A colorectal polyp segmentation network BGGNet （boundary graph and global graph guide network）is proposed，which is dual-oriented by boundary graphs and global graphs. In this paper，three network modules is designed：a single-level multi-scale feature extraction block，which is used to obtain contextual information of deep features；a boundary generation guidance module，which is used to fuse shallow features of adjacent scales to generate a boundary guidance map；a deep global graph guidance module，used to generate global graph-guided features using the global graph and reverse attention. By combining Res2Net with the above model，a progressively refined polyp segmentation area decoder is constructed. This encoder can fuse global graph-oriented features and boundary graph-oriented features step by step，refine the polyp structure step by step，and improve accuracy of polyp segmentation. Experimental results on data sets such as Kvasir and CVC-612 show that the proposed BGGNet network model has better polyp segmentation performance.