Research On Cross-medium Underwater Imaging Technology Based On Image Sequences

The rational and effective exploration,development and management of Marine resources are of great significance to the sustainable development of human society.Therefore,countries all over the world are competing to study the underwater target detection technology and underwater information transmission technology.Due to the complexity of underwater environment and the randomness of water movement,the difficulty of underwater image restoration is much greater than that of air.Especially at the water-air interface,under the influence of random fluctuation of water surface,complex refraction of light and other factors,the underwater target image sequence acquired by the surface shooting equipment across the medium will generate strong geometric distortion,which seriously restricts the exploration and subsequent use of the underwater target.Moreover,the existing traditional iterative reconstruction algorithm for cross-media underwater distorted images has a high time complexity,which cannot meet the demand of real-time processing in real work.In recent years,deep learning technology has made great achievements in the field of computer vision and achieved a lot of good results in image restoration.Especially,it is widely used in medical image registration and remote sensing image registration which can be used for reference in the research of cross-media underwater image reconstruction technology.In this paper,the cross-media underwater optical imaging characteristics and crossmedia underwater distorted image restoration technology were studied.The main works are as follows:(1)The research status and key technologies of cross-medium underwater distorted image restoration are summarized,and the important applications of deep learning technology in the field of image registration are briefly introduced.(2)The optical properties of cross-medium underwater are studied.Combined with the image degradation model,the effects of these characteristics on the cross-medium underwater optical imaging results are described.(3)The basic principle and flow of image registration algorithm are summarized,and the traditional cross-medium underwater distorted image registration algorithm based on iteration is simulated.(4)Based on spatial transformer networks and bicubic B-spline image registration algorithm,a cross-medium underwater distorted image restoration algorithm based on deep learning is proposed.The convolutional neural networks analyze the regions corresponding to the paired distorted sequence image and the sequence mean image,and output local deformation parameters.By using the estimated deformation vector fields and bicubic Bspline interpolation with spatial transformer networks,the geometric space of the distorted sequence image is registered to the space of the sequence mean image to obtain the reconstructed image.(5)Simulation tests are carried out on the distorted images generated by computer and the real cross-medium underwater distorted images and the experiment results show that in the above two kinds of image data sets are got good recovery results.And Compared with the traditional iterative restoration algorithm,the running time is greatly reduced,which can meet the demand of real-time restoration.