Unmanned aerial vehicle(UAV) image matching methods have been extensively applied in various fields such as search and rescue, precision agriculture, etc. Due to the change of the UAV attitude and complexity of flight environment, there are scale morphological changes and multi-view differences between the UAV images and satellite images. This paper proposed a UAV image matching based on graph neural network. First, a method of feature extraction based on multi-type attention graph neural network is introduced to solve the view differences between two images. Then, a multiple kernel clustering based on consensus affinity graph is used to cluster the feature descriptors, which deals with the problem of scale morphological changes. Finally, an optimal transport matching method based on random sampling is proposed to improve the robustness to noise. The experiment results demonstrate that our method outperforms the available methods. The AR of it is 1.97% higher than the original matching model.