As an important research direction in the field of deep learning (DL), human movement recognition has shown broad application prospects in sports movement recognition. Movement recognition aims to accurately determine the behavior category of characters by analyzing video content containing their activities. In recent years, methods based on Graph Convolutional Networks (GCN) have shown excellent performance in describing human skeletal structures, effectively improving the precision of movement recognition. However, in practical applications, there are still problems such as insufficient spatiotemporal information fusion and insufficient dynamic adaptability of graph structures. Therefore, this article proposes a sports movement recognition model based on Dynamic Directed Graph Convolutional Network (DD-GCN), which models representation learning on dynamic graphs as a joint aggregation process of temporal and spatial information. The model integrates the ability of GCN to extract spatial structural features with the ability of Time Convolutional Network (TCN) to capture temporal historical information through causal convolution, and introduces an adaptive parameter update mechanism in the spatial convolution layer, allowing the model to dynamically adjust parameters based on changes in graph structure. The results indicate that our model has high recognition precision and good generalization ability in sports movement recognition tasks.