Classification of seismic signals using scalogram and wavelet based features

In this paper we propose a method to extract image-based features from the scalogram which represents the energy percentage of each coefficient obtained after applying the Wavelet Transform to seismic signals and then identify the most significant energy levels, which are synonymous of a seismic event. The scalogram graphs were worked by digital image processing, to treat the event as a geometric figure and to extract characteristics of it. In addition, the energy coefficient of each wavelet decomposition level was calculated, as well as the energy contained in each image. These coefficients were also used as features from the seismic event. Finally, a bank of 16 features was obtained, which was evaluated by using three different Machine Learning classifiers, with and without feature selection stage. The results obtained corroborated that the selected scalogram and wavelet based features provide enough discriminating guidelines to classify seismic events with low error rates.