To address the problems that existing community detection algorithms ignore the structure of neighbour diversity in local density computation, and that it is difficult to take into account both global structure and local features in the seed selection phase, this paper proposes a local community detection algorithm based on density peak seeding diffusion (DPCD). Firstly, the DPCD constructs a new local density definition by distinguishing between independent and shared neighbours of a node and combining the structural similarity information among neighbours to more accurately portray the degree of aggregation of a node in a local structure. At the same time, it combines the relative distance of nodes and the network average threshold to select representative and spatially distributed seed nodes with good adaptability on both global and local scales. Then, the seed nodes are selected through the neighbourhood local density-based algorithm. Finally, the label update strategy based on the local density of neighbours is used to further optimize the community belonging of the nodes. By comprehensively considering both the global distribution characteristics of the network and the internal connectivity of communities, the DPCD greatly improves detection performance. Experiments on various real and synthetic networks demonstrate that the DPCD significantly outperforms other algorithms.