The purpose of multi-modal entity alignment is to discover entities with the same meaning represented in two multi-modal knowledge graphs. Due to the heterogeneity of structures between different knowledge graphs, this paper proposes a multi-modal entity alignment model based on neighborhood matching. In structural embedding, this paper introduces long-distance neighbor information, uses a gating mechanism to aggregate direct and long-distance neighbor information, and the matching module is used to calculate the similarity between the neighborhood nodes. Meanwhile, multi-modal information such as relationships, attributes, and images are fused to enhance the effect of entity alignment. In this paper, experiments are conducted on DBP15K datasets and the results are significantly better than the baseline method.