Abstract:In recent years, with the development of convolutional neural network, image sensing technology based on deep learning has made great progress. Because the deep learning algorithm does not rely on artificial aided design markers, has strong generalization ability and high detection accuracy, it has drawn attention to scholars at home and abroad in the field of intelligent perception of spatial non cooperative targets. This paper investigates the research status of intelligent perception of non-cooperative targets using deep learning methods, and introduces and summarizes them. Firstly, the on-orbit application and mission phase planning of space non-cooperative target perception are summarized, and then the structural characteristics and surface illumination characteristics of non-cooperative targets are analyzed; Secondly, three methods of establishing non cooperative target data sets are summarized, and the research status of non-cooperative target recognition and non-cooperative target pose detection at home and abroad is classified and summarized; Finally, the key problems and difficulties of non-cooperative target intelligent perception method based on deep learning are analyzed, and the ideas of follow-up research are given.