Aiming at the needs of different domains that need to be faced in the detection process at this stage, the research of cross-domain target detection technology based on deep learning is proposed. Introduce the deep learning target detection technology and lead to the cross-domain target detection technology, combine the previous literature and experiments, summarise the methods used and the shortcomings; list the main cross-domain datasets used at this stage, including CDTD, Cityscapes and Foggy Cityscapes, and M3D, as well as the commonly used evaluation metrics for cross-domain target detection, such as mAP, F scatter and H scatter, all of which can accurately evaluate the effectiveness of the algorithm for cross-domain target detection. The existence challenges of cross-domain target detection are pointed out, including 3 aspects of domain differences, small sample learning, and data enhancement and feature alignment, and the outlook is given in 3 directions, namely, improving adaptivity, enhancing robustness, and strengthening portability. The results show that this study can provide a comprehensive reference framework for research and practice in the field of cross-domain target detection.