Early Detection of Seismic Signal Anomalies Using Raspberry Pi 5 and Lightweight Machine Learning Models
Ahmad Kadarisman 1,2, Ayu Widowati 1, Marrissa Arlinkha 1, Rina Yuniarty 1, Rini Anggraeni 1, Santoso Soekirno 1, Martarizal 1, Hanif Andi Nugraha 2

1. Department of Physics, Universias Indonesia, Depok, Indonesia
2. Directorate of Instrumentation and Calibration, BMKG, Jakarta, Indonesia

Abstract

Real-time detection of anomalous seismic signals is critical for maintaining the reliability of monitoring systems, particularly in environments prone to anthropogenic interference and instrumental instability. This paper presents a lightweight edge computing framework built on the Raspberry Pi 5 and AI-Kit, designed to identify signal anomalies onsite. Seismic data streams in MiniSEED format were segmented and processed to extract statistical and spectral features (e.g., RMS, kurtosis, and Power Spectral Density). Unsupervised learning models, including Isolation Forest and a lightweight Autoencoder, were deployed to detect deviations without the need for labeled datasets. Experimental evaluations using data from the TOJI station showed effective anomaly identification with low inference latency and minimal resource consumption, underscoring the systems suitability for deployment in resource-constrained seismic networks.

Keywords: Seismic anomaly detection, Raspberry Pi 5, real-time monitoring, AI-Kit, unsupervised learning, anthropogenic noise, embedded system, Isolation Forest, Autoencoder

Topic: Instrumentation and Computational Physics