Exploring the feasibility of real-time on-device ECG biometric classification using quantized neural networks

Biometric classification using electrocardiogram (ECG) signals offers a promising pathway for continuous, personalized healthcare monitoring. This work presents a proof-of-concept embedded deep learning system for real-time ECG biometric classification on wearable Holter devices, reducing reliance on continuous cloud connectivity. A quantized convolutional neural network (CNN) was deployed on an STM32H7 microcontroller to identify individuals based on unique ECG patterns, incorporating an initial signal quality assessment stage to ensure that only high-quality segments are processed. Evaluated on the PTB Diagnostic ECG Database with subject-specific training, the system achieved F1 score of 94.51% and a classification accuracy of 94.68% on five-second ECG segments, with an average inference time of 1.35 s, enabling real-time operation on resource-constrained hardware. By performing on-device inference, the system improves data privacy, can reduce power consumption, and minimizes unnecessary data transmission. This embedded implementation demonstrates the feasibility of integrating lightweight ECG biometrics into wearable systems, with potential for future extensions toward personalized healthcare monitoring and early anomaly detection.