Deep Multi-scale Dilated Convolution Neural Network with Attention Mechanism: A Novel Method for Earthquake Magnitude Classification

The magnitude of an earthquake influences the amount of damage and casualties it causes in any society. Although reliable and timely prediction of magnitude earthquakes can help reduce damage, it requires advanced algorithms to extract relevant information from raw seismic data. Because of its high strength in better representation of features, Convolutional Neural Network (CNN) has demonstrated success in earthquake prediction among deep learning algorithms. The features of different received fields obtained by various sizes of convolution kernels can give comprehensive seismic information. Conventional CNN uses a one-size convolution kernel to learn complete information from complex inputs. Furthermore, conventional CNNs have a limited capacity to learn local and global features simultaneously. To address the issues above, we suggested a novel strategy based on a multi-scale dilated convolutional neural network with an attention mechanism (MSDCA). While the multi-scale module automatically obtains the rich features of different time scales, the attention mechanism decides the weight of distinct scales, improving the model’s dynamic adjustment performance and adaptable capacity. The purpose of this study is to classify the magnitude of Himalayan earthquakes into different categories based on their importance. The outcomes and comparisons with state-of-the-art techniques demonstrate the proposed method’s superiority.