Advancements in seismic data collection and analysis through machine learning

Abstract

The evolution of seismic data collection has been driven by the need for stations to capture large volumes of high-frequency signals continuously. These signals typically contain both seismic and non-seismic information. Previous research converted SEED data into CSV format and used principal component analysis (PCA) for feature extraction from the seismic dataset. Machine learning models were then employed, showing an improvement in identifying seismic and non-seismic events. This paper focuses on applying deep learning methods, specifically deep neural networks (DNN) and a hybrid model combining long short-term memory (LSTM) networks with DNN (LSTM+DNN). The proposed deep learning models demonstrate a notable improvement over traditional machine learning technique. Experimental results show a test accuracy of 99.24% using deep learning, compared to an average of 97.80% achieved with machine learning models, indicating a 1.46% enhancement in detection accuracy. This underscores the potential of deep learning in accurately detecting seismic events in real-time monitoring systems.