Research On Registration Method For High Resolution Remote Sensing Images

Obtaining and using information from remote sensing images is the ultimate goal of remote sensing.After several decades of development,aviation remote sensing has been transformed from military applications to civilian technology,especially some aerial images,which have played a huge role in geological monitoring and agriculture.In the specific observation tasks,due to the limitations of the size and flight height of the aerial camera sensor,most of the remote sensing images collected from satellite and aircraft imaging devices cannot be directly applied,and the image coordinate transformation is required to coordinate the coordinate system.This paper takes image registration method research and architecture improvement as the core,and introduces the main techniques and registration process of image registration in detail.For the characteristics of remote sensing images and special application background,the registration of remote sensing images based on feature points is carried out.Improvements have been made.The main work of this paper are as follows:(1)In view of the large amount of data and the large difference in edge imaging of remote sensing images,nonlinear optimization of linear affine transformation parameters is carried out.This paper improves a coarse-to-fine block-based remote sensing image optimization registration algorithm,which simplifies this complex parameter optimization model into the form of image slice.As a result,the speed and accuracy of remote sensing image registration is higher than traditional algorithms.(2)The remote sensing image viewpoints have heterogeneous and nonlinear geometric deformation.On the basis of image segmentation,the Heteroscedastic error-in-variables model is applied to the registration parameter estimation,and the feature points of each region are given different weights.The accuracy of registration in the building area of interest increased by 16.67%.(3)In this paper,a remote sensing image registration framework based on LIFT feature point extraction and MC-CNN hybrid model is constructed.The trained network structure is used to extract and register the remote sensing image,which simplifies the segmentation process and achieves good registration visual effects.