| Abstract [eng] |
Project objective – To investigate the possibilities of using EEG signals for control purposes. Tasks: Analyze the main artifacts that occur during EEG signal measurement; Examine methods for extracting and classifying EEG signal features based on their origin and assess their applicability; Compare vertical and horizontal eye movements and blinks in EEG data; nvestigate the possibilities of detecting eye blinks, movements, and states, and generating control signals from EEG data. Project structure: theoretical part, experimental part, research part, conclusions, and list of references. Software used: “Python” 3.10.12; “BrainAccess SDK”. Conclusions obtained: EEG signal artifacts are categorized as internal (heart rate, eye movements, muscle activity) and external (poor electrode-to-skin contact, low electrode conductivity). Methods such as ICA, DTW, amplitude thresholding, and others can be used to detect these artifacts. EEG feature extraction methods fall into the following categories: time domain (fractal analysis of signals and statistical features), frequency domain (Fourier transform or power spectral density), decomposition (wavelet transform, empirical mode decomposition, or independent component analysis), and spatial domain (CSP algorithm and its variations). For EEG feature classification, algorithms used include machine learning methods (SVM, CNN, DBN, LSTM, GAN) and analytical methods (LDA). Human-machine interfaces based on EEG data interpretation are most commonly used for recognizing movements, emotions, or fatigue. A comparison of vertical and horizontal eye movements and blinks in EEG data showed differences in the 20–45 Hz frequency band – peaks of forced eye blinks were detected using the ICA method along with PSD calculation in this frequency band, with values lower than 0.1 μV2/Hz, whereas vertical eye movements showed values greater than 0.1 μV2/Hz. Eye blinks can be most effectively detected using the ICA method combined with PSD calculation in the 1.5–12 Hz frequency band. Based on the detected blinks and eye movements, an algorithm was developed that converted this information into control signals. The algorithm achieved an average effectiveness of 93.93 proc. in detecting eye blinks and vertical movements when two channels were not available. |
