Dunhuang manuscripts, from the Eastern Jin Dynasty to the Northern Song Dynasty, are precious materials for studying the history and culture of each dynasty. However, due to the corrosion of time and the intrusion of foreign enemies, these manuscripts are scattered everywhere. Therefore, this paper proposes an ancient Dunhuang Manuscripts matching approach based on multi-dimensional feature fusion. We design a multi-dimensional feature description framework with three-dimension, namely edge orientation information, global geometric features, and column spacing features. Then we implement feature dynamic weighted fusion by self-attention mechanism of Transformer. In addition, we construct a dataset simulating real debris tearing to support the training and evaluation of the model. The experimental results show that our framework improves the accuracy and efficiency of fragment matching, which lays a solid foundation for digital protection and research of Dunhuang manuscripts.