MSFFCN: A Multi-Scale Feature Fusion Network for Rapid Intensity Classification of Ground Vibration Signals

Ground vibration events represent a critical challenge to operational safety and risk management in mining engineering. Reliable estimation of ground shaking intensity serves as a key prerequisite for effective seismic risk mitigation in mining environments. This study investigates the application of intensity classification techniques to mining scenarios and introduces a deep learning-based Multi-Scale Feature Fusion Classification Network. The model utilizes waveform data within two seconds following P-wave onset to enable rapid and accurate classification of seismic intensity levels. With an intensity threshold of 4.5 used to differentiate low- and high-intensity events, the model achieves a classification accuracy of 90.27% on the test set. Particularly, it surpasses 99% accuracy for samples with intensity levels ≤3 or ≥6, demonstrating a substantial advantage over traditional approaches that require complete waveform sequences and exhibit limited processing efficiency. Notably, the model maintains over 85% accuracy under -5dB SNR conditions, outperforming state-of-the-art methods in noise resilience, while reducing computational complexity by 78% compared to Transformer-based architectures. The proposed method supports real-time seismic monitoring in mining areas, facilitating rapid assessment of risk levels and providing critical decision support for operational adjustments and emergency response. These capabilities contribute to enhanced seismic resilience and the promotion of safe mining practices.