Neuromorphic Classification of Geophone Signals with Legendre Memory Units

Geophones are acoustic detectors that respond to seismic waves, which generate ground vibrations. In this work, we utilized a neuromorphic (brain-inspired) Legendre Memory Units (LMUs)-driven neural model for low-power, real-time classification of Geophone data, discriminating vibrations, which were generated by human footsteps, a moving vehicle, and ambient noise. We show that our neuromorphic hardware-compatible neural design produces comparable results with state-of-the-art long-short-term memory (LSTM) models, achieving high test accuracy (LMU: 93.88%, LSTM: 95.15%). Our work highlights the potential of LMUs-driven inference models for classifying analog data in energy-constrained environments.