Syntactic pattern recognition for classification of ricker wavelets

A seismic trace includes three kinds of zero-phase Ricker wavelets in our investigations. For poor separability, threshold selection in the decision-theoretic pattern recognition is not easy. So the approach of syntactic pattern recognition is proposed. The block diagram of a syntactic pattern recognition system is shown in Figure 3. In the primitive recognition, Freeman's chain code is used, i.e., amplitude-independent code and amplitude-dependent code. String encoding converts a seismic trace into a sentence, i.e., a string of primitives. A peak recognition algorithm can be applied to detect positive and negative peaks. From these peaks, Ricker wavelets can be extracted. Three syntactic methods are presented for the classification of Ricker wavelets. They are finite-state automaton, minimum-distance classification, and nearest-neighbor (NN) classification. In order to carry out the minimum-distance and the NN classifications, string distance is computed by Levenshtein distance which is the minimum number of symbol insertions, deletions, and substitutions required to transform one string into the other string. This distance can be computed by using dynamic programming. There are two and three-class classifications and the results are very encouraging.