Visual imagery is a cognitive process that generates internal visual representations in the absence of external visual stimuli and has become an important paradigm in brain-computer interface (BCI) research. In recent years, although publicly available simultaneous EEG-fNIRS datasets have been developed for tasks such as motor imagery and mental arithmetic, bimodal neural signal resources specifically designed for visual imagery tasks remain limited. Therefore, this study constructs and releases a simultaneous EEG-fNIRS dataset for three-class visual imagery tasks.The dataset initially collected neural signal data from 23 participants. After quality control, 22 valid participants were retained for subsequent analyses. During data acquisition, 32-channel EEG was used to record scalp electrophysiological signals, while 22-channel fNIRS was used to measure prefrontal hemodynamic responses. The dataset includes raw EEG and fNIRS signals, event markers, channel information, and example preprocessing codes.Furthermore, EEG-based, fNIRS-based, and EEG-fNIRS fusion baseline models were established to evaluate whether the dataset contains discriminative information for three-class visual imagery classification. The benchmark results demonstrate that the dataset contains neural information associated with visual imagery conditions and can support unimodal analysis, multimodal fusion, and BCI-related research.

Background: Visual imagery (VI) is a cognitive process that generates visual representations in the absence of external visual input; however, the spatiotemporal organization of its neural representations remains poorly understood. Methods: In this study, we combined electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) with representational similarity analysis (RSA) and the Schaefer 2018 functional parcellation atlas to compare the correspondence between brain network representations and CORnet-S model representations during visual perception (VP) and VI. We further examined the cross-modal associations between EEG time–frequency representations and fMRI network representations during VI. Results: Significant model–brain representational correspondence was detected in the visual network during VP, and no significant correspondence was detected in any network during VI, yielding different descriptive significance profiles across the two tasks. Cross-modal EEG–fMRI RSA further revealed frequency- and time-dependent representational correspondence involving the frontoparietal, default mode, ventral attention, visual, and limbic networks. Significant cross-modal correspondence was observed across the alpha, beta, and gamma bands, with descriptive variation in the networks and temporal clusters involved in each frequency band. Conclusions: Collectively, these findings suggest that VI is not merely a reinstatement of VP but is associated with distributed representational patterns involving multiple functional networks. This study provides multimodal evidence for understanding the neural mechanisms underlying VI.