Research On Underwater Image Mosaic Method And Detection Method Of Polymetallic Nodules

Seabed polymetallic nodules resources are one of the most important deep-sea mineral resources,and their exploration and development are of great strategic importance.Because the distribution of seabed nodules varies greatly in seabed mining area.Only the nodules density reaches a criteria in seabed mining area,mining area has the economic value of mining.Therefore,in order to exploit the seabed polymetallic nodules,it is necessary to fully estimate and analyze the nodules density of each mining area.At present,most of the exploration methods for seabed nodules are video images from the bathyscaph,which have poor resolution and field of vision and cannot directly reflect the distribution of seabed nodules.Therefore,in order to splice video images into panoramic images with high resolution and wide viewing Angle and to segment them,the splicing algorithm of video images taken by deep submarine and the autonomous image segmentation algorithm of polymetallic nodules was studied in this paper.Finally,density distribution of polymetallic nodules on the seabed is estimated by image segmentation.Firstly,key frame selection method of video image was studied in this paper.According to the characteristics of undersea video images,a key frame selection method based on correlation filtering was proposed in this paper.The relative displacement between images is calculated by correlation filtering,and then the key frame is selected and the overlap area between the key frames is calculated,which saves a lot of computation for subsequent image Mosaic.The accuracy of the key frame extraction based on correlation filtering is verified by comparative experiments.Secondly,studied the image mosaic technology was studied in this paper.First,the feature matching algorithm was studied in this paper.And then,because matching points of GMS matching algorithm are concentrated,and the matching points on moving objects cannot be effectively filtered out,which affecting the effect of image mosaic.A improved GMS feature matching algorithm was proposed in this paper which is improved from the aspects of feature point extraction and matching point filtering.And the superiority of the improved GMS feature matching algorithm in this paper is proved effectively through comparison experiments.Then this paper studied the image fusion algorithm,aiming at the problem of double object in image fusion and its boundary transition is not natural,this paper proposed an improved asymptotic convergence algorithm.And through the comparison experiment,it is proved that the improved gradual in and gradual out fusion algorithm in this paper can effectively solve the double object problem in image mosaic,and the mosaic image is more natural.At last,the improved feature extraction algorithm and image fusion algorithm are used to combine the selected key frame images.Thirdly,aiming at the time-consuming and laboring problem of image segmentation data set,a method of pre-labeling tuberculosis image based on threshold segmentation was proposed in this paper,which can effectively improve the efficiency of data set labeling.Then,random clipping,rotation and noise are adopted to expand the data set and increase the diversity of the data set samples.Finally,image segmentation algorithm was studied in this paper.To solve the problem of large difference in the size of seabed nodules at image,the improved Unet++ algorithm was proposed in this paper,which is improved from the aspects of the convolution structure,sampling methods,the network connection.And through the contrast experiment,verified the improved Unet++ algorithm advantages in accuracy and speed.