Personality recognition holds significant potential in fields such as clinical psychology and human-computer interaction. Despite the promising performance of electroencephalography (EEG)-based personality recognition, its advancement is still limited by insufficient exploitation of the intrinsic spatial topology and ineffective multi-dimensional feature fusion of EEG signals. To overcome these challenges, this paper proposes an adaptive Inter-Channel Relation Guided Feature Fusion Network (CRGFFNet) comprising two dedicated components. The Spatial Proximal-Channel-Reordering (SPCR) algorithm, grounded in reinforcement learning, adaptively reorders EEG channels to preserve three-dimensional spatial structure while attenuating high-frequency noise. Moreover, the Spatio-Temporal-Frequency Hybrid Convolutional Feature Augmentation (STFA) module leverages orthogonal strip convolutions and parallel frequency pooling to extract discriminative local features from multiple signal domains. Outputs from the STFA module are then used as input embedding to a Transformer encoder, thereby enabling modeling of global contextual relationships across the entire sequence. To validate the proposed model, a 64channel EEG dataset was collected from 22 subjects while they were exposed to happy, calm, and sad emotional stimuli. Each subject was then annotated according to the Big Five personality traits. Experimental results demonstrate that CRGFFNet achieves state-of-the-art performance, attaining an average accuracy of 89.94% across all personality dimensions and significantly outperforming existing methods.