Community detection is an important tool for analyzing and understanding large-scale complex networks. It can divide the network nodes into multiple communities, which have dense intra-community connections and sparse inter-community connections. Traditional community detection algorithms focus on non-attributed networks that contain only topological structures and ignore the attribute information on the nodes. Dual-channel attribute network community detection model optimizes the topology and attribute information as two channels, which can make full use of the two types of information and improve the clustering accuracy of the model. As a classical mathematical method of community detection, non-negative matrix factorization is only suitable for linear data, and cannot mine the nonlinear latent structural features. To address these limitations, this paper proposes a dual-channel attributed graph community detection algorithm based on kernel matrix factorization(KDACD). The nonlinear relations between nodes are learned by using kernel trick which projecting attribute features of nodes into high-dimensional Hilbert Spaces, and the robustness of the model is improved by sparse constraints and manifold regularization terms. Extensive experiments on 6 real-world datasets verify the effectiveness of the algorithm.