Abstract: CircRNA is a kind of RNA with a closed ring structure, and in-depth analysis of the interaction mechanism between CircRNA and binding proteins is of great significance for disease prevention and treatment. However, in the task of identifying the interaction sites of CircRNA binding proteins, it is difficult for the existing methods to adequately capture the nonlinear relationships in the complex data and lack of efficient feature extraction mechanisms, for this reason, we propose a new feature extraction model, HiMap-Circ, which can extract richer and more comprehensive features through feature high-dimensional mapping, thus improving the expression effect of the sequence features and overall model performance. In HiMap-Circ, three different encoding methods are used to encode features in CircRNA sequences, and feature fusion is performed by combining feature high-dimensional mapping, multi-scale convolution, and the attention mechanism, so as to capture both local and global features in the sequences. Comparison experiments with other similar methods on CircRNA datasets as well as ablation experiments show that HiMap-Circ outperforms the existing state-of-the-art similar methods and proves to be an effective method for predicting the interaction sites of CircRNA with binding proteins.