employed: 1) the vibration signal is segmented by a sliding-window and its noise reduced; 2) wavelet and spectral analyses detect nonstationary and cyclic patterns, respectively; 3) statistical properties such as mean, variance, and energy of each set of coefficients, in addition to the frequencies corresponding with the largest coefficients of the spectral analyses, form primary feature arrays (PFAs). Principal Component Analysis (PCA) applied to PFAs reduces their size and yields secondary feature arrays (SFAs). Classification of the SFAs using a classification fusion of K-means and self–organizing feature maps provides cluster centers that represent the SFAs of typical vibration activities; 4) the cluster centers are used as the states of a Markov chain with memory. Low probability state transitions are translated as anomalous vibration activities and will be reported; 5) sequential F-statistics will be used to find event locations. Performance of the algorithms (Pd vs. Pfa) will be investigated by adding different noise levels and anomalous vibrations to the signal and we expect Pd>90% and Pfa<1%.