Automatic classification of multi-channel microseismic waveform based on DCNN-SPP

Abstract

The recognition of multi-channel microseismic waveform is important for hazard prediction. This method is widely used to design the corresponding waveform feature of microseismic waveform recognition by hand. However, manual design process is challenging with unsatisfactory classification effect. Thus, this study presents a method with deep convolutional neural network and Spatial Pyramid Pooling (DCNN-SPP) to identify automatically microseismic waveform. We constructed DCNN-SPP to classify multi-channel microseismic waveform. Given the inconsistent channel number of each receiving microseismic signal, we used spatial pyramid pooling to fix the features of multi-channel waveforms. This signal was extracted from the last convolution layer, which can achieve end-to-end training. To extend the dataset, we pre-process raw data by filtering and de-nosing. The experiment shows the practical value of the recognition method with an accuracy rate of up to 91.13%.