Research On Improvement Of Ship Target Tracking Algorithm Based On Broad Learning

With the continuous deepening of world trade and the development of the shipping industry,ships are getting bigger and faster.The traditional cruise ship model is restricted.It has the problems of small field of vision,slow response speed,inability to integrate global information,inability to continuously and effectively track illegal ships,and subsequent forensics and difficulty in handling evidence.UAV can be used as a supplement to the traditional model.The weaknesses of reaction speed and visual field limitation are compensated by its high speed advantage.In particular,UAV has obvious advantages in emergency response and investigation and evidence collection.By using UAV to identify and track the escaping ship in the air,the escaping ship can be prevented.It effectively improves the emergency and timeliness of law enforcement investigations.The use of UAV to collect images to achieve tracking of ships has become the key to realizing intelligent and autonomous tracking of suspicious ships.In this paper,a tracking system based on Broad Learning System is designed to solve the problem of how to use the image data obtained by drones to track suspicious ships.Through the width-learning feature layer mapping framework,FHOG(Felzenszwalb’s Histogram of Oriented Gradient)features and CN(Color Name)features are fused to extract reliable target features.Image information is fully utilized.The problem of deformation of the ship target during the tracking process is solved by the enhanced layer learning framework of Broad Learning System in this paper,which can achieve target recognition and classification more accurately and quickly.For the problem of increasing target data in the tracking process,this paper is solved by the incremental learning algorithm of the Broad Learning System to realize the online update of the target model.Compared with traditional algorithms,it greatly reduces the model update time and provides better accuracy and higher stability for subsequent target tracking.The FHOG feature response graph and the CN feature response graph are fused.It can quickly reduce the target area.It also guarantees the real-time nature of the tracking system.At the same time introducing a compensation factor,it can adaptively select the fusion ratio.This can better adapt to the changing goals and background during the tracking process.