Research On Ship Target Tracking Method Based On Video Satellite

China’s territorial sea is vast and rich in marine resources.With the increasing frequency of maritime activities and the continuous changes of the international situation,marine resource management and territorial sea security are facing severe challenges.As the main means of transportation on the sea,ships have important strategic significance in the civil and military fields.Accurate and rapid detection and tracking of them has become an important means of maritime target supervision and protection of maritime rights and interests.With the rapid development of remote sensing technology,the emergence of video satellites with longer shooting time and higher resolution makes it possible to track ship targets on the sea in real time.Based on the powerful feature learning ability of deep learning,this paper carries out the research of marine ship target tracking method based on video satellite,and finds the method suitable for marine ship target tracking,which provides a new solution for video satellite ship target tracking.The main contents and conclusions of this paper are as follows:(1)The data sets of ship target detection and tracking are constructed.Deep learning needs continuous training to improve the accuracy of the network,so it needs a large number of training samples.At present,there are few open ship data sets based on remote sensing images,which can not meet the experimental needs.Therefore,this paper obtains the optical remote sensing images and ship images in the open remote sensing target detection data set,and constructs the ship target detection data set.Based on the data of “jilin-1” video satellite,the ship target tracking data set is constructed.(2)In order to solve the problem of low accuracy of SSD(Single Shot Multibox Detector)algorithm in ship target detection,a R＿SSD ship target detection algorithm based on Res Net(Residual Network)and optimal prior box size is developed.The algorithm uses Res Net as the feature extraction network,and improves the size of the prior box for the ship target detection data set constructed in this paper.Finally,it is tested in the ship target detection data set,and the results show that the detection accuracy of R＿SSD algorithm is improved by 7.2%,which verifies the effectiveness of the proposed algorithm.(3)In order to solve the problem that the lost tracking target can not be retrieved and the tracking accuracy is low in the SORT(Simple Online And Realtime Tracking)target tracking algorithm,This paper developed a Deep SORT ship target tracking algorithm based on R＿SSD,which uses R＿SSD detection algorithm,the depth apparent feature of the target is introduced on the basis of SORT tracking algorithm,and the ship target tracking is realized based on Kalman filter and Hungarian algorithm.Finally,it is verified on the ship target tracking data set constructed in this paper.The results show that the tracking accuracy of the improved Deep SORT tracking algorithm is improved by 5.8%.For the target lost in the tracking process,it can be recovered in the subsequent frames,which proves the effectiveness of the method in ship target tracking.