Graph-Based Feature Weight Optimisation and Classification of Continuous Seismic Sensor Array Recordings.

Area of Science:

• Geophysics
• Seismology
• Earthquake Engineering

Background:

• Slope instabilities pose significant risks, often characterized by seismic events.
• Effective classification of seismic signals is crucial for mitigating damage and mortality.
• Limited research exists on feature selection for classifying seismic signals from noisy, multi-channel recordings.

Purpose of the Study:

• To develop a novel multi-channel seismic event detection scheme.
• To enhance seismic signal classification using graph-based feature weight optimization.
• To improve the understanding of feature selection's importance in seismic signal analysis.

Main Methods:

• A multi-channel event-detection scheme based on Neyman-Pearson lemma and Multi-channel Coherency Migration (MCM).
• Graph-based feature weight optimization for feature selection, utilizing signal physical characteristics.
• Alternative optimization of feature weights and classification labels using graph smoothness and semidefinite programming (SDP).

Main Results:

• The proposed detection method identified 614 more seismic events over five days compared to the conventional STA/LTA approach.
• Feature selection reduced the number of features by more than half while enhancing classification performance.
• Graph Laplacian Regularisation classifier (GLR) sensitivity improved for rockfall (92%) and slide quakes (88%) after feature selection.

Conclusions:

• The novel multi-channel detection scheme significantly improves seismic event identification.
• Graph-based feature weight optimization is effective for selecting focused feature sets and enhancing classification accuracy.
• This research contributes to better seismic event classification for slope stability monitoring and hazard assessment.