The detector algorithms in use at date rely on negative decision logic: based on a model of the ambient noise process they detect all deviations, but many of them are false alarms. The principal alternative to this approach is pattern recognition, which tests on positive correlation with some known signal patterns. 

The Sonogram-detector realizes this scheme for single seismogram traces. Sonograms display spectral energy versus time. Suitably scaled, these images display only information which is significant to the detection process. Patterns of known earthquakes and noise signals are defined by means of these images. Event detection is performed by recognizing one of the patterns in the actual sonogram. The overall processing scheme is similar to the visual inspection of seismograms by the human observer. 

An off-line test installation for detecting local earthquakes proves the expected low false alarm rate, high timing accuracy and good detection probability of the Sonogram-detector.

Pattern recognition for earthquake detection

Automatic detection of earthquakes in environmental noise is subject to research for more than 25 years. The traditional approaches describe the stationary noise level by some kind of median and standard deviation and determine a detection threshold [1],[2]. Each temporary signal with energy above the threshold triggers a detection. Bandpass- or ARMA-filtering is used to improve the signal to noise ratio. The main disadvantage of these detectors is their high false alarm rate, since they are not able to distinguish between small earthquakes and noise from traffic or industry, which has the same or even higher amplitude. This is due to their decision logic of negative kind: Every signal above the threshold is assumed to be caused by an earthquake.

/pdf/pattern-recognition-for-earthquake-detection-4u2fwqjy02.pdf

Pattern Recognition for Earthquake Detection

Applications of clustering in exploration seismology

A novel methodology for generation of artificial earthquake precursors was tested on Southern California earthquake data in reverse and real time modes. When it was tried as a real time generator of earthquake precursors, it successfully predicted the June, 1992, Landers earthquake. The methodology is based on the use of adaptive neural nets (ANN) that process a set of time-dependent attributes calculated in a moving time-window. The most important of them is a danger function. The structure of the neural net is defined by the properties of input data in the moving time window. Thus, the neural net continuously adapts its structure to the time variant properties of the input attributes. The main problem the authors encountered in training the neural net on the earthquake data was the small size of the training set compared to the number of parameters that describe the structure of the ANN. To prevent instability and over-fitting in the training session, the authors used a technique similar to the damping method in least squares approximation. >

Adaptive neural nets for generation of artificial earthquake precursors

Reliable seismic waves characterization is essential for better understanding wave propagation phenomena, providing new physical insight into soil properties. Many works in this area have been based on detecting special patterns or clusters in seismic data, event detection using parametric models and time-frequency analysis. In this paper we present a new approach making use of the short-term time-frequency Renyi entropy and maximum a posteriori probability (MAP) estimator, operating on time-frequency Renyi entropy, as a new space of decision; this method enables more robust feature extraction and a more accurate classification. This approach was used in simulation and in the analysis of the earthquake records during the Kocaeli, Arcelik seism, Turkey, August 1999, a strong to moderate ground motion.

C.H. Chen

Papers

Seismic pattern recognition

On a segmentation algorithm for seismic signal analysis