Multiscale EEG feature fusion for recognizing 3D object shapes through active touch.

Abstract

Tactile perception has become an important topic in brain function research and in the development of tactile sensing and human-machine interaction technologies. Although prior studies have investigated tactile and haptic processing using electroencephalography (EEG), many of them have focused on relatively simple tactile attributes, passive stimulation paradigms, or single-domain neural features. A systematic EEG framework for active discrimination of multiple three-dimensional (3D) object shapes, together with multi-domain neural characterization and feature-fusion-based decoding of shape-related information, remains limited. In this study, we investigate the neural mechanisms underlying human tactile perception of 3D object shapes during active touch. We design a tactile experiment in which participants grasp objects with different 3D shapes while EEG signals are recorded. We then perform a multi-domain analysis of the collected EEG data, including the time-frequency energy domain, the dynamical complexity domain, and the brain functional network domain. Furthermore, to evaluate whether the complementary information extracted from different neural feature domains can support reliable decoding of shape-related information, we employ an attention-enhanced multi-channel convolutional neural network as a feature-fusion-based validation framework. Experimental results show that the presence or absence of tactile stimulation can be distinguished by event-related synchronization and desynchronization in central brain regions within the alpha and beta bands; that object conditions involving a greater number of edges are associated with lower brain complexity during active touch; and that the local efficiency and clustering coefficient of the brain network in the high-alpha band are significantly correlated with 3D shape categories. Using the above multi-domain features as input, the validation framework achieves an average recognition accuracy of 92.20% for the tactile perception of seven 3D shapes. These findings provide a more systematic active-touch EEG framework for studying 3D tactile shape perception and offer new insights into its underlying neural representations, while serving as a tactile-dominant baseline for future multimodal intelligent interaction research.