AI-Driven Intrusion Detection With Seismic-Acoustic Fusion on Embedded Devices

Modern security systems require robust, real-time solutions to detect anomalies in complex environments. This paper presents a novel multimodal framework integrating the SM-24 geophone and a condenser microphone, leveraging artificial intelligence (AI) to enhance detection performance. By combining seismic vibrations captured by the SM-24 geophone, digitized via the ADS1115 16-bit Analog-to-Digital Converter (ADC), with high-fidelity acoustic signals from the condenser microphone, our approach achieves superior accuracy and robustness compared to single-sensor systems. We propose a lightweight dense neural network (DNN) optimized for resource-constrained embedded platforms, such as the Raspberry Pi 4, enabling efficient real-time processing. Advanced feature extraction techniques, including Mel frequency cepstral coefficients (MFCC) for audio and spectral analysis for seismic data, ensure computational efficiency while maintaining high accuracy. Experimental results demonstrate a detection accuracy of 93.8%, surpassing traditional single-sensor methods by at least 15% in noisy conditions. Evaluated across diverse scenarios, including perimeter surveillance and unauthorized activity detection, the system exhibits exceptional reliability. This work highlights the transformative potential of AI-driven multimodal sensing, offering a scalable, cost-effective solution for next-generation security applications on embedded devices.