Among the approximately 60000 bone stick fragments excavated from the Weiyang Palace ruins of the Han Dynasty, around 57000 are inscribed. Most bone sticks exhibit longitudinal fractures at the time of excavation, resulting in the separation of their upper and lower parts. This fragmentation poses significant challenges to the digital preservation and systematic classification of these cultural artifacts. Traditional manual classification methods are inefficient and may cause further damage to the bone sticks. To address these challenges, this study proposes a parallel multimodal classification model that integrates both bone stick images and inscription information. Visual features are extracted from the images using the Vision-RWKV large-scale model, while textual features are obtained from the inscriptions via the RWKV model. A dynamic cross-modal feature fusion module is introduced to integrate image and text features, followed by a classifier for fine-grained categorization. Experimental results demonstrate that the proposed method achieves an accuracy of 92.85%, significantly outperforming conventional deep learning models and other multimodal approaches. This study provides a robust technical foundation for the efficient classification and organization of bone stick artifacts and establishes a solid basis for the intelligent development of archaeological research.