Pattern Recognition Methods For Extracting Information From Geophysical Data Suites

Abstract

The results of geophysical experiments such as reflection seismic surveys, multiple sensor well logging, and monitoring of seismic activity can be organised as pseudo images whose lines consist of the values recorded by particular sensors. Such images consist of a complex but usually nonrandom background and a signal whose nature and cause may be poorly understood. In spite of these difficulties such signals can be extracted by a combination of string to string matching techniques and careful organisation of heuristic knowledge available for the system. In the case of seismic reflection data the seismic traces are represented as a sequence of peaks which are compared from one trace to the next. Detected matches can be collected into seismic events and these events can be associated by clustering techniques with strata. The primitives for well logs are linear segments of the logs. They are converted to a geologic description by a form of codification of contact recognition rules and automated comparison with a type response. Similar approaches may work for the analysis of nonlinear systems such as earthquake generating zones.