Pioneering Magnetocardiography-Based Human Identification: A Non-Contact Biometric System with Deep Learning

Biometric authentication technologies has been widely explored in recent years, due to their stability and toughness of imitation. Electrocardiogram (ECG)-based human identification has been studied for decades, achieving extensive progresses, while the contact measurement limits the usage of ECG in real life. Magnetocardiography (MCG) collects magnetic signals of cardio events in a non-contact way, which has been proved has higher sensitivity than ECG. This paper presents the first comprehensive study on MCG-based human identification, addressing the significant research gap in MCG-based non-contact biometric identification. A novel identification system has been developed, compromising: (1) data acquisition using a 36-channel optically pumped magnetometer (OPM) system (2) a preprocessing pipeline containing denoising, peak detection and heartbeat segmentation (3) a patch mean removal enhanced two-stage oriented PCA (OPCA) feature extraction method, and (4) a deep neural network classifier with batch normalization.36 healthy adults are recruited in this research. Using the carefully designed identification architecture, the MCG human identification accuracy achieves 99.29% (FPR is 0.0205%, FNR is 1.11%) in test set, comparable to the accuracy of state-of-the-art ECG human identification (99.66%). Besides, the implement of neural network shows significant performance advantages over classical machine learning classifiers (KNN: 93.93%, SVM: 97.86%, LDA: 96.79%). The proposed system validates MCG's viability as a biometric modality, offering inherent advantages including contactless operation, resistance to skin-contact artifacts, and enhanced signal stability in noisy environments. This research pioneers a new paradigm for non-invasive authentication systems and opens avenues for MCG-based security applications.