EEG Brain-Computer Interface Mental State Detection

The patent segments the BCI system into distinct functional modules: EEG signal acquisition unit, feature extraction unit (using Shannon entropy and autoregressive models), mental state classification unit (using SVM and Bayesian inference), and feedback control unit. This modular segmentation allows each component to be optimized independently, improving overall detection accuracy while managing system complexity through organized functional divisions.

The patent transforms raw EEG signals into multiple derived parameters including Shannon entropy values, autoregressive coefficients, and spectral features. By changing the parameter representation from raw time-domain signals to these transformed features, the system achieves superior mental state detection accuracy, resolving the contradiction between simple structure and high precision measurement.

The patent performs preliminary feature extraction by computing Shannon entropy and autoregressive parameters from incoming EEG data streams before full classification. This preliminary processing prepares the data in advance, allowing the main classification algorithms (SVM and Bayesian inference) to operate more efficiently on pre-processed features, thereby reducing overall processing time while maintaining high detection accuracy.

The patent implements a streamlined processing pipeline that selectively computes only the most discriminative features needed for mental state classification. By skipping unnecessary computational steps and focusing on critical features (such as entropy and autoregressive parameters), the system achieves rapid processing of real-time EEG data without sacrificing measurement precision.

The patent extracts and retains only the most informative features from the EEG data, specifically Shannon entropy and autoregressive parameters, while discarding redundant features. This selective extraction approach maintains high classification accuracy by keeping the essential discriminative information while eliminating unnecessary data, thus resolving the contradiction between precision and information loss.

The patent computes a comprehensive set of potential features from the EEG signals but then applies selection criteria to retain only the subset that provides maximum classification value. This partial action approach—computing more features initially but using only the essential ones—ensures high detection precision while managing feature redundancy through selective utilization.

The patent implements a calibration phase during which user-specific parameters and baseline characteristics are extracted and stored before actual BCI operation. This preliminary personalization action allows the system to adapt to individual users without adding complexity during real-time operation, as the personalized processing rules are pre-established during calibration.

The patent enables the BCI system to automatically adapt to individual users through self-calibration procedures that require minimal user intervention. The system performs automated feature extraction and parameter optimization during initial use, allowing it to personalize its processing without requiring complex manual configuration or extensive professional calibration, thus improving user-specific accuracy while managing system complexity.