TinyML ECG authentication reaches 98% accuracy on an ESP32 board
Original title: Heart Signal (ECG) as Vibration Signature for Human Authentication Using TinyML
Lead I models from four public ECG databases were deployed on an Espressif ESP-EYE for continuous checks.
The paper proposes continuous electrocardiogram authentication on microcontrollers, compared with photoplethysmography, as an alternative to one-time biometrics the authors say are exposed to spoofing. Spectrograms and spectral features from single-lead ECG trained DNN, 1D CNN and 2D CNN models in Edge Impulse, reaching 98% accuracy. The model was converted to TinyML and deployed on an Espressif ESP-EYE, with SVM models used for benchmarking. Averaging multiple heartbeats improved noise tolerance.
Unconfirmed: The 98% figure is the authors' experimental accuracy; no test-set size is stated.
- Signals came from UofTDB, MIT-BIH Arrhythmia, Noise Stress Test and PTB.
- The board is an Espressif ESP-EYE with an ESP32 at 240 MHz.
- DNN, 1D CNN and 2D CNN models were trained in Edge Impulse.
- SVM models were applied as a benchmark.
- The authors say ECG varies with activity, health, age and emotion.
- Vendors
- Espressif
- Products
- ESP-EYEESP32Edge Impulse
From the original article
1 min read at arxiv.orgEnsuring data security in the era of Artificial Intelligence (AI) and deepfake technology requires robust biometric systems. Conventional methods such as facial recognition, iris scans, fingerprints, and voice recognition are increasingly vulnerable to spoofing and provide only one-time authentication at login.
By Yogeswar Reddy Thota · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.