Target detection in SAR images has been a research hotspot in recent years， but the characteristics of unclear imaging also make the DETR network model unable to extract its potential features well. At the same time， the DETR network also has the problems of long training cycle and slow convergence. To this end， a Multi-label Assignment DETR （MA-DETR） network was designed for aircraft target detection in SAR images. In this paper， we used a data augmentation module with Large Scale Jittering （LSJ） to enhance the training effect of the network， and then designed a multi-label assignment supervision module to process the data output from the encoder. Among them， multiple supervised auxiliary heads extract potential features and inputted them to the decoder to improve the defects of the one-to-one label assignment method of DETR network. Finally， a matching enhancement module was designed to be added to the decoder to alleviate the matching discreteness caused by the Hungarian matching algorithm and improve the convergence speed of network training loss. The experimental results on the SAR AIRcraft dataset show that， compared with the original method， the proposed method improves the AP_0.5 and AP_0.75 accuracy by 7.9% and 7.4% respectively， and reduces the training cycle by 3.3 times based on the same training network. The new network structure effectively improves the target detection accuracy of SAR images and reduces the training cycle of DETR network.