The relationship between conscious and unconscious perception remains a key issue in cognitive neuroscience. While emotionally salient inputs can be processed unconsciously, whether this alters large-scale network organization remains unclear, particularly across different emotional valences and perceptual gradients. This study examined the unconscious processing of emotions during Continuous Flash Suppression using EEG, graph theory, and the Perceptual Awareness Scale (PAS). Functional connectivity was estimated via weighted Phase Lag Index across frequency bands, and weighted network metrics (efficiency, modularity, path length, degree clustering, centrality) were analyzed through Linear Mixed Effects Models. Results showed that the transition to unconsciousness shifts network topology from integrated global processing to segregated local modules.