Attention-Based CNN for Micro Earthquake Classification in Low SNR
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing earthquake detection methods, such as STA/LTA, autocorrelation-based, and template matching, face challenges in classifying micro earthquakes from noise and artificial events in low SNR environments, leading to inefficiencies and inaccuracies in seismic data analysis.
Innovation Solution
An attention-based neural network method that preprocesses earthquake data, extracts feature maps through multiple convolution layers, measures feature importance using attention techniques, and performs down-sampling to classify earthquake events accurately, even in noisy conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional earthquake detection methods (STA/LTA, autocorrelation, template matching) are used, then detection speed and computational efficiency are maintained, but classification accuracy of micro earthquakes from noise and artificial events deteriorates in low SNR environments
Solution Approach 1:
The patent segments the earthquake detection process into distinct functional modules: data acquisition module, preprocessing module (including normalization and filtering), feature extraction module (using CNN to extract temporal and spectral features), classification module (using attention-based LSTM to classify event types), and post-processing module. This segmentation allows each module to optimize for its specific function, improving overall classification accuracy while managing computational complexity.
Solution Approach 2:
The patent introduces an attention mechanism as an intermediary between feature extraction and classification. The attention mechanism selectively weights different time steps and features, allowing the system to focus on discriminative patterns while suppressing noise. This intermediary layer enhances the model's ability to distinguish micro earthquakes from noise and artificial events by dynamically emphasizing relevant features.
2Measurement precision
If the number of templates is increased to improve template matching performance, then detection accuracy improves, but computational overload increases
Solution Approach 1:
The patent replaces the traditional template matching mechanical process with a deep learning-based feature extraction and classification system. Instead of comparing input signals against multiple stored templates through correlation operations, the system uses CNN layers to automatically extract features and an attention-based LSTM to classify events. This substitution eliminates the need for extensive template libraries and their associated computational overhead while maintaining or improving detection accuracy.
Solution Approach 2:
The patent changes the fundamental parameters of the detection approach by transitioning from template-based correlation (which requires multiple templates) to feature-based classification. The system extracts temporal and spectral features through CNN layers and uses attention mechanisms to weight important features, achieving accurate classification without requiring a large number of templates. This parameter change fundamentally reduces computational complexity.
3Measurement precision
If deep learning models with many parameters are used to extract features, then classification performance improves, but model complexity and training time increase
Solution Approach 1:
The patent introduces dynamic attention mechanisms that adaptively weight features based on their relevance to the current classification task. The attention scores are computed dynamically during inference, allowing the model to focus computational resources on the most informative features for each specific earthquake event. This dynamic adaptation improves classification performance without requiring a fixed increase in model parameters.
Solution Approach 2:
The patent processes earthquake signals by transforming them into multiple feature dimensions through CNN layers, including temporal features, spectral features, and their combinations. The attention mechanism then operates across these dimensions, selectively weighting important features. This dimensional transformation allows the model to capture complex patterns without requiring excessively deep networks, as the information is distributed across multiple feature dimensions rather than requiring increased network depth.
Data Source
AI summary
An earthquake event classification method using an attention-based neural network includes: preprocessing input earthquake data by centering; extracting a feature map by nonlinearly converting the preprocessed earthquake data through a plurality of convolution layers having three or more layers; measuring importance of a learned feature of the nonlinear-converted earthquake data based on an attention technique in which interdependence of channels of the feature map is modeled; correcting a feature value of the measured importance value through element-wise multiply with the learned feature map; performing down-sampling through max-pooling based on the feature value; and classifying an earthquake event by regularizing the down-sampled feature value. Accordingly, main core features inherent in many/complex data are extracted through attention-based deep learning to overcome the limitations of the existing micro earthquake detection technology, thereby enabling earthquake detection even in low SNR environments.