Fig. 2
Summary of the adopted processing workflow, integrating data preprocessing, feature extraction, and unsupervised clustering via Self-Organizing Maps (SOMs). (a) An example of a 15-minute continuous waveform recorded at V0102 station is shown, from which individual 1-minute windows are extracted (highlighted in the red box). (b) For each 1-minute waveform, three features are computed: Linear Prediction Coefficients (LPC), STA/LTA ratio, and Multiscale Entropy. These features were selected to capture the spectral, energy, and signal complexity, respectively. (c) The extracted features are used to construct the input vectors for the SOM training. The resulting SOM map (top) illustrates the organization and density of similar waveform patterns, while the clustering index (bottom) quantifies the temporal variations in the feature distribution.
