A neural network based small seismic event detector

Small seismic events are abundant and can provide a lot of information on the subsurface. The ability to automatically detect small seismic events in an accurate and efficient way is of great interest for the study of geological structure and fault process. Small seismic events are difficult to detect using conventional methods as the amplitudes of signals are relatively small. Here we present an efficient seismic event detector to extract small seismic signals from continuous records. Our algorithm is based on a convolution neural network model. Features are learned from spectrograms of signals and then used to classify windows of seismic records. We demonstrate the performance of our algorithm using dataset from Oklahoma, where seismicity has increased dramatically over the last decade. We show that our algorithm is efficient at detecting small seismic events, resulting in a much more comprehensive catalog compared with the Oklahoma Geological Survey (OGS) catalog.