Compared with that of the medium or large targets, the detection accuracy of small targets with pixels less than 32×32 is poor due to low pixels and lack of feature information, which results in low accuracy of underwater robots, UAVs and other facilities in classification and positioning of small objects.In order to improve the accuracy of the relative facilities in small target detection, analysis is made on the current situation of general deep-learning target detection algorithms at first.Then, the problems hindering the development of the small target detection technology are expounded, and the algorithms of data enhancement, feature fusion, resolution enhancement and context information that can improve the accuracy of small target detection are analyzed.Finally, the performances of AP and APs of all the algorithms in MS COCO data set are summarized, and the future development direction of small target detection is prospected.