Unmanned Aerial Vehicles (UAVs) are increasingly being used for surveillance, search and rescue, and other applications. Human action recognition (HAR) using UAV-captured data is a crucial task that enables real-time detection of human actions from visual input. However, the altitude and mobility of UAVs pose challenges for accurate HAR. In particular, the diminished size of people in aerial views calls for the creation of specialized models to identify human actions precisely. In recent years, Graph Convolutional Network (GCN) models has demonstrated impressive performance in skeleton-based HAR. A graph-based deep learning model is proposed that integrates the spatial convolution module from AAGCN and the temporal convolution module from TD-GDN. The effectiveness of the proposed method is evaluated on the UAV-Human dataset, with experimental results showing superior performance compared to state-of-the-art methods.