Research On Small Target Tracking Algorithm In Sonar Image Based On Particle Filter

In recent years,with the improvement of scientific and technological level,people have a deeper understanding of the ocean,the status of the ocean is also improving.People have realized that the ocean is not only an important energy reserve,but also plays an important role in national strategic security as a sacred and inviolable territory of the country.With the continuous development of underwater weapons to miniaturization,underwater autonomous vehicles,frogmen and other underwater small targets have become the key research objects of underwater attack because of their good concealment and strong mobility.Therefore,it is of great practical significance to detect and track small targets such as frogman in time for the development and protection of our country’s oceans.Sonar is the main equipment to detect the ocean.The forward-looking sonar in sonar has high imaging efficiency,so it plays an important role in the study of underwater target detection and tracking.In this thesis,target detection and tracking based on forward-looking sonar image are deeply studied as follows:Firstly,sonar image preprocessing is carried out.The binary echo signal received by sonar equipment is preprocessed through data format conversion,orthogonal demodulation,matched filtering,down sampling and other steps to obtain a relatively clean and clear echo signal.Then the sonar data is transformed into sonar image in Cartesian coordinate system through beamforming and improved nearest neighbor algorithm.Finally,the small target in sonar image is detected by integral graph algorithm.Secondly,a particle filter sonar image target tracking algorithm based on multi features is proposed to improve the importance sampling part of particle filter.Aiming at the problem of low target tracking accuracy caused by sidelobe effect in forward-looking sonar image,this paper chooses to use robust particle filter algorithm.Considering that a single image feature is easy to be disturbed by background similar features when tracking targets,observation models based on moment invariants and color features are established respectively.The similarity of moment invariants and color features of each particle and the corresponding target template in the two observation models is measured as the feature weight and fused to realize the tracking of small targets in sonar images.Finally,the resampling part of particle filter is improved by using particle filter target tracking algorithm based on artificial immune.The idea of artificial immune is introduced into the particle filter framework,the target feature of sonar image is used as antigen,the regional feature of particle set is used as antibody,and the feature weight is used as the affinity between antigen and antibody.Through the cloning,mutation,selection and other operations of artificial immune algorithm,the diversity of sample set is improved,the disadvantage of particle filter resampling particle dilution is made up,and a good tracking effect is obtained.