Recommender systems have become a crucial intelligent tool, which provides users with personalized services. Graph learning-based recommendation methods treat user-item interactions and the item transitions as pairwise relations but ignore the complex, higher-order interaction information between nodes. Moreover, since most users often interact with few or even no items, graph learning-based recommendation methods suffer from the data sparsity problem, as well as the unbalanced distribution of edges and nodes. To tackle these issues, we propose a Dual Hypergraph-based Self-supervised Learning recommendation model, named DHSL-GM. Specifically, we derive two dual hypergraphs from the user-item bipartite graph, which models the complex high-order user-item interactions by using hypergraph convolution with spectral hypergraph convolution operator. Meanwhile, we design a gated network-based message passing mechanism to dynamically guide message propagation, addressing the problem of the unbalanced distribution of edges and nodes. In addition, to alleviate the data sparsity problem, we design another dual hypergraph convolutional network based on a node discard strategy, which innovatively integrates self-supervised learning into the training of the hypergraph convolutional network. Experimental results on several real datasets demonstrate the superiority and effectiveness of the proposed model.