Detection of bright spots in seismic signals using tree classifiers

A tree classification method is proposed for the detection of bright spots in seismic signals. Envelope, instantaneous frequency, and polarity are used as the features. From the analysis of zero-phase Ricker wavelets, a tree classification technique is adopted. Bright spots are detected if one of three hypotheses is satisfied. The advantage of using a tree classifier is that it is easy to design and computationally efficient. Because of the frequency attenuation effect in a seismogram, a partitioning-method is introduced in conjunction with the tree classification for the detection of bright spots. Results from simulated and relative-amplitude real stacked seismograms are presented. The classification results show that the proposed method can be used to improve seismic interpretation.