Research On Side-scan Sonar Image Mosaic Based On Feature Point

With the shortage of land resources,depositing the ocean resources has been a crucial part,so defending the rights and safety of the ocean has also been a criterion of national defense among many countries in the 21st century.At present,sonar is the major equipment for exploring the ocean.It produces sonar image through receiving the echo of its sonic wave under the water.Therefore,we process the sonar image so as to explore the ocean.However,the most popular sonar is side-scan sonar,whose all images are stripe images which are unsuitable for people to observe the whole scene of the underwater.Thus,many essential fields based on sonar images are restricted,such as topographic mapping,the deposit of underwater resources and undersea warfare.As a consequence,in this paper,we do research in the mosaic algorithms of side-scan sonar images,and the main work are as follows:Firstly,we decode the data(.XTF format)which are derived from the sonar equipment produced by Triton Imaging Inc.Then stripe images are produced by reading information of side-scan sonar,including image data,GPS data,azimuth of the ship,velocity,etc.Besides,the gray scale of sonar image will have clear contrasts because of the loss during the spread process of the sonic wave,so we have to adjust the gray scale.Meanwhile,we have to revise the velocity,because the changing ship velocity would induce the distortion of the longitudinal ratio.Furthermore,there is a problem of sonar image that the target has big transverse compression when it's near the central fade zone,so we have to adjust the slant range.Moreover,we also analyze the reason for the noise of sonar image and then we build the model for the noises.Then we use common denoising methods to remove the noises of sonar image and analyze the results of other researchers' work.Lastly,we use a method which combines self-adaptive median filtering with partial differential equation to filter the sonar image,which is proved to be an effective method to remove the image noises.In order to cater for the real-time requirement of side-scan sonar image montage in the undersea warfare,submarine rescue and other areas,we compare and analyze some commonly used image registration operators,and study fast ORB(Oriented Brief)image registration algorithm.At the same time,in terms of the problems of using ORB operator,such as too many feature points and uneven distribution of the extraction of the texture-rich region,we propose a grid-based segmentation method,so as to achieve a rapid sonar image montage without negatively influencing registration accuracy.As for the requirement of montage accuracy in the field of mapping the seabed topography,undersea resources development and seabed sediments identify,we focus on the research of the operator of KAZE feature points,and then give its sonar image registration results.Then we derive the registration result of the sonar images,and analyze our methods in terms of detection precision,robustness,accuracy,and then comparative analysis results are given.Use RANSAC algorithm to calculate the conversion relation matrix in order to complete the mosaic between side-scan sonar image.For the situation of the image may exist splicing gap,to study the image fusion algorithm,and give the good results and eliminates the splicing gap.Finally,we summarize the research work in this paper,as well as look ahead the future work.