To address the challenges to community detection accuracy and robustness caused by fuzzy community boundaries, structural imbalance, and incomplete local information in complex networks, this paper proposes a community detection algorithm named CDWD, based on weak edge identification and directed propagation mechanisms. The algorithm first identifies and removes weak edges based on the minimum common neighbor criterion, revealing the boundaries of potential communities. Each connected subgraph thus forms an initial community structure. Next, a directed influence graph is constructed to enhance the internal structural connections within communities by leveraging local similarity, improving the directionality and consistency of information propagation. Finally, based on the strength of the topological association between nodes and candidate communities, dynamically determine their optimal affiliation to ensure the integrity and rationality of community division. Experimental results demonstrate that CDWD outperforms mainstream baseline algorithms on multiple real-world networks, synthetic networks, and graph structures constructed from clustered datasets. At the same time, the algorithm parameters are easy to set, making it convenient for practical applications.