چكيده به لاتين
EEG signals are recordings of electrical brain activity, typically measured along the surface of the scalp, resulting from synaptic activity in brain neurons. Authentication and security are critical issues in our lives, and brainwave-based authentication is an addition to biometric authentication systems that offers numerous advantages over others. Recent studies have demonstrated that EEG is a viable signal for biometric authentication, boasting important features such as resistance to spoofing attacks and immunity to coercion or pressure. In this thesis, the BCI IV 2a dataset is utilized, where participants are instructed to imagine the movement of four body parts: right hand, left hand, both feet, and tongue. To classify individuals, a two-dimensional CNN model with 6 convolutional layers was utilized. All potential states were taken into account based on the task type, input signal length, signal frequency band, and the number of EEG channels, with their results presented. According to the obtained results and analyses conducted, the optimal outcomes are associated with employing the combination of frequency bands alongside the maximum number of EEG channels for authentication purposes. Within this thesis, two signal input modes were examined, with durations of 4 seconds and 1 second. The highest accuracy, recorded at 100%, was achieved for the 4-second signal length. Similarly, for the 1-second input signal, accuracy reached 99.42%, notably during users' imagination of left-hand movement. Despite considerable research advancements in recent years, user authentication based on EEG signals still faces unresolved challenges that necessitate further investigation. This is crucial for EEG-based authentication to attain the level of reliability and security comparable to traditional biometric agents like fingerprints, especially in real-world deployment scenarios.
