Research on Seismic Signal Classification and Recognition Based on EEMD and CNN

The accurate classification of seismic signals is of great significance for updating clear earthquake catalog, early-warning the arrival of strong earthquake, and further researches in the field of seismology. This paper first preprocesses the seismic signal data of two types of events: earthquakes and explosions, and then uses the ensemble empirical mode decomposition (EEMD) method to decompose these preprocessed seismic data. A set of intrinsic mode functions being distributed from high-frequency to low-frequency are obtained from each seismic waveform, and then corresponding Hilbert spectrum is obtained by applying Hilbert transform to each of these intrinsic mode functions (IMFs). The image of Hilbert spectrum is scaled to a suitable and comparable size, by repeated trial-and-error experiments, to the size of 32*32 pixels, and then the 32*32 pixels gray-scale image is fed to a convolutional neural network(CNN) to classify the types of seismic signals. In this paper, 1674 waveforms of 54 earthquake events and 1509 waveforms of 63 explosion events are processed as above procedures. By five-fold cross-validation, 80% of the data is used as the training set, remaining 20% data as testing set. The highest classification accuracy rate of testing set is 94.98% with the average accuracy rate 94.41%, this results are significant higher than those of multi-layer perceptron(MLP) and Support Vector Machine(SVM), which implying stronger classification capability of the scheme unifying EEMD and CNN proposed in this paper.