DETECTION OF BRIGHT SPOTS IN SEISMIC SIGNAL USING PATTERN RECOGNITION TECHNIQUES
Abstract
Bright spot is one of the pattern classes in a seismic section and the indicators of gas (hydrocarbon) accumulation. In the past, detection of bright spots depended primarily upon visual examination and the experience of a geophysicist. It is this writer's contention that bright spot detection could be made more confidently by computer-aided analysis. This study concerns with two computer-aided methods. One is the decision-theoretic pattern recognition, the other is the syntactic or structural pattern recognition. Using these two methods, a seismogram is classified into two classes, i.e., bright spot and non-bright spot. In the decision-theoretic pattern recognition, three features from seismic traces extracted, envelope, instantaneous frequency, and polarity. Hilbert transform theorem plays an important role in the analytic signal analysis. Linear and tree classification techniques are applied. The classification result provides candidate bright spots. The second approach to detect candidate bright spots is the utilization of the syntactic pattern recognition technique. Tree classification is used to extract the pattern wavelets of bright spot. Structural information of the pattern wavelets of a bright spot is used, as are Levenshtein distance computation and nearest-neighbor decision rule. A threshold is determined from error probability calculation and is used to detect candidate bright spots. Another factor affecting the detection of bright spot is frequency attenuation. A "partitioning-method" is presented. A seismogram is partitioned into small sections and the tree classification is performed in each section to detect the candidate bright spot. After the candidate bright spots are determined, syntactic pattern recognition technique is used again to recognize the string representation of a bright spot in a two dimensional seismogram. The final result will indicate a bright spot or non-bright spot. It is common in seismic signal analysis to use the zero-phase Ricker wavelets. This study also utilizes these patterns in the simulation to test the proposed techniques as they are applied to the relative-amplitude real seismogram. The classification results obtained from these computer-aided methods can be used to improve seismic interpretation.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.