Seismic buffer recognition using wavelet based features

Valid buffer recognition is an important aspect of seismic signal processing systems, since it differentiates useful information generated by seismic activity from data resulting fi om other causes. Since seismic buffers are characterised by a combination of their time and frequency behaviour, wavelets are natural candidates to capture discriminating features. This paper evaluates different approaches to invalid buffer recognition using wavelet-based feature extraction. The most consistent technique appears to be statistical moments extracted from the envelope of the so-called Single Wavelet Transform. The mutual information criterion was used to reduce the number of features to improve generalisation abilities of the neural classifier.