Strategies for Automated Classification of Seismic Trace Measured by Ocean Bottom Seismic Acquisition Systems

Ocean bottom nodes (OBNs) are a recent technological solution used for seismic data acquisition. Despite of various advantages compared to conventional methods of measurement, the amount of data acquired in OBNs campaigns poses challenges to energy management and data transmission, ultimately limiting the time the device can acquire data on the seabed. To deal with these disadvantages, compression techniques and prediction models have been proposed in the literature and in both approaches the type of trace is an important information. In this work, strategies for developing seismic trace classifier models are assessed aiming to classify seismic traces from ocean bottom nodes into active, passive and microseism. The models were developed based on the machine learning algorithms decision tree and neural networks. Moreover, different features were used in the training process in order to analyze physical quantity dependent and agnostic classifier models. Five different datasets and thousands of traces were used for training and testing the models developed. Models outputs are explored in terms of confusion matrix, accuracy, precision and recall. Results have shown that the use of acceleration and velocity data for classification of microseism and passive traces led to a lower accuracy when compared to the use of sound pressure data. In addition, no relevant difference was found between the decision tree and neural networks for the classification task.