Системи та методики обробки інформації

Permanent URI for this collection

https://elar.naiau.kiev.ua/handle/123456789/25536

Browse

Browsing Системи та методики обробки інформації by Author "Khoma, Yu.V."

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Дослідження придатності електрокардіограм для біометричної ідентифікації. STUDY OF ELECTROCARDIOGRAMS SUITABILITY FOR BIOMETRIC IDENTIFICATION
    (2018) Хома, Ю.В.; Khoma, Yu.V.
    The article describes the requirements for biometric identification systems. It is also pointed out that electrocardiogram is a promising biometric marker for human identity recognition. This paper presents that ECG based biometric systems are almost insensitive to the type and parameters of the ECG measurement instrumentation, the length of records, the number of users in the database, as well as the classification algorithms based on different methods of machine learning. Additional focus on electrocardiogram acquisition for biometric systems has been made. A simple version of the signal aquisition system based on the Arduino Uno platform and e-Health Sensor Platform v.2.0 is presented. The content of the main stages of the signals and data processing in the ECG-based biometric identification system is described. The general research methodology was designed, in particular, for estimating the accuracy of different classification models. Four open-source electrocardiograms that have been used for conducting have been described in the paper. Also presented are six machine learning methods that were used to build a classification model for a system of biometric identification. Current research has shown that the parameters of the ECG measurement instrumentation, in particular the type of electrodes, the sampling rate, and the resolution of the ADC, do not significantly affect the identification of the results. A good correlation was found between the accuracy of all classification algorithms among different databases. It has been established that for well-balanced data, even for a few hundred users, simple algorithms such as k-nearest neighbors (KNN) and linear discriminant analysis (LDA) provide high-accuracy identification. The neural network classifier with one hidden layer also showed high accuracy and stability in all databases, but due to its computational complexity it is expedient to use it in biometric identification systems with heterogeneous datasets of electrocardiograms.