Sparse representation-based classification with two-dimensional dictionary optimization for motor imagery EEG pattern...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Sparse Representation-based Classification (SRC) is effective for motor imagery EEG pattern recognition.
• The performance of SRC heavily relies on the quality of dictionary construction.

Purpose of the Study:

• To propose a novel Two-Dimensional Dictionary Optimization (TDDO) method to improve SRC performance.
• To enhance the accuracy and robustness of motor imagery EEG classification.

Main Methods:

• Constructed an initial dictionary using multi-band features from Filter Bank Common Spatial Pattern (FBCSP).
• Applied Lasso regression for feature selection and K-Nearest Neighbors (KNN) for noise atom cleaning.
• Implemented SRC using linearly represented training samples for classification.

Main Results:

• The TDDO-SRC method demonstrated necessity and rationality.
• Achieved highest average classification accuracies of 86.5% and 92.4% on two public datasets.

Conclusions:

• Dictionary construction quality significantly impacts SRC robustness.
• The optimized TDDO-SRC method substantially improves classification accuracy over original SRC.