Seismic Signal Analysis Based on Variational Mode Decomposition and Hilbert Transform for Ground Intrusion Activity Classification

Yuan Sun, Dongdong Qian, Jing Zheng, Yuting Liu, Cen LiuSensors 2023 · DOI

TL;DR

This paper introduces a novel method for classifying ground intrusion activities using seismic signals. By combining Variational Mode Decomposition (VMD) and the Hilbert Transform (HT), the researchers extract robust time-frequency features that significantly improve classification accuracy compared to traditional methods. This advancement is crucial for enhancing national security and public safety through more reliable intrusion detection.

What Problem Does This Paper Address?

Ground intrusion activities, such as smuggling, illegal entry, and theft, pose significant threats to public security. While various detection methods exist (infrared, electronic fences, video analysis), they often suffer from limitations related to environmental factors, electromagnetic interference, and detectability. Seismic sensing systems offer a promising alternative due to their non-line-of-sight capabilities and ability to be concealed underground. However, accurately classifying the diverse seismic signals generated by different intrusion types (e.g., vehicles, footsteps, excavations) remains a challenge. Existing methods often struggle with the non-linear and non-stationary nature of these signals, leading to suboptimal classification performance.

Methods

The core of this research lies in a novel time-frequency analysis technique that combines Variational Mode Decomposition (VMD) with the Hilbert Transform (HT).

  • Variational Mode Decomposition (VMD): This is a non-recursive, adaptive signal processing method used to decompose the seismic signals into several Band-limited Intrinsic Mode Functions (BIMFs). Unlike traditional methods like Empirical Mode Decomposition (EMD), VMD is based on a solid theoretical foundation and effectively avoids mode aliasing, ensuring more distinct components.
  • Hilbert Transform (HT): Applied to each BIMF obtained from VMD, the HT is used to extract the instantaneous amplitude, phase, and frequency. This allows for the creation of a three-dimensional Hilbert spectrum, which captures the time-frequency-energy characteristics of the signal.
  • Marginal Spectrum: By integrating the Hilbert spectrum over time, a marginal spectrum is generated. This spectrum is crucial as it represents the cumulative amplitude distribution at each frequency point, providing a more accurate reflection of the actual frequency components present in the seismic signal and exhibiting better noise robustness.
  • Feature Extraction: Three key features are extracted from the marginal spectrum:
  • Marginal Spectrum Energy: Quantifies the energy distribution across frequencies.
  • Marginal Spectrum Entropy: Measures the complexity and irregularity of the signal.
  • Marginal Spectrum Dominant Frequency: Identifies the frequency with the highest amplitude.
  • Classification: A Support Vector Machine (SVM) classifier, specifically using the LIBSVM library with a Radial Basis Function (RBF) kernel, is employed to classify the ground intrusion activities based on the extracted features.

Datasets

The study utilized a self-developed continuous seismic data acquisition sensing system. The dataset comprises seismic signals from five categories:

  • Bicycles
  • Vehicles
  • Footsteps
  • Excavations
  • Environmental Noise

Specific recording distances were defined for each target type to ensure comparable seismic energy. The dataset was divided into training (70%) and testing (30%) sets, with 1500 segmented data samples for each class used for analysis.

Key Insights and Ideas

  • VMD-HT Synergy: The combination of VMD and HT proves highly effective in capturing the intricate time-frequency characteristics of seismic signals, which are often non-linear and non-stationary.
  • Marginal Spectrum's Power: The marginal spectrum, derived from the VMD-HT process, offers a more robust and accurate representation of seismic signal frequencies compared to traditional time or frequency domain analyses.
  • Feature Robustness: The three extracted features (energy, entropy, dominant frequency) from the marginal spectrum are shown to be discriminative and robust to noise, leading to improved classification.
  • Superiority over Existing Methods: The proposed VMD-HT approach consistently outperforms other time-frequency methods like EEMD-HT and EWT-HT, as well as basic time and frequency domain analyses, across various performance metrics.

Key Results

The proposed VMD-HT method achieved an average classification performance of:

  • Accuracy: 99.50%
  • Precision: 98.76%
  • Recall: 98.76%
  • F1-Score: 98.75%

These results are significantly higher than those obtained by time-domain analysis, frequency-domain analysis, EEMD-HT, and EWT-HT methods. For instance, compared to time-domain features, the VMD-HT method showed improvements of 25.25% in bicycle classification accuracy and 14.38% in excavation classification accuracy.

Limitations and Future Work

  • Limitations: The current study primarily focuses on classifying single intrusion events. The paper acknowledges that classifying scenarios with more than two seismic signals appearing simultaneously will introduce greater complexity.
  • Future Work:
    • Investigating the classification of mixed seismic signals from multiple simultaneous intrusion activities.
  • Exploring different observation system arrangements (e.g., cross array, diamond array) to improve positioning accuracy and error analysis.
  • Further research into the positioning accuracy and error of various observation systems.

Summary updated

BibTeX
@article{sun2023seismic,
  author = {Yuan Sun and Dongdong Qian and Jing Zheng and Yuting Liu and Cen Liu},
  title = {{Seismic Signal Analysis Based on Variational Mode Decomposition and Hilbert Transform for Ground Intrusion Activity Classification}},
  journal = {Sensors},
  year = {2023},
  doi = {10.3390/s23073674},
}
Figure