Research On Deep Learning Based Maritime Targets Tracking Algorithm

Due to its great value in both military and civil applications,the maritime targets tracking problem has always been the focus of attention,a large number of maritime targets tracking algorithms have emerged.Among the large number of maritime targets tracking algorithms,MHT(Multiple Hypothesis Tracking)algorithm is the one of the earliest successful algorithms.Because of its stable tracking performance,MHT algorithm has attracted a lot of attention.But the complexity of the MHT algorithm is large.And it has a poor performance in high sea state.The MHT algorithm is not mainstream method of maritime target tracking algorithms and rarely appears as the baseline in the assessment.In order to reduce the complexity of MHT algorithm and improve the performance of the algorithm in high sea state,a series of studies are carried out in this paper.The specific research contents are as follows:Firstly,in order to reduce the complexity of the MHT algorithm and improve the performance of the MHT algorithm,this paper designs a light convolutional neural network as the detector before the tracking.This paper designs a series of experiments to verify that the designed neural network meets the design requirements.Secondly,we proposed an improved MHT algorithm based on multi feature fusion to improve the performance of the MHT algorithm in high sea state.By combining time-frequency feature with the original feature,the performance of the MHT algorithm in high sea state is improved.The experimental results prove it.