A Semi-Supervised Approach for Microseisms Classification from Cotopaxi Volcano

Autoencoders as a Characterization Technique and Aid in the Classification of Volcanic Earthquakes

“…The Bayesian neural network (BNN) [6], which is used in the PICOSS platform [36] achieves a performance of 92%, in which the LFCC technique is used to characterize the seismic signals. In the same way this performance is achieved using LPC, LFCC, PCA and MFCC signal characterization techniques in [2], [4], [6], [10], [15], [16], [38]. This work demonstrates that the use of a DAF is a good technique for characterizing volcanic earthquake signals, thus assisting in the classification of seismic records without the need to address complex architectures that require higher computational cost and achieving satisfactory results.…”

“…As is known, there are large amounts of seismic log data acquired by in-situ sensors, however, that require labeling in order to fulfill their purpose of long-term monitoring and interpretation of internal volcanic activity. The results showed that conventional preprocessing techniques applied on volcanic earthquake signals could be improved (LPC and PCA) [2], [16], [24], [38]- [43]. The representation of data by means of resource transformation using methods that are usually successfully applied to signals that are similar to seismic signals, such as speech signals, are not always compatible [7], [15], as is the case with widely used algorithms such as MFCC, LFCC, LPC, PCA, among others.…”

“…It is about transforming the signal domain to a space where the information is less sparse. This approach has been popular in the literature in the last two decades for the characterization of volcanic earthquake signals as part of the pre-processing of the data, among the most common we can mention the Fourier transform, Hilbert transform, wavelet transform, the logarithmic frequency cepstral coefficients (LFCC) [6], [8], [35]- [37], Mel-scale frequency cepstral coefficients (MFCC) [38], Linear Prediction Components (LPC) [2], [15], [24], [39]- [41], Principal Component Analysis (PCA) [16], [38], [42], [43], and among other nonlinear variations derived from the previous ones. This work will address two of the conventional techniques, LPC and PCA, as a basis of comparison for the proposed characterization technique in the preprocessing of seismic volcanic signals using Autoencoders, to transform the signals to a feature subspace as a support in the classification, by reducing the dimension of the data, eliminating redundant information and as a solution when the dataset presents an unbalanced profile.…”

ESeismic-GAN: A Generative Model for Seismic Events From Cotopaxi Volcano

“…They have proved to be successful tools (as a second opinion) for analyzing data in various fields of study, including volcanic seismology. Some examples of MLC applications in the volcano seismic event classification context have been developed from supervised learning models such as artificial neural networks [3], [4], deep neural networks [5], [6], support vector machine (SVM) [7], [8], random forest [9] decision trees [10], Hidden Markov Model (HMM) [11], [12], evolutionary algorithms [13], [14] and Gaussian mixture models (GMM) [15] to other approaches based on unsupervised learning [16], [17] and semi-supervised learning [18].…”

“…Traditional supervised learning models require a certain amount of labeled data for adequately learning the feature space during the training stage and then classify unseen data [9], [18]. However, the number of samples to train and test state-of-the-art MLCs as deep-learning models is a big concern due to data limitations such as insufficient amount of highquality instances in the training data, missing labels, and the imbalanced representation of classes, among others.…”