Aiming at the difficulty of accuracy and stability of UAV vision positioning, the application of object detection network in UAV vision positioning is analyzed. According to the characteristics of UAV vision positioning and the advantages of object detection network, taking feature layer extraction as main design idea, an algorithm is proposed to introduce improved SSD network into UAV vision positioning by modifying the structure of SSD network, designing loss function, changing candidate box selection rules and merging normalized network. The location information of UAV is verified and the feasibility of the algorithm is verified. Experimental results show that the network can meet the requirements of UAV vision positioning, and has certain theoretical and practical reference value.