A deep learning approach for automatic recognition of seismo-volcanic events at the Cotopaxi volcano

• Published in: Journal of volcanology and geothermal research 2021-01, Vol.409, p.107142, Article 107142
• Main Authors: Lara, Fernando, Lara-Cueva, Román, Larco, Julio C., Carrera, Enrique V., León, Rubén
• Format: Article
• Language: English
• Subjects: Convolutional neural networks; Deep Learning; Periodogram; Spectrogram; Volcanic microearthquakes
• ISSN: 0377-0273; 1872-6097
• DOI: 10.1016/j.jvolgeores.2020.107142

The research for developing an automatic recognition system of volcanic microearthquakes have been an important task around the world, based on this, the aim of this paper is to present an automatic recognition system of microearthquakes from the Cotopaxi Volcano based on a deep learning approach. The detection and classification stages were carried out with Convolutional Neural Networks by using spectrograms, which were generated according to the theory of periodograms with different types of windows. In order to enable the training of neural networks with a small database (1187 microearthquakes), the Transfer Learning process was used. This system operates in quasi-realtime, which is able to process records of 20min, accordingly to the requirements of the Instituto Geofísico de la Escuela Politécnica Nacional, with a recognition (detection + classification) time response of one minute, approximately. The system performance presents an accuracy of 99% in the detection stage and an accuracy of 97% in the classification stage. •The proposed system performs the recognition of microearthquakes, integrates detection and classification with CNN.•The proposed scheme operates in quasi real time, according to the generation of seismic records of the Cotopaxi volcano.•The system presents an 99% accurate in detection stage of microearthquakes and 97% accurate in classification stage.