SpikeMind: Physiologically Grounded Spiking Concepts for EEG Emotion Recognition and Rationale Generation

Electroencephalography (EEG) emotion recognition often relies on latent features whose physiological meaning is difficult to compare across subjects or trace through language rationales. SpikeMind uses 12 differential-entropy (DE)-grounded temporal concepts as shared variables for recognition and rationale generation. Their targets describe regional activity and contrasts computed from training-standardized DE features. A coordinate-preserving spiking encoder and fixed band–region readouts learn trajectories aligned with these targets. Reliability-weighted source-subject prototypes classify their temporal summaries and expose predicted-versus-runner-up margin evidence. A planner defines supervision and evaluation facts, while Qwen3-4B organizes the complete continuous concept record into a rationale. On SEED and SEED-IV, weighted accuracy reaches $94.0\%$ and $86.7\%$ under subject-dependent evaluation, and $89.1\%$ and $79.4\%$ under source-only leave-one-subject-out evaluation. The latter exceed EEG Conformer under the same source-prototype rule by $4.1$ and $4.4$ percentage points. Concept-marker correlations range from $0.748$ to $0.842$, and language evidence-retention F1 ranges from $0.832$ to $0.894$. This shared record makes marker agreement, decision sensitivity, and language evidence retention separately testable within one recognition system.