BCI Signal Generation and Classification: Application of Multimodal CNN in EEG Processing

Project Source: Original author/maintainer mcamila1201, GitHub project modelo-generativo-bci (link: https://github.com/mcamila1201/modelo-generativo-bci, release date: 2026-05-27). Core Objective: Solve the problems of signal generation, feature extraction, and classification recognition in BCI systems, and build a complete data processing pipeline. Key Methods: Combine STFT and CWT time-frequency analysis to extract features, and use a dual-branch multimodal CNN architecture to achieve EEG signal classification and evaluation.

Brain-Computer Interface (BCI) establishes a direct communication channel between the human brain and external devices. Non-invasive BCI based on Electroencephalogram (EEG) is widely used due to its safety and ease of use. However, EEG signals have characteristics of non-stationarity, low signal-to-noise ratio, and high dimensionality, making it difficult for traditional methods to effectively extract pattern information.

Signal Generation and Preprocessing: Synthesize EEG signals containing typical brain electrical rhythms such as alpha waves (8-13Hz) and beta waves (13-30Hz), which can be used to verify algorithms or expand datasets. Time-Frequency Feature Extraction: Use STFT (Sliding Window Fourier Transform, efficient and suitable for real-time) and CWT (Scalable Wavelet Basis, multi-resolution suitable for transient signals), and normalize after extraction to eliminate differences. Multimodal CNN Architecture: Dual branches process STFT/CWT feature maps in parallel; the feature fusion layer performs concatenation/weighted fusion; the classification head outputs class probabilities through fully connected layers.

Implement a complete training pipeline (data loading, model initialization, loss calculation, optimizer configuration). In the evaluation phase, output indicators such as accuracy, precision, recall, and F1 score to comprehensively evaluate the model's classification ability.

Technical Significance: Signal generation alleviates the problem of BCI data scarcity; mixed training improves generalization; combined use of STFT and CWT captures both steady-state and transient neural patterns; time-frequency representation visualization enhances model interpretability. Application Prospects: Plays an important role in fields such as medical rehabilitation and human-computer interaction.

Potential improvement directions: Introduce attention mechanisms to focus on important time segments; explore GAN/diffusion models to synthesize more realistic EEG signals; expand to multi-channel EEG processing to support complex interaction scenarios.