Aerial object detection is the process of detecting objects in remote sensing images, such as aerial or satellite imagery. However, due to the unique characteristics and challenges of remote sensing images, such as large image sizes and dense distribution of small objects, annotating the data is time-consuming and costly. Active learning methods can reduce the cost of labeling data and improve the model's generalization ability by selecting the most informative and representative unlabeled samples. In this paper, we studied how to apply active learning techniques to remote sensing object detection tasks and found that traditional active learning frameworks are not suitable. Therefore, we designed a remote sensing task-oriented active learning framework that can more efficiently select representative samples and improve the performance of remote sensing object detection tasks. In addition, we proposed an adaptive weighting loss to further improve the generalization ability of the model in unlabeled areas. A large number of experiments conducted on the remote sensing dataset DOTA-v2.0 showed that applying various classical active learning methods to the new active learning framework can achieve better performance.