Classification of Small Earthquakes and Explosions Utilizing an Enhanced Residual Network

This paper aims to address the challenge of classifying non-natural seismic signals resulting from small earthquakes and artificial blasting activities, which is crucial for refining seismic catalogs and enhancing the accuracy of earthquake warning systems. While traditional methods for identifying source types rely heavily on geophysical features, significant progress has been made in the classification of seismic events using deep learning techniques. Here, we present an enhanced residual network model, ResNet18 with Dual Enhancement (RN18DE), which incorporates two innovative modules: ResSplitConvBlock (RSCB) and DualPoolingAttention (DPA). These modules enhance the accuracy and robustness of the model in distinguishing between natural earthquake signals and artificial blasting signatures. Specifically, the RSCB module improves the network's ability to capture intricate signal details through split convolutional operations, while the DPA module optimizes global feature representations by integrating global pooling and an attention mechanism. Experimental results demonstrate that the RN18DE model outperforms traditional methods and other deep learning models in terms of classification accuracy, achieving an average accuracy of 98.25%. This research not only improves the classification performance but also offers a novel perspective for accurately discriminating between earthquake and blasting signals.