As the core task in the field of computer vision, object detection has made great progress in recent years. In recent years, a target Detection model based on Transformer architecture-DETR (Detection Transformer), has attracted wide attention. The purpose of this paper is to introduce and analyze the characteristics of the earth observation self-built data set, and then use DETR model to train the earth observation self-built data set and evaluate and analyze the experimental results. The challenges of target detection and the limitations of traditional methods are reviewed, followed by an introduction to the Transformer model. Then, the structure, training flow and key components of DETR model are discussed in detail. We experiment the DETR model on common target detection datasets and compare its performance with traditional target detection methods. Under the premise of certain advantages, the model achieves an average precision of more than 40%, which is comparable to the performance of traditional models.