Abstract: UAV aerial images are characterized by complex backgrounds, small object scales, variable object orientation angles. The training data of the model often deviates from the application scenario, leading to significant degradation in detection performance. To address the above problems, a UAV aerial object detection algorithm based on domain adaptation is proposed. The method adopted the overall idea of domain adaptation to improve the generalization of the model in different scenarios. Furthermore, the method designed a feature extraction network which incorporated deformable convolution and ResNet-101, as well as a feature pyramid network and a D-ROI Align to improve the detection accuracy. Domain adaptation experiments based on UAV-related datasets show that the method has better detection performance than the current mainstream algorithms.